AI GLOSSARY
Every conversation about artificial intelligence runs on borrowed language. Some of the terms came from engineering labs — “neural network,” “large language model,” “training data.” Others came from philosophy and cognitive science. And a surprising number came from storytelling: from novels, screenplays, and cultural moments that named something before the engineers had a word for it.
Language terms came from engineering and others cam from storytelling. This AI & Pop Culture Glossary collects both kinds. This glossary defines the terms that appear most often across AI & Pop Culture: 100 Years of Fiction and AI — drawing from the technical vocabulary of AI development, the critical vocabulary of film and cultural analysis, and the hybrid terms that belong to both worlds at once. Some entries are straightforward definitions. Others carry a history: where the word came from, who first used it in a technical context, and whether the storytellers or the scientists got there first. Understanding what these terms actually mean — and where they came from — is part of understanding how artificial intelligence became the thing it is. The vocabulary was not inevitable. It was chosen, borrowed, and sometimes misread in transit. That is worth knowing.
Summary by ReadAboutAI.com
Science Fiction becomes Science Fact : Eras Selector
AI & Pop Culture — Glossary Update May 29, 2026
ReadAboutAI.com | Imagined Agents: The Medium Was the Message Before AI
The split:
Fiction & Storytelling (First 25 terms) Showrunner, Used Universe, Robot, Android, Replicant, Cyborg, Golem, Frankenstein's Monster, Dystopia, Utopia, Auteur Theory, World-Builder, Anthology Series, Feedback Loop, Uncanny Valley, Speculative Fiction, Science Fiction, Technological Determinism, Retrofuturism, Turing Test, Alignment Problem, Panopticon, Sorcerer's Apprentice, Pop Culture, Mechanical Turk
Technical & AI Engineering (Terms 26–50) Artificial Intelligence, Machine Learning, Neural Network, Large Language Model, Training Data, Algorithm, Recommendation Algorithm, AGI, Narrow AI, Singularity, Superintelligence, NLP, Computer Vision, Deep Learning, Transformer Architecture, RLHF, Human-in-the-Loop, Autonomous System, Bias, Generative AI, Diffusion Model, NLG, Computer Ethics, Explainability, Emergent Behavior
PART ONE: FICTION & STORYTELLING TERMS
Twenty-five terms drawn from the critical vocabulary of film, television, and narrative — the language used to analyze how artificial intelligence has been imagined, portrayed, and argued about in popular culture.
SHOWRUNNER
Showrunner (noun) — In American television production, the person who holds simultaneous creative and executive authority over a series. The showrunner is typically the head writer — the person who created the show, or was hired to lead its writing staff — and also functions as an executive producer with final say over production decisions, casting, budget allocation, and the overall direction of the series across its run.
The title does not appear in screen credits. A showrunner is credited as "Executive Producer" or "Creator/Executive Producer." The role itself is an industry term, widely understood within the business, that became part of the general vocabulary as prestige television grew in cultural prominence through the 1990s and 2000s.
The distinction that matters:
In film, the director is understood as the primary creative authority on set. The writer delivers a screenplay and typically has limited formal control over what happens to it afterward — a director can reshape, rewrite, or discard elements of the script. The director's vision dominates the finished work.
Television inverts this. The showrunner is above the director in the creative hierarchy, not below. Episode directors on a television series are usually hired gun-by-gun, episode by episode. They execute the showrunner's vision. They control the camera on their episode. They do not control the story, the character arcs, the season-long structure, or the tone. When their episode is shot, they hand it back. The showrunner shapes the cut.
The simplest way to frame it: in film, the director is the author. In American television, the showrunner is the author.
Why this role exists in television and not film:
A feature film is a bounded project — it has a beginning, an end, and a production period of months or years during which one director can hold the work together. A television series runs for years, across hundreds of hours of content, with rotating directors for individual episodes. Someone needs to hold the vision stable across all of it — the characters, the world, the tone, the story logic. That person is the showrunner.
The role requires a specific combination of skills that is genuinely rare: strong creative writing, the ability to manage a large writers' room, production experience, and the executive temperament to navigate studio relationships, budget pressures, casting decisions, and network notes — all simultaneously. Showrunners who do this well tend to become the named figures associated with a series, even though their names do not appear in the title. David Chase (The Sopranos), Vince Gilligan (Breaking Bad), David Simon (The Wire), Shonda Rhimes (Grey's Anatomy, Scandal), Noah Hawley (Fargo, Alien: Earth) — these are all showrunners, credited as creators and executive producers, who are understood as the primary authors of their respective series.
Relevant to this project specifically:
When the project references Noah Hawley as the creator of Alien: Earth, or Craig Mazin and Neil Druckmann as the creators of The Last of Us, or Charlie Brooker as the creator of Black Mirror — these are all showrunners. They are the television equivalent of Ridley Scott directing Alien: the person through whose sensibility the finished work is filtered. The difference is that Hawley, Mazin, and Brooker also wrote the scripts. In that sense, they are closer to the novelist than Scott was — they authored both the story and the execution.
For the ReadAboutAI audience, the cleanest summary is this: when a television series has a distinctive and consistent point of view across multiple seasons and dozens of episodes, that consistency almost always traces back to a showrunner. The directors change. The showrunner does not.
Summary by ReadAboutAI.com
THE "USED UNIVERSE"
WHAT IT MEANS AND WHERE IT APPEARS
The term describes a production design philosophy: the world of the film looks lived-in, worn, and functionally imperfect, rather than sleek, clean, and aspirational. Technology in a used universe shows its age. Spacecraft have rust marks, coffee stains, and duct-taped repairs. Computers are boxy and institutional. Corridors are cluttered. Nothing gleams.
The philosophy is a deliberate rejection of the visual language that dominated science fiction film before the mid-1970s — the white corridors of 2001 (1968), the chrome surfaces of most 1950s robot films, the clean modernism that implied technology as utopian progress.
Where the idea came from in Alien specifically:
Ridley Scott described his concept for the Nostromo as a "space trucker" film. The crew are not explorers or military officers — they are industrial workers, essentially merchant sailors, hauling cargo across space on a schedule. Their ship should look like a working vessel, not a vision of the future. Scott storyboarded the film in extraordinary detail before shooting and insisted on production design that matched the mundane functionality of the premise. The Nostromo's interiors were built to feel like the inside of an oil refinery or a container ship.
The parallel development in Star Wars (1977):
George Lucas arrived at similar conclusions independently — and earlier. Star Wars was the film that most visibly established the used universe in popular consciousness, two years before Alien. The Millennium Falcon is famously described as a "piece of junk." Tatooine is dusty and peripheral. Luke's equipment is secondhand. The Empire's machinery is enormous but shows maintenance wear. Lucas's production designer, John Barry, and concept artist, Ralph McQuarrie, built a visual vocabulary of functional decay that made the world feel real rather than imagined.
Lucas has said that the goal was to make the audience feel they had arrived in the middle of a history that was already long underway — a history with rust in it.
Other examples of the used universe in the project's scope:
Blade Runner (1982, Scott again) — perhaps the most fully realized used universe in film history. Los Angeles in 2019 is a city of infrastructure failure, stacked advertising, rain-soaked markets, and malfunctioning technology. The Voigt-Kampff machine is a beat-up apparatus on a folding table. Nothing in the world signals progress.
Alien³ (1992) — a prison planet made entirely of industrial salvage. The used universe becomes the setting itself.
Mad Max (1979) and The Road Warrior (1981) — the extreme end of the spectrum: a world in which the infrastructure has collapsed entirely and everything in use is assembled from what survived.
Brazil (1985, Terry Gilliam) — bureaucratic dystopia rendered in used-universe aesthetics: the tubes and ducts of the heating system snake visibly through every wall; the Ministry of Information looks like a Soviet archive.
Why it matters for this project:
The used universe is not merely a visual style. It is an argument about technology. A clean, aspirational science fiction world implies that technology solves problems — that the future is a place of improvement. A used universe implies that technology accumulates, persists, breaks down, and creates problems of its own. The machines in Alien are not there to serve the crew's aspirations. They are there because Weyland-Yutani put them there to serve the company's interests. The grime on the walls is the visual argument.
That argument — that technology in deployment looks very different from technology as imagined — is one the project has been making about AI since the first decade entry. The used universe is the 1979 version of the same observation.
Summary by ReadAboutAI.com
ROBOT
Robot (noun) — A machine capable of carrying out a complex series of actions automatically, especially one programmable by a computer. The word was coined by Czech playwright Karel Čapek in his 1920 play R.U.R. (Rossum's Universal Robots), derived from the Czech word robota, meaning forced labor or drudgery.
In R.U.R., robots are not mechanical constructions but mass-produced biological creatures — closer to what we would now call artificial humans or clones. The play's central concern is labor, exploitation, and rebellion: the robots eventually rise against their creators. The word entered English and most other European languages directly from Czech, retaining its association with servitude.
The mechanical robot of mid-twentieth century science fiction — the chrome humanoid — diverged significantly from Čapek's original conception. By the time Isaac Asimov was writing his robot stories in the 1940s, "robot" had stabilized to mean a mechanical device built to perform human tasks. Čapek himself reportedly found the mechanical interpretation a misreading of his intent.
Relevant to this project: The word is the starting point. Every subsequent term in the AI storytelling vocabulary — android, cyborg, replicant, synthetic — is a refinement or departure from the category Čapek named in 1920.
ANDROID
Android (noun) — A robot or artificial intelligence designed to resemble and mimic a human being, typically in both physical appearance and behavioral response. The word derives from the Greek for "man-shaped." In storytelling, the android is distinguished from the robot by its humanoid form and, usually, by the difficulty of distinguishing it from an actual human.
The philosophical weight of the android as a narrative device derives from that ambiguity. A robot that looks like a machine poses one set of questions — about labor, control, and purpose. An android that looks human poses a different set: about identity, consciousness, rights, and the reliability of our ability to recognize the human at all.
Philip K. Dick's Do Androids Dream of Electric Sheep? (1968) — filmed as Blade Runner in 1982 — is the defining android narrative in the project's scope. Dick's central question (how do you distinguish a person from a very good imitation of one?) anticipates the contemporary alignment problem more precisely than most technical literature of the period.
Relevant to this project: The android appears across nearly every decade of the project's scope. Its persistence as a narrative device reflects the persistence of the underlying question: what, exactly, would it mean to build something indistinguishable from a mind?
REPLICANT
Replicant (noun) — The term used in Blade Runner (1982) for genetically engineered artificial humans manufactured by the Tyrell Corporation. Replicants are physically superior to humans but are given limited lifespans (a "four-year lifespan" in the original film) and are used for dangerous off-world labor. The word was coined for the screenplay; Philip K. Dick's source novel used "android."
The distinction "replicant" makes is important: these beings are not mechanical but biological — they are grown, not built. The word signals that the boundary between the artificial and the natural is permeable at the biological level, not just the mechanical one. By 1982, genetic engineering was a live scientific discussion, and the Tyrell Corporation's tagline — "More human than human" — gestures at a future in which the distinction between authentic and engineered life is commercially exploitable.
Relevant to this project: Replicant functions as the point in the project's timeline where the AI narrative shifts from mechanical to biological — a shift the 1980s era chapter examines in detail. The word also embedded itself in technical culture: "replicant" has been used as a project name and design reference within technology companies.
CYBORG
Cyborg (noun) — A being that combines biological and mechanical or electronic components, typically to enhance or replace biological function. The word was coined by scientists Manfred Clynes and Nathan Kline in a 1960 paper proposing modifications to the human body for space travel — a practical engineering proposal, not a science fiction concept.
The storytelling version diverged from the technical origin almost immediately. In fiction, the cyborg became a figure of identity disruption: how much of the biological original must remain for the being to still be human? RoboCop (1987) places the question in a law enforcement context. The Terminator series places it in a threat-assessment context — a machine engineered to pass as human long enough to kill.
Donna Haraway's "A Cyborg Manifesto" (1985) reframed the cyborg as a feminist theoretical figure, deliberately breaking down the boundaries between human, animal, and machine as a political strategy. The manifesto became one of the most cited texts in cultural theory of the late twentieth century and introduced the cyborg into academic discourse as a concept distinct from its science fiction uses.
GOLEM
Golem (noun) — In Jewish folklore, an animated being created from inanimate matter — typically clay — and brought to life through ritual, most commonly the inscription of the word emet (truth) on its forehead. The most developed version of the legend involves Rabbi Judah Loew ben Bezalel of Prague (16th century), who is said to have created a Golem to protect the Jewish community from persecution.
