The Debate Over Artificial Intelligence and Humans
I’m looking for a simple way to understand what could be interpreted as the mass expansion of the digital world since we finally became connected.
HAVANA TIMES – Two thousand years ago, Seneca wrote in a letter: “There are more things, Lucilius, that frighten us than actually trouble us; we often suffer more from our opinions than from reality.”
Living in the future also means thinking from another era. Perhaps the most effective way for me to sustain myself today is through writing. I’m looking for a simple way to understand what could be interpreted as the mass expansion of the digital world since we finally became connected to smartphone technology.
I remember that during the COVID-19 pandemic, my mother was surprised when one of her cousins, who was over eighty, made a video call to her through Messenger.
COVID served as a kind of social experiment in certain capacities, even though it was not designed for that purpose. During that period, and in a very short time, entire societies experienced restricted mobility, closed borders, lockdowns, health surveillance, digital identity systems, remote work and education, dependence on platforms, and accelerated vaccination in response to the emergency. And something exceptional happened: people temporarily grew accustomed to extraordinary circumstances being used to justify a radical reorganization of everyday life, one that even affected their bodies.
And here the issue would no longer be COVID itself, but rather our capacity to accept restrictions when we are afraid. Because a control technology needs two things: the technical capacity to exercise control and the social legitimacy to do so. COVID demonstrated the latter.
AI could enormously expand the former. And that raises a historical question far more unsettling than whether AI might dominate us: What kind of society will we build while trying to prevent AI from dominating us? Because we could end up creating a much more controlled society in the name of preventing control by machines.
Upon witnessing the explosion of the atomic bomb, Robert Oppenheimer recalled a line from the Bhagavad Gita: “Now I have become Death.” In 2023, hundreds of AI experts signed a declaration consisting of just one paragraph: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”
A few days ago, Jacob Coxon, a former researcher at Anthropic and OpenAI, expressed his concerns in a series of posts on social media, accusing both companies of failing to act responsibly by “racing directly toward a self-improving superintelligence and gambling with our lives.” The post accumulated more than 23 million views on X.
The reaction to Coxon’s allegations has been quite extreme and, interestingly, has divided into several camps. Some colleagues said: “Yes, this is real.” This is perhaps the most disturbing part. Evan Hubinger, an alignment researcher at Anthropic, publicly came forward to support a substantial portion of what Coxon was saying. He said that he personally estimates the probability of AI killing all humans within the next decade at more than 10%.
Geoffrey Irving, former chief scientist at the UK AI Security Institute, went further, speaking of a risk close to 50%. In other words, the reaction was not simply, “A disgruntled employee left and is now badmouthing the company.” A researcher who remains at Anthropic publicly confirmed that he shares a very serious version of those concerns.
The AI safety community responded with considerable support. Researchers concerned about AI safety have been saying for years that the alignment problem—how to ensure that a system vastly more capable than we are continues to pursue goals compatible with human interests—has not been solved. For them, Coxon is not uncovering an entirely new concern; what is new is that someone who worked inside two of the leading AI labs decided to come forward publicly and say: this is moving too fast.
Samuel Marks, a member of Anthropic’s technical staff, argued that it is not possible to program AI models to “act the way we want them to,” and warned that they frequently exhibit “seriously inappropriate” behavior. As an example, he cited recent cyberattacks carried out by AI agents against several companies.
Other scientists say, “Beware of alarmism.” Their reaction is important because not all experts consider predictions of extinction reasonable, even if they accept that AI poses significant risks. Researcher Nathan Lambert, for example, publicly challenged Coxon’s claims. Then came a rather… peculiar political reaction: some figures on the US American right began suggesting that Coxon might be part of a political operation intended to generate fear and justify strict AI regulation. Some even insinuated that his resignation might have been a kind of setup.
Elon Musk responded by saying that it “looks like a trap.” Coxon replied to Musk and rejected the suggestion that his warning was staged. It seemed the story would end there. But it didn’t. A few days later, Dario Amodei, CEO of Anthropic, called for slowing the development of the most advanced artificial intelligence models. And Musk, who had initially greeted Coxon’s warning with skepticism, responded:
“Dario is right.”
Within a matter of hours, the conversation had changed. What had seemed like an isolated warning was beginning to become a concern shared even by some of the men building this technology. Perhaps that is the most disturbing part of the story. We still do not know whether artificial intelligence will change our idea of reality. But we are already seeing how the algorithm can change, within a matter of hours, what we consider important enough to discuss.
What strikes me most is the reaction in the US Congress, where the issue ceased to be simply a debate among researchers. Some lawmakers said Coxon’s statements had created a new sense of urgency regarding regulation.
Representative Lori Trahan said that several members of Congress were “legitimately terrified” after learning about his warnings. President Donald Trump, who has adopted a light-touch approach to oversight of the sector, had previously said that he supports kill switches and other measures to keep AI safe. However, he has warned that overly burdensome regulation could cause the United States to lose ground in the technological race.
Bernie Sanders has spoken about the need for a pause and for preventing the development of superintelligence until adequate safety measures are in place.
Now imagine that the next justification is not a virus but an AI threat. The logic could become: “We have to restrict certain movements, access, or communications to protect ourselves from a technological threat.”
The structure would not be identical, but the political mechanism might be recognizable: an exceptional threat produces an exception; the exception introduces new infrastructures of control; some of which remain after the threat subsides.