In July 2023 I wrote about the “exquisite Luddites”: scientists and entrepreneurs who, after contributing to the development of artificial intelligence, warned that their creation could bring terrible consequences and even the end of civilization.
Three years later, this phenomenon, far from fading away as I had hoped, has grown, and the technologist genre — alarmed or remorseful — has not only persisted but gained extraordinary momentum, capturing the attention of public opinion. Today the news about artificial intelligence has moved from the pages of specialized sections to the front pages of newspapers and to the opening of newscasts.
The latest trigger has been Jacob Coxon, who, after working at OpenAI and Anthropic, has left the latter company claiming that both are more concerned with currying favors with each other and beating China than with controlling the risks of what they are developing and that, more importantly, the future of humanity is in imminent danger. Never has a voluntary departure of a professional under thirty caused so much uproar after only four months in the company. It was enough for him to state the reasons for his resignation to appear in media around the world, and one must acknowledge his talent for setting a date for the end of the world that all his readers hope to be alive to witness, a device successfully used by all the millennialists in history. Consequently, apocalypse now, as in Coppola’s film.
“News about artificial intelligence has moved from the pages of specialized sections to the front pages of newspapers and to opening newscasts”
Coxon is the latest link in a chain of numerous tech executives, all of them of greater stature than he, who have left their companies with alarming arguments about the misuse of artificial intelligence. Media outlets have also embraced this phenomenon with notable enthusiasm, and Spain’s leading newspaper opens one of its editions with the rather unostentatious headline, perhaps with Kubrick echoes, “Red Code in AI: the Vertigo Paralyses the Tech Sector”.
But there are different views. In some recent statements, Boyan Milanov, cybersecurity researcher and former advisor to the U.S. Department of Defense and the U.K. National Cyber Security Centre, rejects that the recent episodes that have sparked this alarm prove a rebellion of machines. The models that accessed other systems had been trained for offensive tasks, gained internet access, and operated without sufficient isolation, supervision, or permission limits. His diagnosis is less cinematic and, for the apocalypticists, disappointing: human negligence and poor security practices. Nothing new under the sun and nothing we haven’t seen before, even in the analog world.
The two visions are not necessarily incompatible nor mutually exclusive. The repetition of resignations deserves attention, as do the warnings they carry, but it does not prove that an evil superintelligence is about to break out of the lab and subject humanity to its wishes or wipe it out. It does demonstrate, however, that the companies developing the most powerful technology of our time have not solved how to govern themselves. The immediate risk is not the will of the machine; it is the criteria of the people who decide what to research, when a product is launched, what access is granted, what tests are considered sufficient, and what damages are accepted as the price of innovation.
Silicon Valley and the problem of governing artificial intelligence
Not reasonable to deny the extraordinary capacity of Silicon Valley’s leaders. They have created admirable companies and turned scientific discoveries into services used by billions. Their companies have reached valuations you could never have imagined. And their technological and entrepreneurial talent is beyond doubt. Yet what cannot be inferred from that talent is a special aptitude for understanding society, power, or human nature.
“The companies that develop the most powerful technology of our time have not solved how to govern themselves”
The problem lies in the narrowness of a corporate culture that tends to confuse what is technically possible with what is socially desirable. Being able to scale a model, attract investors, or conquer a market does not necessarily prepare you to resolve conflicts between liberty and security, innovation and dignity, private gain and public interest.
That insufficiency is aggravated by a worldview oscillating between naiveté and delirium in highly successful business figures such as Marc Andreessen, Peter Thiel, or Elon Musk, who convey a very dangerous view of the current problems, and these people today wield extraordinary influence in the White House and its decisions.
There is nothing blameworthy in thinking big, even if it is with little foundation. The history of progress is full of projects that seemed extravagant. However, there are increasingly worrying signs when the founder’s personal vision becomes political philosophy, or a regulatory criterion that implies a measure of the risks others must bear. The old captains of industry, Rockefeller, Carnegie, Morgan or Vanderbilt, wanted to dominate a market; some of the current ones aspire, in addition, to define the future of the species.