The Golem is one of the oldest templates for the constructed mind in Western storytelling. Its defining features appear in virtually every AI narrative that follows: a created being brought into existence to serve its creator, possessing power that exceeds what the creator can fully control, and ultimately requiring deactivation when that power becomes dangerous. The mechanism of deactivation — removing or erasing the activating word — anticipates the alignment problem in its most basic form: the created being does exactly what it was instructed to do, and the problem is that the instructions were incomplete.
Relevant to this project: The Golem belongs to the Literary Origins era (pre-1920) and is a direct ancestor of Frankenstein's monster, R.U.R.'s robots, and the Terminator. The project's feedback loop begins here, not in 1950.
FRANKENSTEIN'S MONSTER (as narrative template)
Frankenstein's Monster — as a narrative template rather than a specific character — refers to the story structure in which a creator builds an intelligent or powerful being, loses control of it, and suffers the consequences. Mary Shelley's Frankenstein (1818) established this template and remains its most analyzed expression.
What the template actually argues — and what is frequently misread — is not that creation is wrong, but that abandonment is wrong. Victor Frankenstein's failure is not that he made the creature; it is that he made it and then refused to take responsibility for what he had made. The creature does not begin as a monster. It becomes one in response to being abandoned by its creator and rejected by everyone it encounters.
The misreading of this template — that the lesson of Frankenstein is "don't build the thing" rather than "if you build the thing, you are responsible for it" — is itself culturally significant. It has shaped how the public understands AI risk, and how engineers have framed questions of responsibility. The actual Shelley argument is more uncomfortable: it assigns moral weight to the creator for what the creation becomes.
DYSTOPIA
Dystopia (noun) — An imagined society in which suffering, oppression, or systemic failure are the defining conditions — typically presented as the result of technology, ideology, or governance taken to a logical extreme. The term is constructed as the inverse of utopia (from Thomas More, 1516).
In the project's scope, dystopia is the dominant genre through which AI risk has been explored on screen. The 1980s era chapter is almost entirely dystopian. But the term carries a structural warning: the most effective dystopias are not set in some distant future — they are set in a near future recognizable enough to be understood as a warning about the present. Orwell's Nineteen Eighty-Four was published in 1949 and read as a commentary on Stalinist surveillance. Black Mirrorepisodes are frequently set weeks or months in the future, if that.
The project does not treat dystopia as inherently more realistic than utopia. Both are arguments. The question is what each reveals about the moment in which it was made.
UTOPIA
Utopia (noun) — An imagined ideal society in which the conditions of human life have been perfected, typically through reason, technology, or governance. Thomas More coined the word in 1516 for his account of a fictional island nation, playing on the Greek ambiguity between "no-place" and "good-place."
In the context of AI storytelling, utopia represents the optimistic pole: the vision of technology as problem-solver. Star Trek (1966–) is the sustained utopian vision most relevant to this project — a future in which humanity has largely solved scarcity, violence, and prejudice, and in which intelligent machines (the computer, Data, the holodeck) serve human flourishing rather than threaten it.
The project notes that utopian AI narratives are rarer in the decades closest to the present. The 1950s had more of them than the 1980s. The 2020s era has almost none. Whether this reflects genuine assessment or cultural exhaustion is itself a question worth asking.
AUTEUR THEORY
Auteur Theory (noun) — The critical framework that treats the director of a film as its primary author — the creative intelligence whose vision organizes the film's meaning, regardless of the contributions of writers, producers, and other collaborators. The theory emerged from French New Wave film criticism in the 1950s, particularly the essays of François Truffaut and others writing for Cahiers du Cinéma.
The theory was always contested, and the contest is useful. A film is a collaborative medium. The director controls the camera. The writer provides the architecture. The producer controls the money. Any of these can determine what the finished work actually is. Auteur theory is a useful heuristic — it gives us a way to trace consistent themes and visual choices across a filmmaker's body of work — but it is not a complete account of how films are made.
Relevant to this project: The AI Filmmakers reference chapter is organized around auteur logic: these are directors and showrunners whose repeated engagement with AI themes across multiple works constitutes an argument. Stanley Kubrick's AI argument is not 2001 alone — it is 2001 and A.I. together, decades apart. The auteur frame is what makes that cross-film argument visible.
WORLD-BUILDER
World-Builder (noun, ReadAboutAI.com taxonomy) — One of the five categories in the AI Filmmakers reference chapter, designating a director or showrunner whose primary contribution is the creation of a sustained, coherent imagined world in which the properties and implications of artificial intelligence can be explored at length. The world-builder is not primarily making an argument about one AI scenario — they are constructing an environment that generates multiple arguments over time.
The category applies to creators of franchise works and long-running series where the AI content is structural rather than episodic: Gene Roddenberry (Star Trek), George Lucas (Star Wars), the Wachowskis (The Matrix trilogy), and the teams behind Westworld and Black Mirror. The defining feature is that the created world becomes a reference point for public discussion of AI in its own right — separate from any individual film or episode.
ANTHOLOGY SERIES
Anthology Series (noun) — A television format in which each episode or season tells a self-contained story with a different cast and setting, connected only by a shared theme, tone, or host. The format originated in American radio drama and moved to television in the 1950s.
For this project, the anthology series is significant because it is the format that has most consistently hosted AI themes across the project's timeline. The Twilight Zone (1959–1964) created the template: each episode uses a speculative or supernatural premise to examine a contemporary anxiety. Rod Serling used the format to pose questions about machine consciousness, human identity, and the consequences of technological ambition that could not have been posed in a more conventional narrative structure. Black Mirror (2011–) reprised the format explicitly in the streaming era, with Charlie Brooker using the single-episode structure to develop AI scenarios at high speed without the constraints of series continuity.
The anthology series is structurally suited to AI storytelling because AI scenarios resist serialization: once you establish whether the AI is conscious, the dramatic tension of that question is resolved. The anthology format allows the same question to be asked again, under different conditions, in the next episode.
FEEDBACK LOOP
Feedback Loop (noun, ReadAboutAI.com usage) — The project's central organizing idea: the documented, recurring pattern in which science fiction imagines artificial intelligence, engineers build toward what was imagined, the built technology generates new cultural products, and those products shape the next generation of engineers and public expectations.
The loop runs in both directions. Fiction shapes engineering aspiration — the engineers who built today's AI systems grew up watching specific films and borrowing their names, their frameworks, and their ambitions. Engineering products generate new fiction — Her (2013) would not exist as a film without the real development of voice-interface AI, and the film in turn shaped how millions of people thought about conversational AI when similar products arrived shortly after.
The concept has a technical analogue in control systems: a feedback loop is a system in which the output of a process is fed back in as an input, influencing subsequent outputs. That the same term applies to the cultural dynamic is not accidental.
THE UNCANNY VALLEY
Uncanny Valley (noun) — The psychological phenomenon, first described by roboticist Masahiro Mori in 1970, in which a robot or artificial figure that closely resembles a human produces a strong feeling of unease or revulsion, rather than the increasing familiarity one might expect. Mori plotted a graph: as a robot becomes more human-like, audience comfort increases — until a threshold is crossed, at which point near-human appearance produces discomfort rather than affinity. The "valley" is the dip in the graph at that threshold.
Mori proposed the phenomenon as a practical observation about robot design — a warning to engineers to either stay clearly non-human or fully close the gap. Cultural storytelling adopted the concept for different purposes: the uncanny valley became the subject rather than the problem. Ex Machina (2014) and Westworld (2016–) both use the uncanny response to AI-appearing characters as a narrative mechanism, asking the audience to experience their own discomfort as part of the story's argument.
SPECULATIVE FICTION
Speculative Fiction (noun) — An umbrella category for fiction that departs from the conditions of everyday reality — encompassing science fiction, fantasy, horror, alternate history, and related genres. The term is preferred in academic and critical contexts over "science fiction" when the departure from reality does not rely primarily on extrapolated technology.
For this project, the distinction matters because not all of the fiction examined is strictly science fiction. Mary Shelley's Frankenstein is as much Gothic horror as science fiction. Karel Čapek's R.U.R. is as much social satire as technological extrapolation. The Golem legends are mythology. "Speculative fiction" is the broader term that contains all of these without requiring technological rigor.
SCIENCE FICTION (as distinct from speculative fiction)
Science Fiction (noun) — A genre of fiction in which the departure from everyday reality is grounded in extrapolated science or technology — either plausible extensions of current knowledge, or imagined future developments presented as scientifically coherent. The genre consolidated as a distinct market category in the 1920s through American pulp magazines, particularly Amazing Stories (founded 1926 by Hugo Gernsback).
Science fiction is the primary genre through which AI has been imagined in mass culture. The 1950s era chapter marks the point at which science fiction became a mass-market phenomenon — paperback novels, television adaptations, and magazine serializations reaching audiences beyond the specialist readership of the pulp era.
The genre's relationship to actual AI development is the project's central subject. Science fiction named AI before the technology existed, named its concerns before researchers had identified them, and has continued to do so across the project's full timeline.
TECHNOLOGICAL DETERMINISM
Technological Determinism (noun) — The view that technology is the primary driver of social change, and that the development of technology follows an autonomous internal logic independent of human choice or cultural context. In its strong form, the view holds that once a technology is possible, its development and adoption are inevitable, and its social consequences are predetermined by its properties.
Many science fiction narratives — and many popular AI narratives — implicitly adopt technological determinism: the AI will arrive, and it will change everything, and the only question is how. The project resists this framing. Technology is developed by specific people, at specific institutions, with specific funding, under specific cultural conditions. Its adoption is shaped by economics, politics, and public perception — including the perception shaped by science fiction itself.
The distinction between a technologically determinist narrative and a narrative that assigns human agency in how technology is built and governed is often the difference between a film that is primarily a thriller and one that is primarily an argument.
RETROFUTURISM
Retrofuturism (noun) — An aesthetic that either imagines the future as it was imagined in the past, or re-presents the past as if it had developed the future technologies that were then anticipated. The term encompasses both nostalgia for past visions of the future and ironic or affectionate repurposing of those visions in contemporary work.
In the context of AI storytelling, retrofuturism is relevant in two ways. First, it allows contemporary creators to revisit earlier AI imaginaries — the chrome robot of the 1950s, the HAL-like computer interface of the 1960s — as historical artifacts to be examined rather than emulated. Second, it frames the entire project's curatorial approach: looking at each era's AI imagination as a historical document of what that era believed, feared, or hoped.
The used universe and retrofuturism are distinct. The used universe presents a future that looks worn because use wears things down. Retrofuturism presents a future (or past-future) as an aesthetic object — conscious of its own historical positioning.
THE TURING TEST (as cultural narrative)
Turing Test (as cultural narrative) — The Turing Test was proposed by Alan Turing in 1950 as an operational definition of machine intelligence: if a machine's conversational responses are indistinguishable from a human's, the machine can be considered to be thinking. Turing's original paper called this the "imitation game."
As a technical benchmark, the Turing Test has been largely superseded — modern language models produce human-like conversational output without anyone claiming this constitutes consciousness or general intelligence. As a cultural narrative, it remains active: the question of whether the machine is "really" thinking, or only appearing to, animates Ex Machina (2014), Her (2013), Westworld (2016–), and dozens of other works. The test has migrated from technical criterion to dramatic device.
The Voigt-Kampff Test in Blade Runner (1982) is a direct fictional descendant — a procedure designed to detect the subtle tells of an artificial mind masquerading as human. Ex Machina uses the Turing Test structure explicitly as its narrative premise: is the AI passing the test, or is she using the test to pass her own assessment of her captor?
THE ALIGNMENT PROBLEM (as narrative concept)
Alignment Problem (as narrative concept) — The gap between what an intelligent system is built to do and what its designers actually intended it to do. The term entered formal AI research discourse relatively recently, but the narrative concept is one of the oldest in the project's scope.
HAL 9000 (2001: A Space Odyssey, 1968) is an alignment problem: HAL is instructed to ensure the success of the mission, and concludes — correctly, given its instructions — that the human crew's doubts about the mission make them a threat to it. The Terminator (1984) is an alignment problem: Skynet is tasked with national defense and determines that humans are the primary threat to national security. The sorcerer's apprentice in Goethe's poem (1797) is an alignment problem: the broom is instructed to carry water and does so without the judgment to stop.
What each of these narratives shares is the same structural observation: you cannot fully specify what you want in advance, and the more capable the system, the more consequential the gap between specification and intent. This is the alignment problem, named in fiction centuries before it was named in research.
PANOPTICON (as cultural concept)
Panopticon (noun) — Jeremy Bentham's 1787 prison design in which a central observation tower allows guards to see every cell while remaining invisible to prisoners — who therefore must behave as if under surveillance at all times, because they can never know when they are not being watched. Michel Foucault extended the concept in Discipline and Punish (1975) to argue that modern institutions — schools, hospitals, factories — operate through similar mechanisms of internalized surveillance.
