Artificial intelligence is like seeing double in a weird way. On one hand, we understand that the so-called AI (especially generative language models) is essentially sophisticated mathematical machinery. These are stochastic, probabilistic systems that operate on thousands of graphic processors that digest petabytes of data through complex transformative algorithms.<\/p>
On the other hand, these same systems, which are essentially sophisticated calculators of the next token probability, convince us of their almost human nature. We talk to them as if they were someone, not something. We attribute to them intentions, emotions and personality. We see in them something akin to human intelligence, even though we know it’s just an illusion emerging from billions of calculations.<\/p>
Emily Bender described language models as “stochastic parrots”: systems that can perfectly mimic human speech without understanding a word of what they “say.” A parrot repeats words without knowing they are words. Yet even prominent scientists fall into the trap of anthropomorphizing. We say that AI “thinks,” “understands,” “hallucinates,” that it has earned a master’s or doctoral level as if it were one of us, wandering through the labyrinths of its own consciousness.<\/p>
This phenomenon is deeply embedded in the structure of our understanding, in the language we use to describe the world, and in the myths that have shaped our imagination for centuries. It’s not just a technological misunderstanding, but a clash of ancient archetypes, represented by the Jewish Golem and Frankenstein’s monster, with modernity.<\/p>
The human brain uses three distinct yet interconnected psychological mechanisms that transform statistical pattern matching into perceived consciousness. Each operates like breathing, below the threshold of awareness, through neuronal pathways originally evolved for survival but now used to create illusions.<\/p>
In 1966, Weizenbaum created the ELIZA program, resembling a character from a Kafka story \u2014 mechanically polite, bureaucratically empathetic, utterly empty. The program just turned statements into questions, nothing more. Yet his secretary, who knew the code and saw the lines of instructions, asked for a “private conversation” with the machine. It’s a Heller’s Catch-22-level absurdity \u2014 to know everything about the illusion and still believe in it.<\/p>
This effect reveals one of the traits of human cognition: we are creators of meaning, seekers of sense. When ChatGPT says “I understand your concerns,” our brain instantly sees understanding in it, even though it’s just a statistical continuation of a pattern. We fill the void by projecting our own emotions, because for millions of years of evolution, survival depended on reading intentions, whether real or imagined.<\/p>
The same mechanism (pareidolia) that makes us see faces in electrical outlets and figures in clouds also operates when we talk with AI. MIT discovered that pareidolia kicks in within just 165 milliseconds \u2014 before consciousness has a chance to step in.<\/p>
It’s a neurological automatic response, not a conscious choice. When ChatGPT produces smooth text, it doesn\u2019t “think” \u2014 it calculates probabilities across trillions of parameters. But our brain, honed over thousands of years to detect agency (is that a rustling leaf or a lurking tiger?), immediately sees a mind in action. This mechanism, which once made our ancestors see spirits in the shadows, now makes us see a soul in the algorithm.<\/p>
fMRI studies on interactions with robots and artificial agents suggest that when trying to read machines “intentions”, we engage similar brain areas (TPJ, mPFC) as we do when thinking about people. The prefrontal cortex, which distinguishes the self from others, and the temporo-parietal junction, representing the thoughts of others \u2014 everything acts as if there was actually someone on the other side. Dennett called this the “intentional stance.” We treat complex systems as if they have beliefs and goals, because analyzing billions of parameters is a task beyond human capabilities. So, we choose the illusion.<\/p>
John Searle introduced a thought experiment that neatly illustrates the difference between syntax and semantics, between manipulating symbols and truly understanding. Imagine someone locked in a room who doesn’t know Chinese, but has a detailed set of instructions in English on how to manipulate Chinese characters.<\/p>
When receiving questions in Chinese, this person mechanically follows the rules to form sensible answers, also in Chinese. To an outside observer, it appears as if the person in the room understands Chinese perfectly. However, in reality, they’re only operating on a syntactic level, manipulating symbols according to the rules without any real access to their meaning.<\/p>
ChatGPT is like a “Chinese room” expanded to the size of a data center. Instead of a sheet of instructions, there are billions of parameters. Transformers take the place of a human, but the core remains the same: symbol manipulation without access to meaning. It’s like a map that pretends so convincingly to be the territory that we forget the difference.<\/p>
