Project Syndicate
AI giants are promising a superintelligent future, but could the dream of AI saving humanity be hiding a dangerous race for profit and power?
Which AI Myth Will We Embrace?
Author: Richard K. Sherwin
NEW YORK—It is telling that Anthropic chose to call its most powerful AI model “Mythos.” In ancient Greek, the word denotes an original reality that we can imitate or embody through the stories we tell ourselves and one another. Pliny the Elder tells us that the fifth-century Greek painter Zeuxis rendered grapes so realistically as to fool even the birds.
But stories do not only imitate a pre-existing reality. They also engender new realities by expressing what is possible. For example, the revolutionary opening words of the American Declaration of Independence (“We hold these truths to be self-evident”) gave pride of place to individual freedom regardless of class, property ownership, or religious or political belief. Through this mythos, the Founding Fathers laid the foundation for a way of governance conceived in liberty and equality.
The frontier AI models embody both kinds of storytelling. On one hand, they imitate past human thinking, responding to user prompts by seeking out relevant patterns from the vast database of previously archived material that they were trained on.
But they also fabulize a new reality. The founders of Anthropic and OpenAI seem to subscribe to the same vision of the imminent arrival of an all-powerful AI god that will deliver us to a transhuman utopia. Yet there is good reason to question whether this myth is merely a pretext justifying the rush for market dominance, wealth, and power.
Some figures, like Anthropic co-founder Christopher Olah, are now seeking to provide a crash course in morality to the company’s models. But it is a little late in the game for that. If Anthropic sincerely sought to align Claude’s behavior with serving the good of humanity, it would have first trained its engineers in the ways of wisdom and goodness.
Instead, Claude was trained to offer a service: to provide information, to solve problems, and to perform useful tasks autonomously. AI is being sold as a profit-making tool, as a source of cheap, hyper-efficient cognitive labor. Any morality introduced after the fact is mere window dressing—a gloss on the underlying economic raison d’être and a sop to a public that is growing increasingly apprehensive about the technology.
The reality a myth ultimately creates depends on the cultural context in which it functions. If a culture champions AI as a beneficent superintelligent godsend, the dangers associated with its creation (like AI agents failing to act in our best interests) become more likely. But if a culture objects to unbridled arrogance and insists on a counternarrative of social cooperation, it is more likely to generate a check on hubris and the dangers it can create.
Such concerns have deep roots in the Western imagination. The pre-Socratic philosopher Heraclitus wrote: “The sun will not overstep his measures; if he does, the Erinyes, the handmaids of Justice, will find him out.” These are the same Erinyes (or Furies) who pursue Orestes for the murder of Clytemnestra in Aeschylus’ Eumenides. Orestes slew his mother to avenge his father, the great Achaean general Agamemnon, whom Clytemnestra murdered for having sacrificed their daughter, Iphigenia. Her young life was the price the prophets said must be paid for a fair wind to launch the Achaean fleet against Troy. So it goes, for generation unto generation.
It is too soon to know what moral limit, if any, Anthropic’s Mythos will imitate or embody. History suggests we should be wary. It took an awareness of human irrationality, the madness of “mutual assured destruction,” to bring about international accords prohibiting nuclear proliferation in the 20th century. Social cooperation triumphed in the face of an irrational quest for power and dominance.
Human nature has not changed, but the threats that we face have. We know that the dangers associated with out-of-control AI are real, because we have already witnessed them. Fortunately, the spontaneous hacking of Hugging Face by OpenAI’s agents caused minimal harm. But similar episodes have since been revealed, and experts warn that we may not be so lucky in the future. From AI agents gaining unauthorized control over critical infrastructure to carrying out the wishes of terrorist groups or designing novel pathogens, no threat can be ruled out.
As the legal scholar Antara Haldarreminds us, social cooperation and the preservation of the commons have always provided a bulwark against the calamitous effects of human arrogance, greed, and will to power. Though this foundational narrative has been overshadowed by classical economics’ tale of the self-interested “rational man,” it has not been extinguished. Stories that imitate past acts of sacrifice and service on behalf of the group persist, as do those who imagine new moral possibilities to come.
For example, advocates of social cooperation today might call our attention to the courage and diplomatic acumen that it took to prevent nuclear Armageddon after World War II. What would it take to manage the threat posed by AI? One can imagine global coordination among independent fact-finding commissions that are qualified and equipped to acquire first-hand knowledge of the dangers specific models pose. Universal monitoring might then be supplemented with internationally recognized guardrails—minimum mandatory safety standards—capable of pre-empting threats before they arise and effectively managing them when they do.
The mythos (or foundational stories) we tell ourselves will ultimately determine our fate. They need not come only from industry leaders whose interest lies in regaling us with visions of new scientific and technological horizons. Prudence, accompanied by a counternarrative emphasizing social cooperation, might be all that can keep us on the path of survival and mutual flourishing.
Credits: TCA, LLC.