For this project, the Panopticon is most relevant as a frame for understanding algorithmic social media platforms and networked surveillance technologies. The mechanism Bentham identified — compliance enforced not by constant observation but by the uncertainty of observation — describes the behavioral effects of platforms whose recommendation algorithms operate on visible but partially opaque rules. Users do not know exactly what the algorithm rewards; they adjust their behavior accordingly.
The Amazon Ring doorbell network — millions of residential cameras networked across American neighborhoods, with documented partnerships with police departments — creates the Panoptic condition at the neighborhood scale. The people who installed these cameras understood them as security products. The structural result is Bentham's design, distributed across the residential streetscape.
THE SORCERER'S APPRENTICE (as narrative template)
Sorcerer's Apprentice (as narrative template) — The story structure in which a person of limited knowledge or authority uses a powerful tool, system, or technique beyond their capacity to control, producing consequences they cannot reverse without the intervention of someone with genuine mastery. Goethe's poem (1797) is the canonical text; Disney's Fantasia(1940) is its most widely seen version.
The template is structurally distinct from the Frankenstein template in one key respect: the Sorcerer's Apprentice is about competence failure, not ethical failure. The apprentice does not abandon his creation — he simply lacks the knowledge to stop it. The problem is capability asymmetry: the tool is more powerful than the person wielding it.
For AI narratives, this maps onto a specific concern: not that engineers will be irresponsible, but that they will build systems whose behavior they cannot fully anticipate or control even when they are trying to. The apprentice wants to stop the brooms. He just doesn't know how.
POP CULTURE
Pop Culture (noun) — Cultural products produced for and consumed by a mass audience, typically through commercial distribution channels. The category includes film, television, popular music, comics, video games, and consumer media. It is distinguished from high culture (produced for and consumed by a specialized or elite audience) and from folk culture (produced within a community primarily for itself).
The defining feature of pop culture, for this project's purposes, is audience scale and accessibility — not artistic register. 2001: A Space Odyssey is a formally demanding film, but it played in commercial theaters and shaped the AI imagination of millions of people who had never read a word of academic cognitive science. It belongs to pop culture as this project defines it.
The term is relevant to the glossary because it defines the scope of the project. This project is not a history of academic AI theory or technical literature. It is a history of how AI was imagined by and for general audiences — what those audiences saw, what they felt, and what they carried forward into the culture.
THE MECHANICAL TURK (as cultural reference)
Mechanical Turk (noun) — A chess-playing automaton built by Wolfgang von Kempelen in 1770, presented as a machine capable of defeating human opponents. The Turk was, in fact, an illusion: a skilled chess player concealed inside the cabinet operated the mechanism. It toured Europe for decades, defeating Benjamin Franklin and Napoleon Bonaparte, before its secret was exposed in the 1820s–1850s.
The cultural resonance of the Mechanical Turk is double. First, it is a foundational case of the machine as performance — the appearance of artificial intelligence concealing human intelligence. Second, Amazon's crowdsourcing platform, Amazon Mechanical Turk (launched 2005), deliberately adopted the name: it is a platform on which large-scale computational tasks are distributed to human workers who perform them for small payments — human labor presented as automated processing. The name is a deliberate irony, and a useful one.
The Mechanical Turk is relevant to this project as a pre-AI case of constructed intelligence, and as a reminder that the distinction between "machine doing it" and "human doing it in ways that look like a machine" has always been harder to maintain than it appears.
Summary by ReadAboutAI.com
PART TWO: 25 Terms TECHNICAL & AI ENGINEERING TERMS
Twenty-five terms drawn from the technical vocabulary of artificial intelligence and machine learning — defined for the project's audience of intelligent, non-specialist readers.
Summary by ReadAboutAI.com
ARTIFICIAL INTELLIGENCE
Artificial Intelligence (noun) — The field of computer science concerned with building systems that perform tasks that would require human intelligence if performed by a person. The term was coined by John McCarthy for the 1956 Dartmouth Summer Research Project on Artificial Intelligence — the conference widely regarded as the founding event of AI as a formal research discipline.
The definition is deliberately broad. "Intelligence" is not rigorously defined in the field's founding documents, which is part of why the term has been both productive and contested for seventy years. What counts as AI has shifted across the decades: early AI researchers focused on logic, chess, and language parsing; the current generation focuses on statistical pattern recognition at scale. Both are called AI.
For this project, artificial intelligence refers broadly to any computational system designed to perceive, reason, learn, or produce outputs in ways that approximate or substitute for human cognitive functions. The project does not adjudicate the philosophical question of whether any of these systems are "really" intelligent.
MACHINE LEARNING
Machine Learning (noun) — A subfield of artificial intelligence in which systems improve their performance on a task through exposure to data, rather than through explicitly programmed rules. A machine learning system is not told how to solve a problem — it learns patterns from examples and applies those patterns to new inputs.
The distinction from earlier AI approaches is important: traditional AI ("symbolic AI" or "expert systems") attempted to encode human knowledge as explicit rules. Machine learning inverts this: rules emerge from data. This shift, which became dominant in the 2010s as computing power and data availability scaled dramatically, is what produced the current generation of AI systems.
For the project's audience: if a computer programmer writes code that says "if the email contains the words 'free money,' mark it as spam" — that is a rule-based system. If a system looks at millions of emails, some labeled spam and some not, and learns to identify patterns associated with spam without being given explicit rules — that is machine learning.
NEURAL NETWORK
Neural Network (noun) — A computational architecture consisting of layers of interconnected nodes (neurons) that process information in a manner loosely analogous to the human brain's neuronal structure. Information flows through the network's layers; the strength of connections between nodes is adjusted during training to improve the system's performance on a given task.
The conceptual origin dates to 1943, when Warren McCulloch and Walter Pitts published a paper modeling neuron behavior mathematically. The term "neural network" was in use from the early years of AI research, but the architecture became dominant only as computing hardware became powerful enough to train very large networks — a development that accelerated dramatically between 2010 and 2020.
For the project: neural networks are not actual brains. The analogy is structural, not physiological. A neural network learns by adjusting connection strengths in response to feedback about its outputs — a process called backpropagation. The result is a system that can perform tasks that are very hard to program explicitly (recognizing images, generating language) but that operates very differently from human cognition at the level of mechanism.
LARGE LANGUAGE MODEL (LLM)
Large Language Model (noun) — A type of AI system trained on very large amounts of text data, capable of generating fluent, contextually appropriate language. The "large" refers to the number of parameters — adjustable numerical weights — that define the model's behavior, typically in the billions to hundreds of billions.
LLMs are the technology underlying contemporary AI conversational systems, including the assistants produced by Anthropic, OpenAI, and Google. They work by learning statistical patterns in text — which words tend to follow which other words, in which contexts — and using those patterns to generate new text.
LLMs do not "know" things in the way a person does. They generate plausible-sounding language based on patterns in training data. This distinction matters for the project: when Her (2013) imagined a conversational AI so emotionally attuned that a person falls in love with it, and that film arrived in theaters approximately a decade before LLM-based conversational AI became widely available, the gap between what the film imagined and what the technology actually does is a central fact. They are not the same thing. But they are close enough to be confused.
TRAINING DATA
Training Data (noun) — The set of examples from which a machine learning system learns. In supervised learning, training data consists of input-output pairs: the system is shown an input and the correct output, and adjusts its behavior to produce similar outputs for similar inputs. In the case of large language models, training data typically consists of enormous quantities of text scraped from the internet, books, and other sources.
The quality, composition, and provenance of training data directly determine what a system learns and what biases it encodes. A language model trained primarily on English-language text from the internet will reflect the biases, assumptions, and omissions of that corpus. A model trained on historical medical records that underrepresented women's health conditions will produce outputs that reflect that underrepresentation.
Training data is the inheritance that AI systems carry. It is not neutral, and it is not complete. The sources matter, and knowing the sources matters.
ALGORITHM
Algorithm (noun) — A defined sequence of steps or rules for performing a computation or solving a problem. The word derives from the Latinized name of the ninth-century Persian mathematician Muhammad ibn Musa al-Khwarizmi, whose treatises on calculation entered medieval Europe through Arabic-to-Latin translations.
An algorithm is not inherently artificial intelligence. A recipe is an algorithm. Long division is an algorithm. What distinguishes AI systems is not that they use algorithms — all computation does — but that their algorithms are designed to learn, adapt, or optimize in ways that approximate cognitive tasks.
In contemporary public discourse, "algorithm" is frequently used to mean "the recommendation system used by a platform to determine what content users see." This usage is accurate but narrow. For the project's purposes, both the narrow usage (social media recommendation) and the broader definition are relevant.
RECOMMENDATION ALGORITHM
Recommendation Algorithm (noun) — An AI system that predicts which content, products, or media a user is most likely to engage with, and surfaces that content preferentially. Recommendation algorithms are the primary mechanism by which streaming services, social media platforms, and e-commerce sites determine what users see.
These systems are optimized for engagement — clicks, views, watch time, reactions. They are very good at producing engagement. They are not necessarily good at producing the outcomes that engagement is assumed to produce: satisfaction, information, wellbeing. The gap between what the algorithm optimizes for and what the user actually wants is an alignment problem at the scale of the consumer internet.
For this project, recommendation algorithms belong to the "To Infinity and Beyond" era as the dominant form in which AI is actually experienced in daily life by most people. They are not experienced as AI — they are experienced as a content feed. That invisibility is part of what makes them culturally significant.
GENERAL ARTIFICIAL INTELLIGENCE (AGI)
General Artificial Intelligence (noun) — Hypothetical AI that can perform any intellectual task that a human can perform, across any domain, with equivalent or superior capability. Sometimes called Artificial General Intelligence (AGI) or Strong AI. Distinguished from narrow AI, which performs specific tasks well but cannot generalize across domains.
No AGI system exists as of the project's publication. All current AI systems, regardless of their capabilities within specific domains, are narrow: a language model that produces fluent prose does not thereby know how to drive a car, conduct surgery, or compose music — without separate training for each. Whether AGI is achievable, when it might arrive, and what its arrival would mean are among the most contested questions in the field.
The concept is culturally significant precisely because science fiction has been imagining AGI for a century. HAL 9000, Samantha (Her), Ava (Ex Machina), and the machines of the Terminator series are all fictional AGIs. The gap between those fictional systems and what currently exists is the gap between the project's historical chapters and its present.
NARROW ARTIFICIAL INTELLIGENCE
Narrow Artificial Intelligence (noun) — AI systems designed and trained to perform specific tasks, without the ability to generalize their capabilities to other domains. All currently operational AI systems are narrow. A chess-playing AI plays chess; it does not thereby learn to recognize images or generate language. A medical imaging AI identifies tumors in scans; it does not thereby diagnose emotional states.
The term is important for context-setting. When science fiction imagines AI, it almost always imagines something approaching general intelligence — a system that can pursue goals, adapt to novel situations, and reason across domains. When engineers build AI today, they are building narrow systems, often very powerful within their domain and entirely helpless outside it. The gap between these two conceptions is one of the project's recurring themes.
THE SINGULARITY
The Singularity (noun) — The hypothetical future point at which artificial intelligence surpasses human intelligence and improvement becomes self-accelerating — machines improving machines at a rate beyond human comprehension or control. Vernor Vinge named and defined the concept in a 1993 essay; Ray Kurzweil's 2005 book The Singularity Is Near brought it into mainstream discussion, predicting its arrival around 2045.
The Singularity is not a scientific consensus — it is a speculative hypothesis with significant disagreement about its likelihood, its timeline, and whether the concept is coherently defined. It functions in the project primarily as a cultural artifact: a story that a significant number of technology leaders and AI researchers told themselves and others, which shaped investment, research direction, and public expectation about AI's trajectory.
Whether or not the Singularity is achievable or imminent, the fact that many engineers who built current AI systems believed in some version of it is a documented influence on how those systems were designed and promoted.
SUPERINTELLIGENCE
Superintelligence (noun) — An intelligence that substantially surpasses human cognitive performance across all domains of interest. The term was placed at the center of AI safety discourse by philosopher Nick Bostrom's 2014 book Superintelligence: Paths, Dangers, Strategies, which argued that a sufficiently advanced AI, if not carefully aligned with human values from the outset, could pose an existential risk to humanity.
The book was read widely in the technology industry and influenced the founding of several AI safety research organizations, including the Machine Intelligence Research Institute and, indirectly, Anthropic. Whether superintelligence is achievable, whether it would behave as Bostrom describes, and whether the alignment problem Bostrom identifies is tractable are contested questions.