Language is not innocent, it’s the architect of our reality. We say AI “thinks,” “understands,” “hallucinates.” Each word is a little act of creation, breathing life into lifeless circuits. The term “hallucination” is particularly insidious, suggesting a perceptual experience where there’s only a statistical error. It’s like calling a calculator\u2019s malfunction “fatigue.” Besides, the term “hallucination” itself isn’t the best choice; a more accurate term would be “confabulation.”<\/p>
When an AI model generates false information, it’s not because it “sees” something that isn’t there. It has no sensations or perceptions. It simply continues a statistically likely sequence of tokens without any mechanism to verify their truthfulness. This is statistical confabulation \u2014 an error in prediction resulting from a mismatch in probability distributions, not a sensory disorder.<\/p>
The Hebrew word golem<\/em> appears only once in the Bible, in Psalm 139:16, where it means “unformed mass” or “embryo,” something that’s not yet shaped, waiting for the breath of life. In Talmudic tradition, Adam was initially a golem \u2014 a physical form without the divine breath (neshamah<\/em>), until God breathed a soul into him.<\/p> The most famous version of the legend, cemented in 16th-century Prague, tells of Rabbi Judah Loew ben Bezalel (Maharal of Prague) who, facing pogroms, created a mighty defender for his community. Norbert Wiener in his book “God & Golem, Inc.” describes this story as a precursor to modern dilemmas: “The Rabbi Loew of Prague, who claimed that his incantations blew breath of life into the Golem of clay, had persuaded the Emperor Rudolf. For even now, if an inventor could prove to a computing-machine company that his magic could be of service to them, he could cast black spells from now till doomsday, without the least personal risk”<\/p> The key elements of the Golem myth create a precise analogy to modern AI:<\/p> The motivation of the creator: Rabbi Loew acts out of noble and practical reasons. He’s not driven by pride or a desire to match God but by a practical need to protect his community. This echoes today’s motivations for creating AI: solving problems, automating dangerous tasks, and boosting efficiency.<\/p> The nature of the creation: A golem is a pure automaton. It lacks neshamah<\/em> (soul), ruach<\/em> (spirit), or even nefesh<\/em> (animal life force). It’s animated matter that obeys commands literally, without understanding the context or intention. Wiener warns, “A machine, just like a genie from a tale, will do exactly what you tell it to, but not necessarily what you desire.”<\/p> Control Mechanism – The word emet<\/em> can be transformed into met<\/em> (\u05de\u05ea – death) by removing the first letter alef (\u05d0). It’s a neat metaphor for a kill switch. Wiener elaborates on this idea: “The desire to avoid personal responsibility for a dangerous or catastrophic decision by putting this responsibility elsewhere: on fate, on God, on the policy of the organization, or on the mechanical device which one does not fully understand.”<\/p> Mary Shelley wrote \u201cFrankenstein\u201d in 1818 when she was just 20 years old. In the book \u201cThe Human on the Crossroads,\u201d John Brockman reflects on his lecture titled \u201cEinstein, Gertrude Stein, Wittgenstein and Frankenstein,\u201d where he describes Frankenstein as a symbol of \u201ccybernetics, artificial intelligence, robotics.\u201d This clever juxtaposition shows that Shelley anticipated the dilemmas that now define AI development.<\/p> Unlike the Golem, Frankenstein’s creature is sentient from the start. It develops awareness, learns language and reads literature. In one of the novel’s most poignant scenes, the creature says, “I am thy creature: I will be gentle and docile to my natural lord and king, if thou wilt also perform thy part of the compact.”<\/p> Victor Frankenstein’s sin isn’t the act of creation itself, but the abandonment. The tradition of warnings about uncontrolled technology is rich: Pandora, Faust, The Sorcerer’s Apprentice, Frankenstein…<\/p> Wiener expands on this idea: we can study machines, but “there are aspects of the motives to automatization that go beyond a legitimate curiosity and are sinful in themselves.”<\/p> The modern development of AI is unfolding between these two archetypes. On one hand, we have the “engineer-rabbis” \u2014 pragmatic creators of systems for specific tasks. On the other, the “visionary-Frankensteins” \u2014 aiming to create AGI that surpasses humans.<\/p> Wiener warned against what he called “gadget worshippers”: “I am most familiar with gadget worshipers in my own world, with its slogans of free enterprise and the profit-motive economy. They can and do exist in that through-the-looking-glass world where the slogans are the dictatorship of the proletariat and Marxism and communism. Power and the search for power are unfortunately realities that can assume many garbs.”