The Ancient Forerunner of AI

  • Themes: Culture, Technology

The myth of Talos gave us the original AI dilemma.

Talos, the bronze giant.
Talos, the bronze giant. Credit: Everett Collection Inc / Alamy Stock Photo

The first robot was imagined in Greek mythology more than 2,500 years ago. Talos was a giant bronze automaton fabricated by Hephaestos, the blacksmith god of innovation and technology. Commissioned by Zeus to defend the island kingdom of Crete, ruled by King Minos, Talos actually fits the definition of a robot. Talos was self-moving, with inner workings (an internal ‘artery’) and a power source (ichor, life-fluid of immortal gods). This vivisystem was sealed with a bolt on his ankle. Half-machine, half-human, Talos marched around the island at high speed three times a day, on the lookout for strangers. ‘Programmed’ with something like Artificial Intelligence, Talos was able to interpret and interact with his surroundings. When he spotted invaders approaching by sea, he hurled boulders to sink the ships. At close range, Talos heated his bronze body red-hot and crushed men to his chest, roasting them alive.

In the myth, Jason and the Argonauts were doomed to become victims of Talos when they anchored on Crete. Fortunately for the mythic heroes, the sorceress Medea figured out how to demolish the giant android, with her knowledge of his inner works and a clever guess that the robot had human-like traits. She played on his vulnerabilities—his emotions and the bolt on his ankle. Medea convinced Talos that she could make him immortal, but only if he allowed her to remove the bolt that sealed his internal workings. Unaware of his own nature, susceptible to persuasion, and fearing death, he agreed. When Medea and Jason unsealed the bolt, Talos’s ichor bled out and he was destroyed.

The myth of Talos, from the time of Homer, evokes the practical and ethical dilemmas of modern AI. In posing knotty questions about links between tyranny and technology and how to control automatons, the story foreshadows the qualms that surround new AI technologies. Medea’s destruction of Talos shows that robots and AI do not always behave as expected by their builders and users. Like Talos, they might make disastrous decisions on their own. No matter how advanced the technology, there is always the danger of a techno-wizard like Medea hacking the system by exploiting its weaknesses.

A security system that dispatches artificial agents created by superior intelligence to automatically perform preordained duties triggered by specific situations, as Talos was meant to do, may seem desirable. But what if the situation shifts or it becomes necessary to interrupt the automatic response? How can humans control, disable, or destroy a powerful, unstoppable machine? How does one incapacitate AI once set on track?

In Plato’s account of the myth, Talos carried a bronze tablet of laws and rigorously enforced them on Crete. In 1596, the notion of technological justice appeared in the poet Edmund Spenser‘s allegorical epic The Faerie Queene. Spenser created a Talos-like figure—an android knight made of iron—that he named Talus. A goddess dispatches Talus to help Sir Artegall, a righteous cavalier, in his quest to mete out justice to villains. The seeker of virtue and his automated enforcer-squire travel around the countryside bringing law and order to the realm.

Invincible and swift-acting, Spenser’s robotic Iron Knight takes his job literally. An inflexible killing machine without mercy, Talus carries out an inhumane, unbending form of justice. Unmoved by empathy or pity, with no capacity for compassion, the Iron Knight imposes Iron Law. The grim automaton has no interest in wrongdoers’ experiences, current situations, backstories, or motives. Notably, on their journey to serve justice, Artegall and Talus encounter legal cases that strike contemporary chords for modern readers. The first episode is a domestic murder case, in which Talus relies on circumstantial evidence to identify and punish the guilty party. The second case involves extortion and the extortioner’s attempt to bribe Talus, who deals out a severe sentence. Next the Iron Knight banishes ‘enemies of society,’ ‘lawlesse multitudes’ of outlawed religious sects. In other episodes, Talus deals harshly with defrauders and imposters, settles disputes over property rights, kills a female monster, and ‘inflicts grievous punishment’ on rebels against the English crown. Spenser describes the Iron Knight violently beating a mob of women demanding ‘unnatural’ political and legal rights. Talus goes on to slaughter another ‘rabble’ of women who have captured Artegall, due to his own foolish error. Here, Talus turns against Sir Artegall, refusing to rescue him because of a legal technicality in the Iron Knight’s original ‘contract’. In the last stanzas, we see Artegall preventing his robot squire from attacking a slanderer with excessive violence.

Spenser’s ‘legend of justice’ is a medieval allegory about the rule of Elizabeth I, but the poet’s retooling of the bronze robot-guardian of Greek myth into a merciless iron robot-vigilante offers a striking catalyst for thinking about the wisdom of giving AI a role in our judicial system. One can envision employing AI to find precedents, search DNA archives to identify victims or suspects, or predict crime patterns. Some US courts already use algorithms to set bail. An AI system’s access to volumes of legal material and sentencing based on equations might seem to promise fairness under the law. But Talus’s actions raise questions about whether AI should be ‘programmed’ with ethical values. Should we trust AI to become artificial moral agents? We have already seen that AI is infected by biases embedded by programmers and the data they feed it. Can moral values be replicated humanely in algorithms? Will ‘objective’ AI judges understand forgiveness or compassion better than humans could? Can AI distinguish remorseful wrongdoers from pathological criminals? Or grasp extenuating circumstances and nuances of guilt or innocence?

Moreover, both Talos and Talus were ancient forerunners of the Black Box dilemma of AI and machine learning. AI entities’ access to unthinkably vast stores of data means that they will make decisions based on unfathomable processes, thus becoming inscrutable to makers and consumers. Remarkably, the first inklings of such issues arose millennia before sweeping advances in technology have made these questions so urgent.


Adrienne Mayor