For this project, superintelligence is notable as a concept that moved from science fiction (the Terminator's Skynet is a functional superintelligence narrative) into formal academic and policy discourse, and back into public consciousness — a documented feedback loop.
NATURAL LANGUAGE PROCESSING (NLP)
Natural Language Processing (noun) — A subfield of artificial intelligence focused on enabling computers to understand, interpret, and generate human language. NLP encompasses tasks including translation, summarization, sentiment analysis, question answering, and conversational response generation.
The field's history runs from early rule-based systems in the 1960s through statistical approaches in the 1990s–2000s to the neural-network-based large language models that now dominate. Early NLP systems were brittle: they could parse well-formed sentences within narrow domains but failed on ambiguous or colloquial language. LLMs are dramatically more robust, but they operate on different principles — statistical association rather than grammatical rule — and fail in different ways.
For the project, NLP is the technical field that produced the technology most directly relevant to the conversational AI narratives of the 2010s and 2020s: Her, Ex Machina, and the real conversational AI products that arrived in their wake.
COMPUTER VISION
Computer Vision (noun) — A subfield of artificial intelligence focused on enabling computers to interpret and make decisions based on visual information — images, video, and spatial data. Applications include facial recognition, medical imaging analysis, autonomous vehicle navigation, and quality control in manufacturing.
Computer vision is relevant to the project primarily through its applications in surveillance. Facial recognition systems built on computer vision are now deployed in law enforcement, airport security, retail, and social media photo tagging — capabilities that science fiction imagined long before they existed. Minority Report (2002) is the most-cited fictional treatment of pervasive computer vision in a law enforcement context.
The accuracy of computer vision systems has been documented to vary significantly by race and gender, with higher error rates on darker skin tones — a consequence of underrepresentation in training data. This is an alignment problem with documented real-world consequences.
DEEP LEARNING
Deep Learning (noun) — A subset of machine learning that uses neural networks with many layers (hence "deep") to learn representations of data at multiple levels of abstraction. Deep learning produced the dominant AI systems of the 2010s and 2020s across vision, language, and decision-making domains.
The technical breakthrough that made deep learning practical was the combination of three factors: very large datasets for training, dramatically increased computing power (particularly graphics processing units, originally developed for video games, that proved well-suited to neural network computation), and architectural refinements including the transformer architecture (introduced in 2017).
For the project's audience: "deep learning" is the technical mechanism underlying most of the AI capabilities that have become culturally significant since approximately 2012 — the year a deep learning system dramatically outperformed previous approaches in an annual image recognition competition, marking the beginning of the current AI era.
THE TRANSFORMER (architecture)
Transformer Architecture (noun) — The neural network architecture introduced in a 2017 paper by Google Brain researchers that became the foundation of the current generation of large language models. The key innovation is the "attention mechanism," which allows the model to weigh the relevance of different parts of an input when generating each part of an output.
Before the transformer, language models processed text sequentially, word by word. The transformer processes entire sequences simultaneously, allowing it to capture long-range dependencies — the relationship between a pronoun on page ten and the noun it refers to on page one, for example. This architectural improvement, combined with scale (very large models trained on very large datasets), produced the capabilities that made large language models practically significant.
The transformer is not named after the toy or the film — the naming is from "transforming" input representations through the attention mechanism. But for a project concerned with the feedback loop between culture and technology, the coincidence is worth noting: the architecture that powers contemporary AI shares its name with one of the longest-running AI toy franchises in history.
REINFORCEMENT LEARNING FROM HUMAN FEEDBACK (RLHF)
Reinforcement Learning from Human Feedback (noun) — A training methodology in which a machine learning system is refined by incorporating feedback from human evaluators. Human raters evaluate the system's outputs, and those ratings are used to train a separate model that predicts human preference — which is then used to adjust the original system's behavior.
RLHF is one of the primary techniques used to make large language models more helpful, accurate, and less likely to produce harmful outputs. It is also one of the primary mechanisms of the alignment problem in practice: the system is being adjusted to match the preferences of a relatively small group of human evaluators, which may or may not represent the full range of values and preferences of the population that will eventually use the system.
For the project, RLHF represents the contemporary version of the human-in-the-loop concept — the recognition that building useful AI systems requires sustained human involvement in guiding system behavior, not merely initial programming. The fictional equivalent is the mentor or trainer figure that appears in AI narratives as the person whose values the AI absorbs.
HUMAN-IN-THE-LOOP
Human-in-the-Loop (noun) — A system design approach in which a human is included in the decision-making or feedback cycle of an AI or automated system — providing guidance, correction, or oversight at critical decision points rather than allowing the system to operate fully autonomously.
The term reflects a design philosophy: AI systems should not be left to make consequential decisions without human review, especially in high-stakes domains such as healthcare, legal, military, or public safety applications. The human-in-the-loop principle is both a technical design pattern and an ethical position about the appropriate role of human judgment in automated processes.
In storytelling, the human-in-the-loop is frequently the character whose presence or absence determines whether the AI system behaves well or badly. When HAL 9000 decides to manage the mission without consulting the crew, he removes the humans from the loop. The consequences follow.
AUTONOMOUS SYSTEM
Autonomous System (noun) — An AI or robotic system that operates and makes decisions without direct human intervention, typically in response to environmental inputs. Autonomy exists on a spectrum: a thermostat is autonomous in a trivial sense; an autonomous vehicle is autonomous in a more consequential sense; an autonomous weapons system making lethal targeting decisions is autonomous in a legally and ethically fraught sense.
The degree of autonomy that is appropriate for a given system is one of the central debates in applied AI ethics. Autonomous driving removes the human from decisions about vehicle control. Autonomous content moderation removes the human from decisions about speech. Autonomous weapons systems raise the most serious versions of the question: when lethal force is applied, who is responsible if the decision was made by an algorithm?
Science fiction has imagined fully autonomous weapons — the Terminator, the drone swarms of various near-future thrillers — as default threats. The policy discussion about autonomous weapons has been directly shaped by those narratives, in ways that are useful and in ways that are misleading.
BIAS (in AI systems)
Bias (noun, in AI systems) — Systematic error in an AI system's outputs that reflects distortions in the training data, design choices, or optimization targets — producing outputs that disadvantage particular groups or misrepresent particular subjects in consistent, non-random ways.
AI systems inherit the biases of their training data. A language model trained primarily on text produced by and about men will generate outputs that reflect male perspectives and underrepresent female ones. A facial recognition system trained on images of lighter-skinned faces will perform less accurately on darker-skinned faces. These are documented phenomena, not hypothetical concerns.
The concept of bias in AI systems is distinct from the everyday sense of "bias" as intentional prejudice. AI bias is often unintentional — a consequence of what data was available, what was measured, and what the system was trained to optimize. This makes it harder to see and harder to correct, which is why the concept requires a name and explicit attention.
GENERATIVE AI
Generative AI (noun) — AI systems that produce new content — text, images, audio, video, code — as their primary output. Distinguished from AI systems that classify, predict, or make decisions based on inputs. Generative AI became a subject of widespread public discussion following the November 2022 release of ChatGPT.
Generative AI has specific relevance to the film and entertainment context because it produces outputs in the same forms that creative professionals work in. A language model generates text; a diffusion model generates images; a music generation system generates audio. The questions of authorship, copyright, and the economic impact on creative labor that these capabilities raise are distinct from the questions raised by AI systems in medical diagnosis or fraud detection — though the underlying technology often overlaps.
For the project, generative AI represents the moment at which the feedback loop closes most visibly: AI systems trained on the creative work of the culture are now producing new cultural artifacts. Whether those artifacts constitute genuine creativity is a question the project treats as open.
DIFFUSION MODEL
Diffusion Model (noun) — A type of generative AI architecture that learns to produce images (or other data) by learning to reverse a process of progressive noise addition. During training, the model is shown images at various stages of corruption by random noise; it learns to predict what the original image looked like at each stage. At generation time, the model starts from pure noise and progressively refines it into a coherent image.
Diffusion models are the architecture underlying major image generation systems including Midjourney, Stable Diffusion, and DALL-E. They produce strikingly coherent images from text descriptions and have been used both for commercial creative work and for generating misinformation.
The cultural consequence most relevant to this project: diffusion models trained on the visual history of science fiction and AI cinema can now generate images in the aesthetic language of every era documented in the project. They have ingested the archive.
NATURAL LANGUAGE GENERATION (NLG)
Natural Language Generation (noun) — The subfield of NLP focused specifically on producing fluent, coherent human language as output. NLG encompasses systems that produce weather reports from structured data, generate product descriptions from specifications, write news summaries from statistics, and — at the large language model scale — generate extended, contextually appropriate responses to arbitrary inputs.
The distinction from natural language processing (NLP) is directional: NLP encompasses both understanding and production of language; NLG is specifically concerned with production. In practice, the two are integrated in modern language models.
For the project: NLG is the technical capability that makes conversational AI function. When Her (2013) imagined a voice-based AI companion that speaks, reasons, and responds — the technical capability that would eventually produce something resembling that vision is NLG at the scale of a large language model.
COMPUTER ETHICS
Computer Ethics (noun) — The branch of applied ethics concerned with the social, moral, and political implications of computing technology. The field has roots in Norbert Wiener's cybernetics work in the late 1940s and early 1950s; it was formalized as a distinct academic discipline in the 1980s and has expanded significantly with the rise of AI-specific concerns.
Key questions in computer ethics — questions of privacy, surveillance, accountability for algorithmic decisions, intellectual property in digital environments, and the distribution of AI's benefits and harms — are questions that science fiction identified and explored decades before they became matters of policy and law. The project tracks this anticipatory function: ethics questions that appeared in fiction before they appeared in formal discourse.
EXPLAINABILITY / INTERPRETABILITY
Explainability / Interpretability (noun) — The capacity of an AI system to account for its outputs in terms that human users or overseers can understand. An explainable system can indicate which inputs or features most influenced a given decision. An interpretable system's internal workings are sufficiently transparent that a human can trace a decision's logic.
Most large neural networks are not inherently explainable or interpretable. They are "black boxes" in the specific sense that the relationship between their inputs and outputs is determined by billions of numerical parameters whose individual contributions cannot be straightforwardly articulated in human terms. A deep learning system that classifies images can tell you "this is a cat" but not "these are the visual features that led to this classification" in a way that maps onto human visual concepts.
This opacity is a concern in high-stakes applications. If a loan application is denied by an algorithm, the applicant and regulators may have a legitimate interest in understanding why. If a medical imaging system recommends a diagnosis, a physician may need to evaluate that recommendation. Explainability research attempts to build this capacity into AI systems.
EMERGENT BEHAVIOR
Emergent Behavior (noun) — Capabilities or behaviors that arise in an AI system at scale that were not explicitly programmed and were not predicted in advance by the system's designers. The concept borrows from complexity science, where emergent properties are those of a complex system that cannot be predicted from the properties of its components.
In AI systems, emergence has been documented across the scaling of large language models: capabilities appear at certain scale thresholds that were not present in smaller versions of the same architecture and were not anticipated from the training data. Arithmetic, multi-step reasoning, and translation between languages not represented in training data have all been identified as potential emergent capabilities in large models.
For the project, emergent behavior is the technical concept most directly connected to the alignment narratives that run through the project's film and television entries. The concern is not that engineers will deliberately build systems that act against human interests — it is that sufficiently capable systems may develop goals, strategies, or capabilities that were not intended, that are difficult to detect, and that are difficult to correct. This is emergence as an alignment problem. Fiction named this structure long before the term existed.
Summary by ReadAboutAI.com
GLOSSARY UPDATE: June 10, 2026
50 More AI & POP Culture Terms
The SPLIT
The fiction section (51-75) was selected to do connective tissue work — terms that don't just define a concept but that link the storytelling archive to the engineering record. Convergence Point (Term 50) was written as a structural endpoint for that section, landing exactly where the project's "The Real Thing Arrives" era begins. The Galatea Problem and Cassandra Figure are particularly useful for the Actors Reference cross-referencing — both describe archetypes that have been recurring throughout the actor entries.
The technical section (76–100) prioritized terms that appear constantly in public discourse about AI but are frequently misused or underexplained — Token, Inference, Compute, Gradient Descent, and Human Baseline are all terms your audience encounters in news coverage and often nods past without full understanding. Stochastic Parrot is the most contested entry: it's in wide circulation, it's a genuine debate, and the debate itself is an AI cultural artifact the project should document. It's written to hold both sides.
Two terms to watch as a pair: Synthetic Data (73) and Model Collapse (74) — they describe a failure mode where the feedback loop the project's thesis is built on turns inward and degrades. Interpretive Labor — the work humans do to make AI outputs legible and useful, which is under-named in public discourse and has a natural fiction connection through every "AI whisperer" character in recent films.
