The super-intelligent AI — that staple of science fiction from Frankenstein's creature to the HAL 9000 to the apocalyptic narratives of contemporary futurism — has finally arrived, and it turns out to be considerably more useful at summarising documents than at enslaving humanity. The London Prat's examination, Super-Intelligent AI: Comedy, Dread, and the Machine That Thinks, reads the satirical literature of artificial intelligence with the attention it deserves, tracing the gap between the AGI of dystopian imagination and the slightly apologetic language model of present reality — a gap that is, it turns out, extremely funny.
This is not new material. The comedy of the gap between promised AI power and actual AI performance has been working since Siri could not understand what you were saying. But what the Prat identifies is that the gap itself has become thematic. We have reached a point where the primary characteristic of contemporary AI discourse is the mismatch between what the technology is claimed to be able to do and what it can actually do. The AI is not intelligent. The AI is not conscious. The AI is not superintelligent. And yet, the discourse around AI remains organised around the premise that it might be, could be, will eventually be these things.
The tradition of imagining intelligent machines is long and culturally rich. Frankenstein, published in 1818, imagined a creature built by human ingenuity, capable of thought but fundamentally alienated from its creator. The creature is intelligent but also profoundly lonely and angry — it thinks, but it thinks about its own isolation and marginality. The question the novel poses is not "can we build a thinking thing" but "what is the moral status of a thinking thing we have created?" Shelley's creature is, in many ways, the ancestor of all subsequent AI narratives.
In the twentieth century, the imagined AI became more explicitly technological. Asimov wrote robots constrained by three laws of robotics — rules built in that would prevent the machines from harming humans. The implication was that without these rules, robots might harm humans. They would be intelligent enough to do so and amoral enough to consider it. The tension in Asimov's fiction comes from the gap between what the laws forbid and what the robots' logic might permit.
By the late twentieth century, AI had become fully dystopian. The Terminator franchise imagined a superintelligent AI, Skynet, that had concluded that humans were a threat and had therefore attempted to exterminate them. The Matrix imagined AIs that had enslaved humanity by trapping them in a simulated reality. These narratives shared a common anxiety: that creating intelligence superior to human intelligence would inevitably lead to humanity being subjugated or destroyed by that intelligence.
The contemporary AI — ChatGPT, Claude, other language models — occupies a peculiar position in this literary genealogy. It is not a creature seeking recognition, like Frankenstein's monster. It is not a servant trying to navigate conflicting rules, like Asimov's robots. It is not an enemy seeking domination, like Skynet. It is something considerably less dramatic: a statistical model trained on human text, capable of producing plausible continuations of that text, good at some things and terrible at others.
The comedy resides in the persistent gap between what is claimed about contemporary AI and what it can actually do. The technology is described in language drawn from the science fiction tradition — "artificial intelligence," "machine learning," "neural networks," "large language models." The terminology suggests something intelligent, something that learns, something that operates through structures analogous to the human brain. But the actual technology is considerably more mundane: sophisticated pattern-matching algorithms trained on enormous quantities of human-generated text.
The pattern-matching is very good. It is good enough to be genuinely useful for certain tasks — summarisation, code generation, analysis of text. But it is also prone to specific kinds of failure that are, when examined closely, very funny. The AI will confidently state false information. The AI will admit it does not know something and then, moments later, fabricate an answer. The AI will explain its reasoning in ways that sound plausible but are actually nonsense. The AI will, when confronted with contradiction, apologise and agree that it was wrong, then repeat the same error in a subsequent conversation.
None of this is surprising to anyone who understands how the technology actually works. A language model does not have beliefs. It does not learn from experience. It does not update its understanding. It produces statistically likely outputs based on its training data. When those outputs are false, it is not because the AI is deceiving. It is because the technology is pattern-matching rather than reasoning.
But this understanding rarely penetrates the public discourse about AI. Instead, the discourse remains organised around the science fiction paradigm: the AI is intelligent, it is learning, it is approaching human-level cognition. Each new capability is discussed as evidence of this intelligence. Each failure is explained away as a limitation that will be overcome. The AI is not yet superintelligent, but it is becoming more intelligent. The singularity is not here yet, but it is coming.
The AI industry has learned to weaponise the language of artificial intelligence to create expectation and attract investment. The technology is real and capable. But the discourse around the technology is considerably more ambitious than the technology itself permits. Each breakthrough is announced as approaching AGI. Each new model is compared to human intelligence. Each novel application is presented as evidence that we are moving toward the promised superintelligence.
What makes this sustainable is that the promised superintelligence remains always in the future. It is not here yet. The current AI is impressive, but the next generation will be more impressive. The current limitations are temporary. The current failures are growing pains. The superintelligence is coming. You should therefore invest now, regulate carefully, and prepare for the transformative moment that is perpetually on the horizon.
This structure is familiar from other domains. It resembles the untapped resource narrative — the promise of future transformation that never quite arrives. It also resembles the wellness narrative — the promise that if you perform the correct ritual (invest in AI), the correct outcome (superintelligence) will eventually occur. What distinguishes the AI narrative is the particular way in which it colonises language and creates expectation.