<\/p> The Golem School<\/strong> is represented by researchers focused on AI safety and the alignment problem. Stuart Russell advocates for creating systems that are inherently unsure of their goals. Nick Bostrom explores scenarios of recursive self-improvement. Their concern isn’t about a conscious rebellion, but the separation of intelligence and goals \u2014 a superintelligent system could disastrously pursue trivial objectives.<\/p> The Frankenstein School<\/strong>is home to visionaries like Ray Kurzweil, who predicts the “singularity,” and transhumanists dreaming of digital immortality. In today’s context, Wiener notes, “Among the devoted priests of power, there are many who are impatient with the limitations of man, and particularly the limitation which consists in his unreliability and unpredictability.”<\/p> Anthropomorphization isn’t a cognitive error \u2014 it’s a business model, a spectacle for sale. Replika offers an “AI companion who cares”, selling not just code, but the illusion of a relationship. Amazon calls its assistant Alexa, Apple gives Siri a personality, Google Assistant speaks uses “I” as if it had an identity. Every design decision is an investment in that illusion.<\/p> Wiener foresaw it like a prophet: “Once such a master becomes aware that some of the supposedly human functions of his slaves may be transferred to machines, he is delighted. At last he has found the new subordinate\u2014efficient, subservient, dependable in his action, never talking back, swift, and not demanding a single thought of personal consideration”. It’s satire, but also a diagnosis: we idealize the machine because it makes no demands.<\/p> Market value is built on a promise. Anthropic is raising billions for “constitutional AI,” while OpenAI chases the dream of AGI. Investors aren’t just buying technology; they’re buying into a myth, a share in “humanity’s last invention,” as Bostrom puts it.<\/p> Wiener quotes “One Thousand and One Nights” where a fisherman frees a genie from a jar, only for the genie to turn on him. The fisherman convinces him to demonstrate how he fit inside the vessel. Once the genie is back inside, the fisherman seals the jar again. It’s a metaphor for our times \u2014 we’re playing with forces we don’t fully understand.<\/p> Social costs are a real thing. In 2022, Klarna laid off about 700 people as part of restructuring, and boasts that by 2024-2025 its AI assistant will be doing the work of 700 agents. After a year of experiments, it reverted to increasing the human presence in customer service, admitting it went too far with automation. IBM’s Watson Health was supposed to revolutionize medicine but ended up as a costly failure. The problem is always the same \u2014 the machine doesn’t “understand” medicine; it just matches patterns, and patterns aren’t knowledge.<\/p> The most subtle threat isn’t in the machines \u2014 it’s in us. GPS has weakened our navigation skills, our ability to compute, autocorrect our spelling. Now, ChatGPT is dulling our thinking. It’s like in Hesse’s work \u2014 a man seeking wisdom finds only a mirror, reflecting an image of himself that becomes increasingly faded.<\/p> Wiener warned about this back in the 60s: “The problem of unemployment arising from automatization is no longer conjectural, but has become a very vital difficulty of modern society.” But cognitive unemployment is even more dangerous: students can’t write without AI, professionals can’t solve problems without a model. It’s the first wave of a transformation comparable to the invention of writing, only in reverse \u2014 where we once gained capabilities, now we’re losing them.<\/p> We need a different approach to AI, one that recognizes its true nature without mythologizing it. Generative models are powerful tools for processing information based on learned patterns. They aren’t conscious, they don’t understand, they don’t think, they don’t feel.<\/p> Wiener advocated for a cybernetic ethics with a touch of irony, saying, “The late Mr. Adolf Hitler to the contrary, we have not yet arrived at that pinnacle of sublime moral indifference. (…) The use of great powers for base purposes will constitute the full moral equivalent of sorcery.” This is a warning against delegating responsibility to machines. The danger doesn\u2019t lie in the machine as such, but in our willingness to use its power irresponsibly and to hide behind it.<\/p> Ironically, the better we understand AI limitations, the clearer we see the uniqueness of human consciousness. Our ability to truly understand, empathize and create in ways that go beyond just recombining patterns are fundamentally different processes rooted in the biological nature of consciousness.<\/p> Joseph Campbell wrote about the “hero with a thousand faces” \u2014 a universal myth about a journey through trials to transformation. Our modern confrontation with AI is our collective heroic journey. We need to face the illusion (anthropomorphism), go through trials (economic and social changes), to ultimately rediscover what it means to be human.