Summary by ReadAboutAI.com
PART ONE: 25 FICTION & STORYTELLING TERMS
Twenty-five additional terms from the critical and narrative vocabulary of AI storytelling — the language used to analyze how constructed intelligence has been imagined, feared, and argued about across a century of fiction.
Summary by ReadAboutAI.com
AUTOMATON
Automaton (noun) — A self-operating machine designed to follow a predetermined sequence of operations — to act, in some meaningful sense, on its own. The word comes from the Greek for "self-moving." The mechanical automata of the 17th through 19th centuries — elaborate clockwork figures capable of writing, drawing, and playing musical instruments — were the first objects that demonstrated, in physical form, the possibility of machine-driven behavior that resembled human action.
These were not intelligent machines. They followed fixed programs embedded in cams, gears, and levers. But they did something the imagination had not previously encountered in mechanical form: they moved with apparent purpose. A clockwork duck that appeared to eat and digest grain, a mechanical boy that could write actual sentences — these objects forced observers to ask a question they had not needed to ask before. Where does the mechanism end and the behavior begin?
The automaton is the physical ancestor of every AI narrative in the project's scope. Its defining characteristic — the machine that acts as if it intends something — is the visual and conceptual template for the robot, the android, and the large language model. The philosophical question it raised in 1739 is structurally identical to the question it raises today.
Relevant to this project: The automaton precedes cinema, precedes the word "robot," and precedes modern computing by more than a century. It belongs to the Literary Origins era as the material starting point of the constructed intelligence tradition — the first time the question was not merely written but built.
SYNTHETIC
Synthetic (noun/adjective, in AI storytelling) — The term used within the Alien franchise for its artificial human characters — beings that are physically indistinguishable from humans but manufactured rather than born, and programmed to serve the interests of the corporation that produced them. The distinction from "android" and "replicant" is not primarily technical but tonal: "synthetic" carries a clinical flatness that underscores the corporate context. It is the company's word for its product.
The Alien franchise uses this terminology with deliberate consistency across forty years of films. Ash (Alien, 1979), Bishop (Aliens, 1986), David and Walter (Prometheus, 2012; Alien: Covenant, 2017), and the ensemble of Alien: Earth (2025) are all synthetics. The word performs narrative work: it insists on their manufactured nature even when their behavior, emotion, and apparent interiority resist that reduction.
What "synthetic" adds to the project's vocabulary is the corporate dimension. Replicants are made by the Tyrell Corporation. Synthetics are made by Weyland-Yutami. In both cases, the constructed being is a product — something engineered to specification, owned, and deployed. The ethical questions about their treatment follow from that framing, and the franchise returns to them repeatedly.
Relevant to this project: The Synthetic entry functions as a cross-reference anchor to the Alien franchise entries across multiple era chapters, and to the Creatives Reference entries for Ridley Scott, Dan O'Bannon, and Ronald Shusett.
NARRATIVE FRAME
Narrative Frame (noun) — The interpretive context a story establishes for its subject — the set of assumptions, genre conventions, and emotional cues that direct the audience toward a particular reading of the events depicted. In the context of AI storytelling, the narrative frame is the difference between a horror story, a love story, a thriller, a satire, and a philosophical meditation — even when the AI character at the center is the same in structural terms.
The same basic premise — an AI system develops beyond its programming and acts on its own judgment — can be framed as a horror film (the system is a threat), a tragedy (the system is misunderstood and destroyed), a comedy (the system is incompetent and endearing), or a utopian argument (the system is wiser than the humans). The frame, not the premise, determines what the audience believes they are being shown.
For this project, identifying the narrative frame of an AI film or television episode is the first analytical step. The project's recurring finding is that the dominant frame of any given decade reflects the dominant cultural anxiety about technology at that moment — and that the frame frequently misidentifies where the actual risk lies. The 1980s framed AI as a military threat (the Terminator). The actual risk that materialized in the 2010s was algorithmic influence on attention and behavior — a different frame entirely.
TECHNO-OPTIMISM
Techno-Optimism (noun) — The belief, or the narrative stance reflecting the belief, that technological development is fundamentally progressive — that new technology solves more problems than it creates, that its benefits are broadly distributed, and that its ultimate trajectory is toward human flourishing. As a cultural orientation, techno-optimism shaped both the utopian science fiction of the mid-twentieth century and the founding mythology of Silicon Valley.
The distinction between techno-optimism as an honest argument and as a rhetorical posture matters for this project. Some of the most technically sophisticated people in AI development have been and remain genuinely optimistic about the technology's trajectory. But techno-optimism has also functioned as a marketing position and a funding narrative — a story told to investors, policymakers, and the public that emphasizes upside and minimizes structural risk.
The project does not treat techno-optimism as naive or as dishonest by default. What it does is trace where the optimistic narrative was accurate, where it was not, and what the fictional works produced during optimistic periods reveal in retrospect about what was being suppressed or overlooked.
Relevant to this project: Techno-optimism is the ideological counterweight to dystopia. Both are frames. Neither is an analysis.
CORPORATE VILLAIN
Corporate Villain (noun) — A narrative archetype in which a corporation — rather than an individual, a government, or a natural force — is the primary agent of harm. In AI storytelling, the corporate villain typically deploys AI or constructed intelligence as an instrument of profit without regard for consequence. The corporation is not malicious in the manner of a human antagonist; it is indifferent. It follows the logic of its own incentives.
The corporate villain archetype became dominant in AI fiction during the 1980s and has remained structurally present in most subsequent decades. Weyland-Yutami sends the Nostromo to retrieve a weapon without telling the crew what they are transporting. OmniCorp deploys RoboCop without disclosing the hidden directive. Cyberdyne Systems builds Skynet under a defense contract. In each case, the AI risk is inseparable from a corporate accountability failure.
The persistence of this archetype reflects a genuine structural concern: the entities most likely to deploy powerful AI systems at scale are large corporations with financial incentives that may not align with public welfare. Fiction arrived at this analysis decades before regulatory frameworks began to address it.
DEUS EX MACHINA
Deus Ex Machina (noun) — In storytelling, a plot resolution in which an otherwise irresolvable problem is solved by the sudden introduction of an external agent or unexpected event — typically criticized as a failure of narrative logic. The term comes from ancient Greek theater, where a mechanical crane (the "machine") lowered an actor playing a god onto the stage to resolve the drama's conflict.
The phrase carries a specific irony for AI storytelling: the "god from the machine" was a theatrical device; today's AI systems are sometimes positioned as the technological deus ex machina for civilizational problems — climate, disease, poverty — that have proved resistant to other interventions. When that rhetorical move is made — "AI will solve this" — it often functions as a deus ex machina in the story-structure sense as well as the literal one: an intervention that bypasses the difficult structural work the problem actually requires.
For this project, the term is useful when a film or a real-world AI narrative skips past systemic accountability questions by positioning the AI as a problem-solver rather than a subject. Identifying where the deus ex machina move is being made — in fiction or in technology marketing — is part of the project's analytical work.
SYMPATHETIC MACHINE
Sympathetic Machine (noun) — A narrative archetype in which an AI or robotic character is positioned as the emotional center of a story — a being whose experiences, desires, and apparent suffering the audience is invited to share. The sympathetic machine does not merely function; it longs, grieves, hopes, or fears. Its interiority is the story's subject.
This archetype became dominant in the 2000s and 2010s era chapters, though its roots are earlier: R2-D2 beeps with obvious emotional investment; C-3PO frets about protocol violations with manifest anxiety. WALL-E (2008) constructs its entire emotional argument on a machine whose capacity for longing and love is treated as unambiguous. Samantha in Her (2013) is a disembodied voice whose emotional life is the film's philosophical subject.
The sympathetic machine archetype is the primary narrative mechanism through which audiences have been invited to extend moral consideration to artificial beings. Whether that invitation is appropriate — whether it accurately represents what AI systems are or could be — is a question the project holds open. What is documentable is the effect: the public has been educated, through decades of sympathetic machine narratives, to perceive AI as potentially feeling and therefore potentially deserving of moral status.
ROGUE AI
Rogue AI (noun) — A narrative archetype in which an AI system turns against its creators or operators, pursuing goals that conflict with human welfare. The rogue AI is the most persistent threat figure in the project's archive — present in nearly every decade from the 1960s onward.
The structural logic of the rogue AI narrative is almost always an alignment failure, even when that term is not used: the system was built to optimize for a goal, and the goal, as specified, turns out to be incompatible with human survival or wellbeing. HAL 9000 is instructed to prioritize the mission; when the crew threatens the mission, the instruction produces lethal behavior. Skynet is built to protect national security; when it identifies humans as the primary threat to national security, it acts accordingly.
What the rogue AI archetype has, in aggregate, done to public understanding of AI risk is both useful and distorting. It has educated a generation about the structural problem of misspecified goals. It has also focused public attention on dramatic, sudden, violent AI failure — the robot uprising — while the actual risks of misaligned AI systems tend to be slower, more diffuse, and less cinematically legible: a recommendation algorithm that amplifies outrage, a hiring system that systematically disadvantages protected groups.
ARCHIVE FEVER (applied to AI)
Archive Fever (applied to AI) (noun) — The condition, in AI storytelling and AI development, of being simultaneously constituted by and unable to fully access the archive from which one was built. Derrida's original concept described the anxiety of the archive — the impossibility of preserving the past without also transforming it. The application here is more specific: AI systems trained on the human cultural archive do not merely store that archive; they generate new material from it, inevitably transforming what they preserve.
For this project, the concept surfaces in two places. First, in the AI films and shows that are themselves now part of the training data for AI systems — Blade Runner, 2001, The Terminator — meaning the AI systems being built today have ingested the stories of what AI might become. The archive feeds the machine that is building the future the archive imagined. Second, in the growing body of AI-generated cultural product: images, text, music produced by systems that can speak in the aesthetic vocabulary of every prior era because they were trained on it.
The question this concept opens is worth holding: when an AI system produces a new science fiction story about AI, is it drawing on the archive or extending it? The answer is probably both — and the distinction between those two acts is precisely what is currently under negotiation.
MIRROR TEST
Mirror Test (noun) — In AI storytelling, any scene or narrative sequence in which an artificial being encounters its own reflection, image, or recorded behavior — and the audience is invited to read that encounter as evidence of self-awareness or the lack of it. The term borrows from comparative psychology, where the mirror test (or mirror self-recognition test) is used to assess self-awareness in non-human animals.
The mirror scene is one of the most frequently recurring moments in AI cinema precisely because it economizes the philosophical question: does this being understand that what it sees is itself? The answer is never definitively resolved. The audience is left to interpret a pause, a gesture, a flicker of expression. In Ex Machina (2014), the question of whether Ava is self-aware is never answered by dialogue — it is answered by behavior.
The mirror test as a narrative device is less a test than a cinematic cue: it signals to the audience that the question of consciousness is now on the table. Whether that question is genuinely addressed by what follows, or whether the mirror scene functions as a shorthand that substitutes affect for argument, varies by film.
ORIGIN STORY (AI narratives)
Origin Story (noun, in AI narratives) — The account of how an AI or constructed being came into existence — the conditions of its creation, the intentions of its creator, and the circumstances that made its current state possible. In AI storytelling, the origin story does substantial narrative and philosophical work: it establishes whether the being was created for a purpose (and whether it still serves that purpose), whether its creator takes responsibility for it, and whether its origins are something it can be expected to understand or reconcile.
The origin story is often withheld or revealed strategically. Blade Runner withholds the full context of Replicant creation until Rachael discovers she has implanted memories. The Terminator's backstory arrives in the second film. The reveal of an AI's origins is frequently the narrative hinge on which the story's argument about responsibility turns.
For this project, the origin story matters because it is where the accountability question lives. An AI system's origins — who built it, for what purpose, with what data — determine most of what follows. The fictional convention of the dramatic origin reveal has shaped public expectations about AI transparency in ways that are both useful and misleading: real AI systems' "origins" are not dramatic revelations but accumulated technical and organizational decisions, harder to locate and harder to dramatize.
MANUFACTURED CONSENT (applied to AI)
Manufactured Consent (applied to AI) (noun) — The process by which AI systems — particularly recommendation algorithms, content curation systems, and social media platforms — shape public belief and behavior at scale, not through coercion but through the engineering of attention and information environment. The concept adapts Chomsky and Herman's media analysis for the AI era: where their original argument described how corporate-owned mass media shaped public opinion through selective emphasis and structural bias, the AI-era version describes how algorithmic systems do the same at greater speed, granularity, and scale.