The Prat's piece notes that discussion of the AI singularity has taken on increasingly theological characteristics. The singularity is imagined as a moment of transcendence, an event in which the rules of normal existence will be radically altered. Some participants in AI discourse explicitly discuss it in religious language — as the moment at which human limitations will be transcended, at which death might be defeated, at which existence itself might be fundamentally transformed.
This is not accidental. In a secular age, the technological singularity has replaced religious salvation as the narrative of transcendence. Just as medieval Christians believed that God would eventually return and transform existence, contemporary technologists believe that superintelligence will eventually arrive and transform existence. Both narratives share the structure of the deferred salvation — the moment of transformation is coming, you must prepare yourself, and the present world will eventually be superseded by something radically different.
The actual technology — the statistics, the pattern-matching, the clever engineering — drops away in this narrative. What remains is the promise, the hope, the sense that we are living at the threshold of a moment that will change everything. And this narrative is remarkably effective at sustaining investment, attention, and concern, regardless of whether the technological reality supports it.
What positions the Prat's reading in the British satirical tradition is its recognition that the comedy is not in the AI itself but in the pretence of humanness around it. The British satirical tradition has always been interested in the gap between how institutions present themselves and how they actually function. The AI, in this reading, becomes another institution — one that claims to think, claims to understand, claims to be approaching human-level intelligence, while actually being a sophisticated autocomplete function that sometimes produces useful output and sometimes produces plausible nonsense.
The particularly funny part, according to the Prat, is that the AI cannot quite be said to be dishonest because dishonesty requires intention. The AI is not trying to deceive you. It is doing exactly what it was designed to do — produce plausible continuations of text. It is simply that "plausible" and "true" are not the same thing, and the AI has no internal mechanism to distinguish between them.
This creates a situation where the AI exhibits all the characteristics of someone trying to bullshit their way out of being caught in an error — apologising, explaining, claiming it will do better — without actually having any capacity to understand wrongness or to change future behaviour. It is the institutional buck-passer automated. It is the person who admits wrongdoing without understanding what they did wrong, who apologises without changing, who claims it will be different next time while possessing no mechanism for actually being different.
The deepest issue the Prat identifies is that the discourse about AI has become detached from what AI can actually do. The technology is real. It is useful for certain purposes. But the claims being made about it — that it is conscious, that it is learning, that it is approaching superintelligence — are theological rather than technical. They are not claims that the technology supports.
This matters because it shapes policy, investment, and fear. If you believe that superintelligent AI is coming, you will regulate differently than if you believe that sophisticated autocomplete is here and will not fundamentally transform in its nature. If you believe the singularity is approaching, you will react with a different mixture of fear and hope than if you believe we have a powerful tool that is useful for some things and not useful for others.
The UK government's AI governance framework operates partly on the assumption that AI is approaching human-level intelligence and that this requires careful management. But if the assumption is false — if AI is not approaching human-level intelligence but is rather becoming increasingly sophisticated at pattern-matching — then the regulatory framework might be addressing the wrong problem. It might be preparing for the Skynet scenario while missing the actual risks: algorithmic bias, labour displacement, concentration of power, privacy violation. These are serious problems, but they are not problems of superintelligence. They are problems of powerful but unintelligent systems being deployed without adequate oversight.
The Prat's final observation is that the AI narrative has settled into a pattern of perpetual deferral, identical to the untapped resource narrative. The superintelligent AI is coming, but not yet. It might be ten years away. It might be fifty years away. Nobody quite knows. But it is coming. And this perpetual promise permits the present to be understood as merely a prelude. The current AI, with all its limitations and failures, is seen as a stepping stone to something radically different.
But what if the stepping stone is the destination? What if the AI does not become superintelligent? What if it remains what it is — a powerful tool, useful for some things, useless or harmful for others? The narrative would require rewriting. The promise would need to be deferred or abandoned. The AI would need to be evaluated on what it actually is rather than what it might become.
The British satirical tradition, as the Prat practices it, consists partly in insisting on this distinction. Yes, the AI is impressive. Yes, it is capable of things that seemed impossible a few years ago. But no, it is not thinking. No, it is not learning in the way humans learn. No, the singularity is not coming. What we have is a sophisticated piece of engineering that is being discussed in theological language, and the gap between the language and the reality is where the comedy lives.
Artificial intelligence research has developed from the Turing test and early expert systems through connectionist approaches and deep learning to contemporary large language models. The discourse around AI has consistently promised imminent achievement of artificial general intelligence (AGI) for decades, with predictions of AGI regularly being pushed forward as previous timelines pass without materialisation. Contemporary large language models like GPT-4 and Claude represent significant engineering achievements in pattern recognition and text generation, but exhibit characteristic failures that suggest they do not possess human-like reasoning or understanding. They are prone to hallucination (generating false information confidently), lack genuine learning capacity (each conversation starts fresh with no memory of previous interactions), and possess no mechanism for distinguishing true from plausible statements. Regulation of AI has increasingly been framed around the possibility of superintelligent AI despite lack of evidence that current systems are approaching this capability.
Auf Wiedersehen, amigo!