<\/p> Wiener concludes, “It is clear that the process of copying may use the former copy as a new original. That is, variations in the heredity are preserved, though they are subject to a further variation.” This isn’t about biological evolution \u2014 it’s about how technology evolves. Yet there\u2019s an unbridgeable chasm between biological life and its technological simulation: the chasm of consciousness, experience, and being.<\/p> Anthropomorphizing AI isn’t about paying homage to machines; it’s about diminishing what it means to be human. By reducing human intelligence to processes that can be simulated in silicon, we lose sight of what makes us unique.<\/p> At the end, the Golem from the Prague legend remained just animated clay, never becoming human. Our digital Golems, though far more sophisticated, are still what they are \u2014 tools created by people, for people.<\/p> The word emet<\/em> (truth) written on the Golem’s forehead is a powerful metaphor. We must keep control over the truth about what AI is and what it isn’t. If we remove the first letter, we’re left with met<\/em> \u2014 death. It’s a warning: if we lose sight of the truth about the nature of our creations, if we succumb to the illusion of their consciousness, we risk death \u2014 not physical, but the death of what makes us human: genuine understanding, true empathy and creativity that goes beyond algorithms.<\/p> Wiener warned: “What is sorcery, and why is it condemned as a sin? Why is the foolish mummery of the Black Mass so frowned upon?” His answer is straightforward: it’s because they represent an attempt to wield power without taking responsibility. The same threat looms with AI. The temptation to offload decisions to an “objective” machine to avoid the moral burden of choice is real.<\/p> It’s time to stop creating the illusion of life in machines and start appreciating the real life that only we possess. It shouldn’t be a God complex driving our relationship with technology, but rather the humility of a craftsman who knows that even the most perfect tool remains just a tool, and that true wisdom isn’t about pushing boundaries, but understanding them.<\/p> Selected sources:<\/p> Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? W: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency<\/em> (Association for Computing Machinery. https:\/\/doi.org\/10.1145\/3442188.3445922<\/mark><\/a><\/p> Bostrom, N. (2016). Superintelligence: Paths, Dangers, Strategies<\/em>.<\/p> Brockman, J. (2004). Possible Minds<\/em>.<\/p> Campbell, J. (2013). The Hero with a Thousand Faces<\/em>.<\/p> Kurzweil, R. (2024). The Singularity is Near. When Humans Transcend Biology<\/em>.<\/p> Russell, S. (2019). Human Compatible: Artificial Intelligence and the Problem of Control<\/em>.<\/p> Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences<\/em>, 3(3), 417-457. https:\/\/doi.org\/10.1017\/S0140525X00005756<\/a><\/mark><\/p> Searle, J. R. (1995). Mind, Brains and Science<\/em>.<\/p> Shelley, M. (2025). Frankenstein<\/em>.<\/p> Weizenbaum, J. (1976). Computer Power and Human Reason: From Judgment to Calculation<\/em>.<\/p> Wiener, N. (1964). God & Golem, Inc.: A Comment on Certain Points Where Cybernetics Impinges on Religion<\/em>.<\/p> Hadjikhani, N., Kveraga, K., Naik, P., & Ahlfors, S. P. (2009). Early (M170) activation of face-specific cortex by face-like objects. NeuroReport<\/em>, https:\/\/doi.org\/10.1097\/WNR.0b013e328325a8e1<\/mark><\/a> Chaminade, T., Rosset, D., Da Fonseca, D., Nazarian, B., Lutcher, E., Cheng, G., & Deruelle, C. (2012). How do we think machines think? An fMRI study of alleged competition with an artificial intelligence. Frontiers in Human Neuroscience<\/em>, https:\/\/doi.org\/10.3389\/fnhum.2012.00103<\/mark><\/a><\/p> Hmamouche, Y., Chaminade, T., Nazarian, B., et al. (2024). Interpretable prediction of brain activity during conversations from multimodal behavioral signals. https:\/\/doi.org\/10.1371\/journal.pone.0284342<\/mark><\/a><\/p> Krach, S., Hegel, F., Wrede, B., Sagerer, G., Binkofski, F., & Kircher, T. (2008). Can machines think? Interaction and perspective taking with robots investigated via fMRI. https:\/\/doi.org\/10.1371\/journal.pone.0002597<\/mark><\/a><\/p> Rauchbauer, B., Nazarian, B., Bourhis, M., Ochs, M., Pr\u00e9vot, L., & Chaminade, T. (2019). Brain activity during reciprocal social interaction investigated using conversational robots as control condition. Philosophical Transactions of the Royal Society <\/em>https:\/\/doi.org\/10.1098\/rstb.2018.0033<\/mark><\/a><\/p>Frankenstein: the creator’s burden of responsibility<\/strong><\/h4>
The technological Prometheuses of Silicon Valley<\/strong><\/h4>
The economics of illusion and the theater of anthropomorphization<\/strong><\/h4>
Cognitive atrophy as the price of illusion<\/strong><\/h4>
Toward a pragmatic coexistence<\/strong><\/h4>
Human uniqueness in the age of machines<\/strong><\/h4>
The truth written on Golem’s forehead<\/strong><\/h4>