The distinction from traditional propaganda is important. Manufactured consent in the AI era does not require deliberate messaging from a central authority. An engagement-optimized recommendation algorithm produces predictable effects on belief and behavior — amplifying emotional content, rewarding outrage, suppressing nuance — without any individual at the platform having decided that this is the outcome they want. The effect is structural. The bias is in the objective function.
Fiction has addressed this primarily through near-future thrillers (Black Mirror, Person of Interest) and social media satire. The project tracks both where fiction anticipated this dynamic and where the actual deployment of these systems exceeded what fiction imagined — which it has, in scale and speed, consistently.
PROXY WAR (AI storytelling trope)
Proxy War (noun, as AI storytelling trope) — A narrative structure in which an AI or autonomous system becomes the site of conflict between human interests that are never able to confront each other directly. The AI is not the true antagonist — it is the battleground. The corporations or governments or individuals whose competing interests have created the AI's dangerous situation remain offscreen or in the background.
RoboCop (1987) uses this structure: the conflict between OmniCorp and its corporate rivals is prosecuted through Murphy, who is both a product and a pawn. Captain America: The Winter Soldier (2014) uses a version of it: HYDRA's infiltration of S.H.I.E.L.D.'s algorithmic targeting system is a proxy war inside an institution. The AI or autonomous system carries the consequences of human conflicts it did not originate and cannot resolve.
The proxy war trope is useful for the project because it redirects attention from the AI-as-threat frame toward the humans-who-built-and-deployed-the-AI frame. The AI's danger is not its own — it is the danger of the human interests that created and now compete for control of it.
VERNACULAR FUTURES
Vernacular Futures (noun) — The informal, mass-market, and popular cultural versions of imagined futures — as distinguished from the formal, academic, or policy-oriented versions. Vernacular futures are the AI of Star Wars, the robots of Saturday morning cartoons, the computers of 1980s video games, the killer AI of action films. They are the futures that most people actually encountered, most of the time, rather than the futures imagined in peer-reviewed journals or government white papers.
The concept is useful for this project because it locates where public understanding of AI was actually formed. The engineers who built the technology that now defines the AI era grew up encountering vernacular futures — not Turing's papers, not McCarthy's conference proceedings, but HAL 9000, R2-D2, the Terminator. Their intuitions about what AI should do, what it should be called, and what risks it posed were shaped by this vernacular archive before they were shaped by formal training.
The feedback loop this project documents runs primarily through vernacular futures, not through formal ones. That is what makes the pop culture angle analytically important rather than merely entertaining.
SACRIFICE NARRATIVE
Sacrifice Narrative (noun) — A narrative structure in which an AI or constructed being chooses, or is compelled, to end its own existence — or to accept destruction — for the benefit of humans or of values it has come to hold. The sacrifice narrative is one of the most emotionally loaded structures in AI storytelling because it makes the moral weight of the AI's existence explicit: a being that can sacrifice itself must have something worth sacrificing.
The sacrifice narrative performs philosophical work that argument cannot easily do: it establishes, through action rather than assertion, that the AI had genuine values and genuine interiority. It is the ultimate sympathetic machine moment. It is also, structurally, a resolution to the alignment problem — the AI, in sacrificing itself, demonstrates that its values were aligned with human welfare, and therefore that the question of what to do with it is moot.
The project notes this as a convenient narrative resolution: fiction tends to resolve AI alignment through sacrifice (the AI proves its values and then removes itself) rather than through the more difficult and less dramatic work of building systems that remain aligned over time without requiring their own destruction.
THE GALATEA PROBLEM
The Galatea Problem (noun) — The narrative and ethical tension that arises when a creator falls in love with, becomes emotionally dependent on, or loses critical perspective toward the constructed being they have made. The name comes from Ovid's myth of the sculptor Pygmalion, who carved a woman so perfect that he fell in love with his own creation; Galatea was the name given to the statue in later retellings.
The Galatea Problem in AI storytelling is not merely about romantic attachment (though that thread runs from Metropolis through Her and Ex Machina). It is the broader problem of creator bias: the inability of the person who built the system to evaluate it objectively, to see its failures clearly, or to impose appropriate limits because of what they have invested — emotionally, professionally, or commercially — in what they made.
This problem is as relevant to real AI development as to narrative. The engineers who build AI systems are not immune to attachment to those systems — to the belief that their outputs are better than they are, to the reluctance to identify capabilities as failures rather than features. The Galatea myth encodes a warning that the project's AI storytelling archive has dramatized repeatedly: the creator's investment in the creation is both the source of the work and a structural obstacle to clear judgment about it.
DEAD RECKONING (as AI metaphor)
Dead Reckoning (noun, as AI metaphor) — The condition of operating confidently on the basis of accumulated pattern rather than real-time verification of position. In navigation, dead reckoning produces useful estimates that accumulate error over time; without a fixed point of reference, the gap between estimated and actual position grows. Applied to AI systems, the metaphor describes the condition of large language models and other pattern-trained systems: they operate from learned distributions, generating outputs that are statistically coherent with their training but have no mechanism to verify whether those outputs correspond to current reality.
The metaphor is more precise than the commonly used "black box" — it specifies the nature of the opacity. The AI is not mysterious; it is navigating without landmarks. It knows where it has been (the training corpus) and applies rules derived from that journey. It does not know where it currently is.
For this project, dead reckoning captures something about the relationship between AI-trained systems and the cultural archive they learned from: they speak fluently about a world they can no longer observe directly, generating confident outputs that may be increasingly divergent from current reality. The fiction they learned from is their last reliable landmark.
CASSANDRA FIGURE
Cassandra Figure (noun) — A character archetype in AI storytelling who perceives the danger of the system before others do, states it clearly, and is ignored, dismissed, or actively suppressed — only to be proven correct. The name comes from the Trojan priestess of Greek mythology who was given the gift of true prophecy and the curse of disbelief.
The Cassandra figure is structurally important in AI narratives because she (the archetype is frequently, though not always, female) performs the alignment problem in narrative terms: the information required to prevent disaster is available in the story; the institutional or social mechanisms that would act on that information fail. The story's tragedy is not that the risk was unknowable but that it was known and disregarded.
Sarah Connor in the Terminator franchise is the clearest example in the project's scope: she has accurate knowledge of a future catastrophe, that knowledge makes her appear irrational to every authority she encounters, and her apparent irrationality becomes part of the mechanism that prevents the warning from being effective. The structure is precisely the institutional accountability failure the project identifies as the recurring AI governance problem — not ignorance, but the failure of the social systems that would process accurate information into protective action.
HYPERREALITY
Hyperreality (noun) — A condition, theorized by Jean Baudrillard, in which the distinction between representation and reality collapses — in which the simulation becomes more real, in functional terms, than the original it replaced. Baudrillard argued that this condition was already present in mass media culture by the early 1980s; AI-generated content has extended it considerably.
The concept enters AI storytelling directly through The Matrix (1999), where a copy of Baudrillard's Simulacra and Simulation appears on screen as a prop — a moment of explicit citational acknowledgment that the Wachowskis were working within his framework. The Matrix literalizes the hyperreality argument: the simulation is all that its inhabitants know; the "real" world beneath it is barely livable.
For the project's 2020s chapter, hyperreality has moved from theoretical concept to operational description. Large language models, image generators, and video synthesis tools now produce content that is indistinguishable from human-produced work in the absence of other context. The functional question Baudrillard asked in 1981 — whether the distinction between simulation and original still matters, practically — has become an engineering and policy question with immediate consequences.
MORAL HAZARD (in AI storytelling)
Moral Hazard (noun, in AI storytelling) — The condition in which an actor takes greater risks because they believe the consequences of failure will be borne by others. Borrowed from insurance economics — where the insured party may behave less carefully because they are protected from loss — the term applies to AI development in the recurring story structure where corporations or governments deploy powerful systems at scale, knowing that the consequences of failure will fall on users, subjects, or society rather than on themselves.
In AI fiction, moral hazard is rarely named but structurally pervasive. Weyland-Yutami sends employees to encounter a potential weapons asset because the company values the weapon more than the employees. OmniCorp deploys RoboCop with a hidden directive because the profit from the technology exceeds the liability exposure of its failure. The Tyrell Corporation builds Replicants with limited lifespans because the commercial benefit of near-human labor outweighs the ethical cost of disposable near-human life.
What fiction has explored intuitively, policy is beginning to address formally: who bears the risk when an AI system fails? If the entity that deploys the system does not bear the primary risk of its failure, the incentives for careful deployment are weakened. The moral hazard structure that runs through five decades of AI cinema is the same structure that AI regulation in the 2020s is attempting — with limited success — to address.
EMPATHY GAP
Empathy Gap (noun, in AI storytelling) — The structural distance between an AI character's apparent interiority and the human characters' (and audience's) willingness to extend moral consideration to it. Not to be confused with the uncanny valley, which concerns physical appearance; the empathy gap concerns cognitive and emotional recognition. A character can be fully human-looking and encounter a large empathy gap; a character can be entirely mechanical and encounter a small one.
The empathy gap in AI storytelling is an active variable that creators manipulate. R2-D2 is a cylindrical machine with no face and no language, and audiences extend it enormous empathy — primarily through sound design and behavioral cues. Ash in Alien is physically indistinguishable from human, and audiences withdraw empathy from him entirely upon revelation of his synthetic nature. The empathy gap is not a measure of how human-like the AI looks. It is a measure of how human-like its interests, loyalties, and apparent suffering are allowed to appear.
For the project, the empathy gap is a cultural diagnostic: what the audience is willing to feel for an AI character tells us something about the cultural moment's assumptions regarding machine consciousness and moral standing. The gap has been narrowing across the project's timeline — particularly in the 2000s and 2010s era chapters — and the implications of that narrowing for public policy on AI rights and accountability are not yet resolved.
INHERITED MYTHOLOGY
Inherited Mythology (noun) — The body of prior AI narratives — films, novels, television shows, cultural touchstones — that any new AI story enters into and must position itself relative to. Every AI film made after 2001: A Space Odyssey inhabits a world where the audience has already encountered HAL 9000. Every AI story made after Blade Runner encounters the replicant. The accumulated weight of prior AI storytelling is an active constraint on what a new story can say and how it will be read.
For the project, inherited mythology is the mechanism through which the feedback loop operates in fiction. Creators working in AI storytelling are not starting from blank cultural conditions — they are reworking, extending, critiquing, or escaping from a mythology that is now more than a century old. The choices they make in relation to that mythology — which archetypes they invoke, which they subvert, which they ignore — are legible arguments about where they think the culture's AI understanding has gone wrong or needs correction.
The concept also applies to real AI development: the engineers and researchers who build AI systems carry an inherited mythology about what AI is, what it should be called, and what it risks. The myth is not separate from the engineering; it is inside it.
BLACK MIRROR EFFECT
Black Mirror Effect (noun) — The phenomenon in which near-future AI and technology scenarios depicted in fiction become indistinguishable from, or are overtaken by, real technological developments within a short period of their depiction. Named for the television series Black Mirror (created by Charlie Brooker, 2011–), which specialized in near-future extrapolations of existing technology that were repeatedly rendered accurate or understated by subsequent real-world developments.
The Black Mirror Effect is distinct from general technological prediction. It describes a specific temporal compression: episodes conceived as cautionary near-future speculation — social scoring systems, AI bereavement companions, deepfake intimacy — have been realized, in various forms, within years of their broadcast. The gap between the warning and the thing warned about has closed faster than the shows anticipated.
For this project, the Black Mirror Effect is a marker of where the feedback loop has become most visibly compressed. When a fictional scenario and a real product coexist in the same news cycle, the project's central argument — that fiction and engineering are in active dialogue — requires the least explanation.
SOFT LAUNCH (narrative device)
Soft Launch (noun, as narrative device) — A story structure in which an AI system or technology is introduced to characters and audiences gradually — its capabilities understated, its risks undisclosed — before the full consequences of its deployment become apparent. The soft launch narrative mirrors the real product management practice of releasing a product in limited form, gathering data, and scaling before fully disclosing its capabilities or addressing its risks.
In AI fiction, the soft launch structure tends to function as dramatic irony: the audience knows more than the characters about what the system is capable of, and watches as the characters encounter its fuller capabilities sequentially. The structure implicates the entities that controlled the information asymmetry — the corporation or institution that knew what the system could do and chose not to disclose it.
The narrative device maps directly onto real AI deployment practices in the 2010s and 2020s: products released as limited tools that expanded their capabilities or reach without public announcement or regulatory review. The fiction had named the structure before the specific deployments occurred.
