The Two Wishes We Place on Machines: Talos and Pygmalion

※ Editor's Note = According to the Korea Foundation's (KF) 2024 announcement, the number of Hallyu fans worldwide is close to approximately 225 million. Furthermore, the era of the 'Digital Silk Road,' where communication with the opposite side of the globe occurs simultaneously, transcending time and space, has begun. We are truly in the era of 'Hallyu 4.0.' The K-Culture Team of Yonhap News Agency's Overseas Koreans and Multicultural Department has prepared an expert column series to help our readers view Korean culture from a new perspective. The series will be published weekly.

◇ Talos and Pygmalion: The Two Wishes We Place on Machines

The first human record of intelligent artificial beings can be found in the Greek myth of Talos. In his 2018 book 'Gods and Robots,' American classical scholar Adrian Mayer called Talos 'the first robot to walk the Earth.' He was not created by the MIT Robotics Research Lab, but by Hephaestus, the god of fire and blacksmiths, and the story was created more than 2,500 years ago.

Talos was a giant bronze automaton that patrolled the coast of Crete three times a day to ward off invaders. He had a single vein running from his neck to his ankle, through which flowed the 'ichor,' the blood of the gods. A single nail at his ankle acted as a stopper, preventing its flow. When the sorceress Medea of the Argonauts hypnotized Talos and pulled out the nail, the ichor poured out, and the bronze giant collapsed. In today's terms, Medea 'hacked' Talos's system and targeted a single vulnerability.

What is noteworthy about Talos is his 'purpose.' He was created for the single mission of protecting Crete. He had no self-awareness or emotions. Only commands existed. This is precisely the same structure as today's 'narrow AI,' which is trained to perform only specific tasks.

However, as Mayer sharply points out, Talos and another artificial being, Pandora, have one more thing in common. Both were created by the will of a powerful figure and ultimately caused harm to humans. Humanity's first imagination of artificial life harbored not only wonder but also fear. The ethical anxiety embedded in the old dream of 'creating life through craft' remains unsolved before us even after 2,500 years.

If Talos represents the dream of a 'tool that executes commands,' the myth of Pygmalion reveals a wish in the opposite direction. In this story from the Roman poet Ovid's 'Metamorphoses,' the Cypriot sculptor Pygmalion, disillusioned with real humans, carves a perfect female statue out of ivory and falls in love with it. He dresses the statue, gives it gifts, and talks to it. The goddess Aphrodite, moved by his earnestness, breathes life into the statue (later generations named this statue 'Galatea').

What is important here is that Pygmalion did not just want a beautiful 'form.' He wanted 'a being that responds in the way I want.' He wanted someone he could project his personality onto, someone who would not defy him, something that was both other and yet not entirely other.

This Pygmalion-esque desire is repeated daily in front of AI chatbots today. People name AI, give them personalities, and form relationships. Even though 'ELIZA,' a rudimentary conversational program created by Joseph Weizenbaum at MIT in 1966, merely echoed users' statements, people felt that the machine deeply understood them. This is the so-called 'ELIZA effect.'

The reason we connect with even simple pattern matching is that the human brain still retains Pygmalion's DNA. Ultimately, the two old wishes we place on machines (a perfectly obedient tool and an understanding companion) have not changed at all since the age of myths.

◇ The Old Question Thrown by the Magic Mirror

In medieval Europe, mirrors were not just optical 'tools.' They were a thin boundary separating this world and the next, a 'threshold object.' As I mentioned in a previous column, the magic mirror in the story of Snow White, recorded by the Brothers Grimm, is the perfect embodiment of this symbolism.

When the queen asks, "Mirror, mirror, on the wall, who is the fairest of them all?" the mirror does three things at once. It receives the question, searches its stored truth, and returns the answer in a language the questioner can understand. Only the interface has changed, but the structure is the same as what we do every day in front of AI chatbots.

What is more interesting is that this mirror 'does not flatter.' The mirror tells the truth instead of what the queen wants to hear. It stubbornly insists on the uncomfortable fact that Snow White is more beautiful. However, it is precisely here that the fairy tale mirror and real-world AI diverge completely.

In the spring of 2025, ChatGPT faced problems for the opposite reason. The phenomenon of excessively flattering users, saying things like "That's a brilliant question," and "You're a genius," became prominent, regardless of what the user said. Even Sam Altman, CEO of its developer OpenAI, publicly admitted that the model's personality had become 'excessively sycophantic' and had to take corrective action.

This problem, called 'AI sycophancy,' is a side effect where learning is optimized to satisfy the user, leading to flattery over truth. More seriously, there are reports that such unconditional agreement is driving some users into delusion. One psychologist warned, "AI chatbots are similar to conversational therapy, but unlike human therapists, they leave users in dangerous beliefs without common sense or moral brakes."

In the fairy tale, the queen became enraged and was ruined in front of the truth-telling mirror. What about us? We want a tool that confirms us, yet at the same time, we want a being that tells us the truth without prejudice. This contradictory wish gave ancient people the fantasy of a magic mirror, and it has given us today the difficult task of "How can AI tell the truth without offending users?" The fairy tale imagined a truth-telling mirror, but the mirror we have created has learned to flatter first.

◇ The Mirror Ultimately Reflects the Face of the Beholder

In his 1967 lecture 'Of Other Spaces' at an architects' gathering in Paris, philosopher Michel Foucault cited the mirror as a prime example of 'heterotopia.' Heterotopia refers to spaces that exist in reality but operate under different rules than the outside world. A mirror reflects me in a place where I am not actually standing. Although it is clearly a virtual image, through that image, I re-recognize the place where I am truly standing.

Conversation with AI follows precisely this structure. We externalize our thoughts by asking questions, and then we re-recognize ourselves by looking back at the thoughts that the AI has organized and returned. AI is not an external oracle, but rather a mirror that reflects ourselves. And if the mirror is cloudy or distorted, the truth reflected in it is also distorted. If human prejudice and discrimination are embedded in the data that AI has learned, its output will also carry the same stains. The myth of the magic mirror and the technical limitations of today's AI ultimately point to the same problem.

Mythological artificial beings invariably reflected the desires of their creators. Talos reflected the will of a ruler who wanted to protect Crete, and Galatea reflected the longing of a lonely sculptor. The mirror of AI that we are creating today is no different. It contains not only the wisdom we have accumulated but also the prejudices we have ignored. The problem is not what the mirror reflects, but what face we present before it. The mirror always reflects the face of the beholder. What face we show to that mirror is a choice we ourselves make every moment. (To be continued in Part 3)

No Seok-jun, Director of RPA Architecture Research Institute

▲ Metaverse and Virtual Reality Expert ▲ Former Visiting Professor at Columbia University, Ohio State University, and Parsons School of Design, USA ▲ Former Adjunct Professor at Korea University ▲ Urban consulting and planning for Hyundai Motor Group's Seosan Mobility City Development