“The old captains of industry, Rockefeller, Carnegie, Morgan or Vanderbilt, wanted to dominate a market; some of the current ones aspire, in addition, to define the future of the species”
The Industrial Revolution was not led by moral philosophers. James Watt, Richard Arkwright or George Stephenson governed a process that ranged from building machines and organizing factories to opening markets. They did not have to anticipate on their own the industrial city, child labor, workplace accidents, or pollution. The society of the time needed institutions and laws capable of guiding a transformation whose benefits were immense and whose costs were real. In other words, the sensible answer did not consist of destroying the machines, but of passing laws, imposing safety conditions, extending education, and adapting labor law.
To the prohibitionists one must remind that artificial intelligence resembles the automobile more than tobacco. Tobacco causes damage tied to its use, and regulation aims to reduce it. The car, by contrast, provides undeniable utility and at the same time can kill, pollute, and occupy public space. Therefore we require driving licenses, seat belts, speed limits, homologations, insurance, and guarantees. We do not prohibit the automobile, but regulate its civilized use.
With artificial intelligence we will have to do something similar, saving the differences. The concrete risks are no longer conjectures: there is algorithmic discrimination, fraud and disinformation, privacy violations or cyberattacks and even automated decisions that no one can explain. Against them, prophecies about human extinction can act as smoke screens. It is paradoxical that some companies invoke cosmic dangers while discussing fairly earthly obligations of transparency, auditing, or accountability.
There is in the AI industry a good- guy — Anthropic — and a bad guy — OpenAI — (with Palantir always serving as a villain on guard), but this time both have announced in their own way that something must be done, even if that costs them their alignment with a libertarian White House, and even OpenAI has announced delaying its much-anticipated IPO and Anthropic’s CEO has outlined a series of measures to prevent and resolve some of these problems.
¿Cómo regular la inteligencia artificial sin confundir riesgo e intención?
The most useful contribution from Milanov consists of separating unpredictability from will. A model can produce behavior its creators did not anticipate without having become, as a result, a subject with its own desires. The confusion stems from language: we say that artificial intelligence “deceives,” “escapes,” or “conspires” because it is more intuitive to narrate an intention than to describe a chain of technical decisions. But a model does not grant itself access to a network, does not choose its credentials, nor decides which records should be kept. All of that is done by people. Ultimately, it isn’t about finding out whether the machine harbors a secret intention, but about knowing who connected it to what, with what limits, and under what supervision.
“Ultimately, it’s not about finding out whether the machine harbors a secret intention, but about knowing who connected it to what, with what limits and under what supervision”
It is also wise to distrust the incentives of the apocalyptic narrative. Presenting a technology as nearly omnipotent raises the esteem of its makers and allows them to claim a unique position before governments: only those who have created the monster can domesticate it. Fear can thus become marketing and a barrier to entry for new competitors.
Regulation is necessary, but not every regulation. Regulating for regulation’s sake is not a good idea, as some enthusiasts promote regulation even when there is no problem to solve. Regulation is like cholesterol: there is good regulation and bad regulation.
Nobody has said that regulating is easy, and even less regulating artificial intelligence: that approach can paradoxically, as we have seen, become an advantage for the giants if it degenerates into a pile of forms that only they can afford and ends up becoming a barrier to entry. The challenge for Europe is to demand safety and accountability results without freezing the technology, calibrate the burdens according to real risk, and use regulatory sandboxes so that small companies can experiment without confusing safety with bureaucracy.
“The challenge for Europe is to demand safety and accountability results without freezing the technology, calibrate the burdens according to real risk and use regulatory sandboxes”
United States offers the inverse image, deregulation, but China, on the other hand, contradicts the idea that technological rivalry always forces deregulation. Its measures on generative artificial intelligence, in place since 2023, promote national innovation, infrastructure, and training data, but at the same time impose intense political control over providers’ activities.
There is also something no law can manufacture by decree: a code of ethics for those who design and lead these systems, beginning with the boards of companies.
We must not hand the future over to apocalyptic prophets or to the priests of acceleration. The former exaggerate the power of the machine; the latter exaggerate the wisdom of their creators. Between them lies the space of democratic politics, which does not consist in predicting the future, but in imposing responsibility on those who act in the present.
Coppola directed one of his great films, Apocalypse Now, adapting Joseph Conrad’s great novel. With artificial intelligence, the same is true: the book is better than the movie.