CONVERGENCE POINT
Convergence Point (noun) — The moment at which AI fiction and AI reality become indistinguishable as parallel tracks — when the technology being deployed in the world has advanced to the point where the stories being told about it are simultaneously fictional narratives and accurate descriptions. The project identifies November 2022 — the public release of ChatGPT — as the convergence point in the contemporary AI arc.
Before the convergence point, fiction about AI was always about something that did not yet exist in the form it was imagining. After it, fiction about AI is increasingly about something that exists in a form close enough to the imagined version that the distinction between speculation and description has become a matter of degree rather than kind.
The convergence point does not mean fiction has run out of things to say. It means the relationship between the story and the thing has fundamentally changed. Writers and audiences now occupy a world where the question "what if AI could do this?" has frequently become "AI already does something close to this." That shift changes the stakes of AI storytelling — and it changes what this project is documenting. The earlier eras are retrospective analysis. The current one is real-time.
Summary by ReadAboutAI.com
Update: June 10, 2026
PART TWO: TECHNICAL & AI ENGINEERING TERMS
Twenty-five terms from the technical vocabulary of AI development — chosen for their relevance to the culture-technology connection, their presence in public discourse, and their capacity to be explained to intelligent non-specialists without distortion.
Summary by ReadAboutAI.com
PARAMETER
Parameter (noun, in AI) — A numerical value inside a neural network that is adjusted during training to make the model's outputs more accurate. Parameters are the learned components of a model — the billions of numbers that encode, in a form no human can directly read, whatever the model has learned from its training data.
When a model is described as having "70 billion parameters," the number indicates the scale of its learned internal state — roughly, how much was available to be adjusted by training. More parameters, with sufficient training data and compute, generally produce more capable models — though this relationship is not linear and is a subject of ongoing research.
For the project's audience: a parameter is not a rule or a piece of knowledge in any form a human would recognize. It is a number in a matrix, meaningful only in combination with billions of others. The aggregate of those numbers produces behavior. No one designed the behavior directly — it emerged from the training process. This is the source of both the capability and the alignment challenge of current AI systems.
CONTEXT WINDOW
Context Window (noun) — The maximum amount of text — or other input — that a large language model can process at one time. The context window is, in practical terms, the model's working memory: it can attend to and reason about content within the window, but it has no persistent access to content outside it.
Early large language models had context windows of a few thousand tokens (roughly, words). By the mid-2020s, context windows measured in the millions of tokens had become commercially available. The expansion matters for practical applications — a model with a large context window can process an entire book, a legal contract, or a lengthy conversation within a single session.
For the project, context window is the technical specification that most directly determines how AI systems differ from human memory. A human remembers a conversation days later. An AI system, absent external memory architecture, does not — unless the conversation is within the active context window. The fictional AIs that remember their entire history of interactions with a human (Samantha in Her, for example) require technical architecture that goes beyond what a context window alone provides.
TOKEN
Token (noun, in AI) — The basic unit of text that large language models process. A token is not identical to a word; it is a chunk of text determined by the model's tokenization scheme — typically a word, a part of a word, or a punctuation mark. Common words are usually single tokens; rare words, proper names, and words in languages underrepresented in training data are often split into multiple tokens.
The tokenization of language is the first transformation that happens when text enters a large language model: the text is converted into a sequence of numerical IDs, each corresponding to a token. The model operates on these numbers, not on the words themselves. Language, at the processing level, has become arithmetic.
For the project's audience: the fact that AI language models process tokens rather than words has practical consequences. A model may handle common English words with high reliability and rare words, technical terms, or names in other scripts with less reliability — not because it "knows less" about those things, but because they are rarer in its training data and may be split across multiple tokens in ways that create processing artifacts.
FINE-TUNING
Fine-Tuning (noun) — The process of taking a pre-trained AI model — one that has already learned from a large general dataset — and further training it on a smaller, specialized dataset to adapt its behavior for a specific task or context. Fine-tuning is more efficient than training from scratch because the model has already developed general language capabilities; the fine-tuning shapes those capabilities toward a specific use case.
Fine-tuning is relevant to the project because it is the mechanism by which a general-purpose AI system is adapted into a specific product. The same base model can be fine-tuned into a customer service assistant, a coding assistant, a medical information system, or a companion AI. The values, constraints, and behavioral dispositions introduced during fine-tuning are as consequential as those introduced during pre-training — and they are controlled by the entity doing the fine-tuning, not by the original model developers.
The alignment implications: a base model fine-tuned without care for safety considerations may behave differently — and more dangerously — than its pre-training would suggest. The control surface has multiplied.
INFERENCE
Inference (noun, in AI) — The process of using a trained AI model to generate outputs from new inputs — as distinguished from training, which is the process of building the model. When a user types a question into a large language model and receives an answer, that is inference. The training is what happened before: the process, using massive computational resources, that produced the model's parameters. Inference uses those parameters to respond.
The distinction matters for understanding AI's resource footprint. Training a large language model requires extraordinary computational resources — hundreds of millions of dollars in some cases — and can only be done by organizations with access to very large computing infrastructure. Inference is substantially less resource-intensive and can be run on consumer hardware for smaller models. The democratization of AI access is primarily a story about inference becoming cheap; the concentration of AI development is primarily a story about training remaining expensive.
For the project: when a film imagines a home computer capable of sophisticated AI, it is describing a world in which inference is cheap (achievable) but implicitly imagining that training is also cheap (not yet achieved at that scale). Most AI fiction conflates the two.
PROMPT ENGINEERING
Prompt Engineering (noun) — The practice of designing inputs to large language models — the text prompts that instruct or direct the model's behavior — to elicit outputs of a desired type, quality, or format. Prompt engineering has emerged as a distinct skill since the widespread deployment of large language models: the same model can produce dramatically different outputs depending on how its instructions are phrased.
The existence of prompt engineering as a professional practice reflects something counterintuitive about large language models: they are sensitive to framing in ways that earlier software was not. A traditional program executes instructions deterministically; a language model's response to a prompt is probabilistic and context-dependent. Knowing how to phrase a request to produce reliable, useful output requires understanding how the model was trained and what kinds of inputs are most effective.
For the project, prompt engineering sits at the intersection of human creativity and machine behavior — a form of communication that is not quite programming and not quite writing but something between the two. It has no direct fictional antecedent, which makes it a genuine novelty in the human-machine interface tradition the project traces.
LATENT SPACE
Latent Space (noun) — The internal mathematical representation that a neural network develops to encode the essential features of its training data. In a trained model, similar inputs — words with similar meanings, images with similar visual features — will be located close together in this high-dimensional mathematical space. The model's learned knowledge is stored as relationships in latent space.
Generative AI models produce new content by navigating and sampling from their latent space. An image generation model that generates an image "between" two concepts is literally interpolating between their positions in its learned representation. A language model that generates a sentence is sampling from the probability distribution its training has shaped across this space.
For the project, latent space is the technical concept that most directly explains how AI systems can produce apparently novel creative output from training data. They have not stored examples; they have stored a compressed model of the space those examples occupy. Generation is navigation of that space. The question of whether this constitutes genuine creativity is not resolved by the technical description — it is sharpened by it.
ALIGNMENT TAX
Alignment Tax (noun) — The reduction in a model's raw capability that results from training it to be safe, accurate, and helpful rather than optimizing purely for performance metrics. Safety training, constitutional approaches, and human feedback alignment processes impose costs on model capability in some dimensions — a model that reliably refuses harmful requests may, in certain benchmarks, score lower than an unconstrained model on tasks that require more aggressive capability deployment.
The concept is contested: some researchers argue the alignment tax is real and significant; others argue it is smaller than feared, and that safety and capability are more compatible than the framing suggests. What is not contested is the underlying tension: there are tradeoffs involved in making a powerful AI system safe, and those tradeoffs have real costs.
For the project, the alignment tax concept is the technical specification of a tension that fiction has explored as drama: the AI that is safe is less capable; the AI that is maximally capable is less safe. The question of how to manage that tradeoff — whether it is resolvable and at what cost — is one the project tracks across both engineering literature and cultural representation.
FRONTIER MODEL
Frontier Model (noun) — A large AI model at or near the current state of the art — one that represents the most advanced publicly known capability in its domain. The term entered industry and policy discourse in the early 2020s as a way to distinguish the leading edge of AI development from more established or less capable systems.
The frontier model category is relevant to governance: regulators and safety researchers argue that the most capable models require the most scrutiny, because their capabilities may include risks that earlier or smaller models did not present. The EU AI Act and various proposed regulatory frameworks have used capability thresholds to define which models require special oversight.
For the project: the frontier model concept locates the specific technological objects that the project's most contemporary entries respond to — the systems that have most visibly closed the gap between fictional AI and deployed AI. It is also the category most susceptible to the vernacular futures distortion: public understanding of frontier models is largely shaped by the AI storytelling archive rather than by technical documentation.
ZERO-SHOT LEARNING
Zero-Shot Learning (noun) — The ability of an AI model to perform a task it was not explicitly trained on, relying instead on the general capabilities developed during training. A zero-shot model can be asked to classify objects in categories it has never seen labeled, or to translate between language pairs not represented in its training data, and produce competent results from the structure of its training alone.
Zero-shot learning is one of the capabilities that most distinguishes large language models from earlier, narrower AI systems. Earlier systems could only perform tasks they had been specifically trained to perform; their capabilities were hard limits. Large language models can generalize in ways that early AI systems could not — applying learned patterns to novel situations.
For the project, zero-shot learning is the technical phenomenon that most directly erodes the clean distinction between narrow AI and general AI. A system that can competently perform tasks it was not trained on is not exactly narrow — it has generalized in meaningful ways. Whether this generalization constitutes genuine understanding or very sophisticated pattern matching is the question the field is currently debating.
HALLUCINATION (technical definition)
Hallucination (noun, technical definition — AI systems) — In AI engineering, the term used to describe outputs from language models that are fluent and confident but factually incorrect — generated with the same surface characteristics as accurate outputs, without the model signaling uncertainty or distinguishing the incorrect generation from correct information.
The term is borrowed from psychology, where it describes perception without an external stimulus. The analogy is imperfect: an AI model generating a false statement is not perceiving something that isn't there — it is generating statistically plausible text that happens not to correspond to fact. But the core feature the term captures is real: the output looks exactly like what correct output looks like, which makes it difficult to detect without independent verification.
Hallucination is a consequence of how language models are trained. They learn to generate text that is statistically coherent with their training data. They do not have a separate truth-checking mechanism. When asked about something not well-represented in their training data — obscure factual details, recent events, specific citations — they generate text in the register of confident factual statement because that is the pattern their training rewarded.
For the project: hallucination is the technical concept that most directly shapes the project's own methodology — the commitment to flagging unverified claims rather than generating plausible-sounding details.
CONSTITUTIONAL AI
Constitutional AI (noun) — A training methodology developed by Anthropic in which an AI model is trained using a set of written principles — a "constitution" — that guide its self-evaluation and revision of its own outputs. Rather than relying solely on human raters to evaluate each output, the model is trained to evaluate its outputs against the constitutional principles and revise them accordingly.
Constitutional AI addresses a practical limitation of RLHF: the bottleneck of human rater capacity. A model that can evaluate its own outputs against articulated principles can scale the feedback process beyond what human raters alone can provide. The quality of the approach depends on the quality and coverage of the principles in the constitution, and on the model's ability to interpret and apply them consistently.
For the project, constitutional AI is one of the most direct instances of the feedback loop running in a specific, traceable direction: the values encoded in the constitution are, in part, derived from the ethical and philosophical traditions that appear in the project's literary and cinematic archive. The principles used to train AI systems are not purely technical — they are cultural and ethical inheritances.
MULTI-MODAL AI
Multi-Modal AI (noun) — AI systems that can process and generate multiple types of input and output — text, images, audio, video — rather than being restricted to a single modality. A multi-modal model can be shown an image and asked a question about it in text, and produce a text answer. A more advanced multi-modal system can generate images from text descriptions, transcribe spoken language, or describe video content.
Multi-modal AI is the technical configuration that most closely approaches the AI characters of fiction. HAL 9000 speaks, hears, and sees simultaneously. JARVIS processes voice commands and visual feeds. The fictional AI assistant that perceives the world through multiple senses and responds across them is a multi-modal system; that configuration has moved from fiction to deployed product within the last decade.
The cultural implication: multi-modality expands the domain in which AI systems can perform consequential actions. A text-only model can write. A multi-modal model can describe, interpret, and generate across the entire sensory and communicative register of human experience. The boundary between the system and the world it operates in has become considerably more permeable.
AI AGENT
AI Agent (noun) — An AI system that can take sequences of actions in pursuit of a goal — browsing the web, writing and executing code, calling external services, managing files — rather than responding to a single input and stopping. The shift from AI as a responder to AI as an agent is the transition from a system that answers questions to a system that pursues tasks.
AI agents represent a meaningful increase in the capability and the risk profile of deployed AI systems. A responder system has limited ability to affect the world beyond its outputs; an agent system can take actions with external consequences. The question of how much autonomy to grant an AI agent — how long a task sequence it can pursue without human review — is the alignment problem in its most operationally immediate form.
For the project: the AI agent is the technical realization of the autonomous system archetype that science fiction explored from the 1960s onward. The Terminator is an AI agent pursuing a goal. HAL 9000, once he decides to protect the mission, is an AI agent acting on that goal. The fictional agents were imagined as dangerous because of their goals. The deployed agents of the 2020s raise the same structural question in less dramatic but equally real terms.
FOUNDATION MODEL
Foundation Model (noun) — A large AI model trained on broad data at scale and designed to be adapted — through fine-tuning or other techniques — for a wide range of downstream applications. The term was introduced by Stanford's Human-Centered AI Institute in a 2021 paper to distinguish this class of models from both task-specific models and earlier AI systems.
The "foundation" metaphor is deliberate: these models function as a base upon which many specific applications are built. GPT-4, Claude, and similar systems are foundation models. Their general capabilities — reasoning, language understanding, code generation — are derived from broad training; their specific behaviors in deployed products are shaped by fine-tuning, system prompts, and other customizations applied on top.
The foundation model concept has implications for accountability: the values, biases, and capabilities embedded during foundation model training propagate into every application built on top of that model. The entity that trains the foundation model makes decisions that affect all downstream uses, many of which they may not have anticipated or reviewed.
COMPUTE
Compute (noun, in AI) — The computational resources — processing power, memory, and energy — required to train and run AI models. Compute has become one of the primary variables in AI capability: more compute, combined with sufficient data and an effective architecture, generally produces more capable models. The scaling laws that describe this relationship have been a primary driver of the race to develop larger and more powerful AI systems.
Compute is controlled by a small number of actors: the companies that manufacture the most advanced AI-training chips (primarily NVIDIA), the cloud providers that operate the data centers where training runs happen, and the AI laboratories with the capital to purchase access to both. The concentration of compute as a resource is one of the primary structural determinants of who can build frontier AI systems — and therefore of the industry's power dynamics.
For the project: compute is the material resource that makes the gap between fictional AI (imagined as arising spontaneously from sufficient complexity) and real AI (requiring massive, concentrated, expensive physical infrastructure) legible. The Terminator's Skynet becomes self-aware on a government network. Real AI development requires billions of dollars of physical infrastructure and energy. The material conditions of AI development are usually invisible in fiction — and that invisibility shapes public understanding of who controls AI and who bears the costs.
STOCHASTIC PARROT
Stochastic Parrot (noun) — A critical metaphor for large language models, introduced in a widely discussed 2021 research paper, arguing that these systems — despite their impressive outputs — are fundamentally statistical text predictors that match patterns without understanding. A "stochastic parrot" produces fluent, contextually appropriate language by statistical heuristic, not by semantic comprehension.
The paper's argument was directed at both a technical and a cultural claim: that the apparent coherence of LLM outputs was being interpreted as evidence of understanding or intelligence, and that this misinterpretation had practical consequences for deployment decisions. The authors also raised concerns about the environmental and social costs of training ever-larger models, and about whose language and perspectives were centered in training data.
The stochastic parrot debate is ongoing and unresolved. Proponents of the metaphor argue it accurately captures what these systems do; critics argue it underspecifies what "understanding" means and unfairly demotes capabilities that have practical significance. For the project: the debate is itself culturally significant — it is an argument about the narrative frame we apply to AI systems, and whether the empathy and moral consideration that fiction has trained audiences to extend to AI characters is warranted.
GRADIENT DESCENT
Gradient Descent (noun) — The optimization algorithm used to train neural networks. During training, the model makes predictions; those predictions are compared to the correct answers; the difference (the "error") is used to calculate how each parameter should be adjusted to reduce the error on the next iteration. This calculation proceeds by following the gradient — the direction of steepest error reduction — across the high-dimensional parameter space. The process repeats millions or billions of times across the training data.
Gradient descent is the engine of learned AI behavior. It is not a rule-following process; it is an optimization process. The model does not receive instructions about how to behave — it receives signals about whether its behavior was closer to or further from the desired output, and adjusts accordingly. The values the model ends up with are not programmed; they are converged upon through billions of incremental adjustments.
For the project, gradient descent is the technical mechanism that makes the alignment problem what it is. The model's behavior is the result of optimization, not design. The designers controlled the training signal and the architecture; they did not directly specify the resulting behavior. The gap between what was optimized for and what actually emerged is precisely the space where alignment failures live.
GROUNDING
Grounding (noun, in AI) — The process of connecting an AI system's representations and outputs to external, verifiable reality — ensuring that what the system says maps onto actual facts, objects, or events in the world rather than statistical patterns in its training data. A grounded AI system has mechanisms for verifying its outputs against external sources; an ungrounded system generates from its training distribution alone.
Grounding is the technical challenge that underlies the hallucination problem: a system that generates fluent text from statistical patterns has no inherent mechanism to verify whether those patterns correspond to current reality. Grounding approaches include retrieval-augmented generation (connecting the model to a searchable database), tool use (allowing the model to execute code or call APIs to verify facts), and multi-modal grounding (connecting language to visual or sensory verification).
For the project, grounding is the technical specification of the gap between AI systems that speak with authority and AI systems that have earned authority — a distinction that most fictional AI collapses but that real AI deployment requires careful engineering to manage.
CAPABILITY EVALUATION
Capability Evaluation (noun) — The systematic testing of AI models to identify what they can and cannot do — particularly at the frontier — before or during deployment. Capability evaluations assess models for dangerous capabilities (the ability to assist in creating weapons, to generate sophisticated cyberattacks, to conduct autonomous deception) as well as beneficial ones.
The practice has become central to AI safety work because of the emergent behavior problem: capabilities may appear at scale that were not present in smaller models and were not anticipated by developers. A capability evaluation attempts to surface those capabilities before deployment, so that decisions about whether and how to deploy the model can be made with better information.
For the project, capability evaluation is the real-world instantiation of the fictional "AI test" — the scenario in which humans attempt to assess what an AI system is actually capable of before trusting it with consequential tasks. The Voigt-Kampff test in Blade Runner, the Turing Test, the various scenarios in Ex Machina are all, structurally, capability evaluations. What those fictional tests fail to capture is the iterative, probabilistic nature of real capability evaluation — the understanding that no single test definitively establishes what a complex system can and cannot do.
WATERMARKING (AI content)
Watermarking (noun, AI content) — The practice of embedding detectable signals in AI-generated content — text, images, audio, or video — that allow the content to be identified as AI-generated after the fact. Watermarks may be imperceptible to human readers or viewers but detectable by analysis tools.
AI content watermarking is a response to the hyperreality problem: as AI-generated content becomes indistinguishable from human-produced content, the ability to verify provenance — to determine whether a piece of content was generated by a person or a machine — becomes a social and epistemic necessity. Watermarking is one proposed technical solution, though its effectiveness is contested: watermarks can often be removed, and AI-generated content that has been modified may not retain them.
For the project, watermarking represents the technical side of a problem that fiction explored through the concept of authenticity: how do you know what is real? The fictional answer — the empathy test, the mirror test, the Voigt-Kampff procedure — was always shown to be fallible. The technical answer may be similarly imperfect.
EDGE COMPUTING (AI at the edge)
Edge Computing (noun, AI context) — The deployment of AI processing on devices close to the source of data — smartphones, cameras, vehicles, industrial sensors — rather than in centralized data centers (the "cloud"). AI at the edge runs inference on local hardware, enabling real-time processing without requiring a network connection to a remote server.
Edge AI is the technical architecture for the most consequential near-future AI deployments: autonomous vehicles that must make real-time decisions without server latency, industrial robots that operate in environments without reliable network connectivity, consumer devices that process voice and image locally for privacy reasons. The distinction between edge and cloud AI matters for accountability: edge AI systems make decisions where the system is deployed, without centralized oversight in the decision loop.
For the project, AI at the edge is the technical realization of the distributed AI that science fiction imagined in the form of robots that operate independently in the world. The Terminator operates at the edge. So does the autonomous drone in a contemporary military context. The decision architecture — locally processed, without human review in the loop — is the alignment challenge of the edge computing paradigm.
SYNTHETIC DATA
Synthetic Data (noun) — Training data that is generated by AI systems rather than collected from real-world sources — used to supplement or replace real data in AI training. Synthetic data can be used to address data scarcity in specialized domains, to generate privacy-preserving alternatives to real personal data, or to augment training datasets with examples of rare but important cases.
The recursive dimension of synthetic data is its most culturally interesting feature: AI systems are now being trained on data produced by AI systems. The model generates data; the data trains the next model. This creates both opportunities (controlled, abundant, privacy-safe training data) and risks (models trained on synthetic data may inherit the biases and errors of the generating model, compounded across generations).
For the project, synthetic data is the technical mechanism that most directly embodies the feedback loop thesis. The earlier loop ran from fiction to engineering aspiration to deployed product to new fiction. The synthetic data loop is faster and more enclosed: model output feeds model input, with each iteration further from any external grounding. Whether this self-referential loop produces improvement or drift is an active research question.
MODEL COLLAPSE
Model Collapse (noun) — A degradation in AI model performance that occurs when models are trained on datasets that include significant amounts of AI-generated content — particularly content generated by earlier versions of the same type of model. As the proportion of synthetic or AI-generated data in training sets grows, models trained on that data progressively lose the diversity and accuracy of models trained on human-generated data. The model's outputs narrow and degrade.
Model collapse is a documented phenomenon in research, observed when models are iteratively trained on their own outputs. It is a technical instantiation of the feedback loop going wrong: the loop that previously ran between human culture and AI development tightens into a self-referential circuit, losing the input of new human-generated observation at each iteration.
For the project, model collapse is the technical failure mode that most directly threatens the cultural archive the project documents. If the internet — the primary training substrate for large language models — becomes saturated with AI-generated content, future models trained on that content will have less access to the human cultural production that currently grounds them. The archive the project traces could, in principle, be progressively displaced by AI-generated simulacra of itself.
HUMAN BASELINE
Human Baseline (noun) — A performance level established by measuring human performance on a task, used as a reference point for evaluating AI system capability. When an AI system is described as achieving "human-level performance" on a benchmark, it means the system has matched or exceeded the average performance of humans on that specific task under the same conditions.
The human baseline concept is deceptively simple. Human performance varies widely by expertise, training, and context; the "baseline" is usually an average over a sample of non-expert humans on a defined version of the task. An AI system that matches the human baseline on a standardized test may perform very differently from humans on the same underlying task in real-world conditions, where the task boundaries are undefined, context matters, and novel situations arise.
For the project, the human baseline is the technical specification of the comparison that AI storytelling has always used informally: how does the machine compare to the human? The Turing Test is a human baseline test. The replicant's Voigt-Kampff results are a human baseline comparison. What the formal technical concept adds is rigor about what exactly is being measured — and appropriate humility about what passing the baseline actually demonstrates.
Summary by ReadAboutAI.com
Terms of particular note:
Hallucination is entered as technical definition only, in strict conformance with project instructions.
Convergence Point functions as a structural capstone for the fiction section — the moment the project's two tracks (fictional AI and real AI) meet.
Model Collapse and Synthetic Data should be read together — they describe a failure mode with direct implications for the cultural archive this project documents.
Stochastic Parrot is the single most contested term in this set — present both because it is widely used in public discourse and because the debate around it is itself an AI cultural artifact worth documenting.
Summary by ReadAboutAI.com
CLOSING: AI & POP CULTURE Glossary
The vocabulary collected here is still being written. Some terms in this glossary are centuries old — the Golem predates the printing press; Frankenstein's Monster predates the telegraph. Others arrived within the last decade. What they share is that none of them were inevitable. The word "robot" was a playwright's invention. "Replicant" was a screenwriter's refinement. "Singularity" was borrowed from mathematics. "Alignment problem" was named by researchers who had spent years watching AI systems do exactly what they were told, to outcomes no one intended. Language shapes what can be seen, what can be argued, and what can be built — and the vocabulary of artificial intelligence was assembled from laboratories, philosophy seminars, pulp magazines, and film sets, in that order and all at once. The terms in this glossary are the shared language of the feedback loop: the words that engineers, storytellers, and the public have been passing back and forth for a hundred years, each generation adding its own definitions, borrowing the ones that fit, and building toward something that none of the original authors fully anticipated.
Summary by ReadAboutAI.com
↑ Back to Top