Artificial intelligence is developing so rapidly that alarm is now being raised not only by its critics, but also by the people who themselves stood at the origins of today's AI industry.
And it is important to clarify: most of them are not opposed to AI as such. They are saying something else — that the race for ever more powerful models is moving faster than clear rules of safety, control and accountability are emerging.
One of the most famous voices is Geoffrey Hinton, one of the "godfathers" of modern AI. After leaving Google, he became far more open in speaking about the risks: from the mass displacement of people from the labour market to the emergence of systems that may one day become harder to control than we assume.
Another key figure is Yoshua Bengio, also one of the pioneers of deep learning. In recent years he has become increasingly active on AI safety issues and calls on governments not to wait for major incidents, but to create independent oversight and restrictions for the most powerful systems in advance.
In 2023, Bengio supported an open letter calling for a pause of at least six months in the training of systems more powerful than GPT-4.
Among the signatories was also Elon Musk. He supported the idea of a pause in the race of the largest AI models, so that during that time more reliable rules, safety standards and methods of control could emerge.
Apple co-founder Steve Wozniakjoined the same appeal. Later he spoke separately about how developers should be more transparent and bear responsibility for the harm the technology may cause.
Among the best-known intellectuals is the historian and writer Yuval Noah Harari. He has long warned that the particular danger of AI lies not only in "smart machines", but also in the ability of algorithms to mass-generate texts, images and persuasive content. In his view, this could seriously change the information environment and the ways in which people make decisions.
One of the most consistent critics of the uncontrolled race is a professor at the University of California, Berkeley Stuart Russell, co-author of one of the best-known textbooks on artificial intelligence.
His position is fairly simple: the problem arises if we create systems capable of achieving goals more effectively than a human, yet cannot guarantee that their goals truly coincide with ours.
And it is especially telling that even the heads of the largest AI companies have begun to speak more cautiously.
The head of Anthropic Dario Amodei publicly called for a slower and more controlled development of advanced models, so that safety methods could keep pace with their capabilities. He advocates independent audits, common standards and international cooperation.
Even Sam Altman, the head of OpenAI, has repeatedly spoken of the need to regulate the most powerful systems and of the fact that the development of AI must not turn into a race in which speed becomes more important than safety.
Similar ideas were expressed by the head of Google DeepMind Demis Hassabis. The largest AI companies are increasingly discussing joint approaches to controlling the most serious risks.
As a result, a rather paradoxical picture emerges.
Hinton, Bengio, Russell, Musk, Wozniak, Harari, Amodei, Altman and others may differ greatly in their assessment of the scale of the threat and in how much development should be slowed. But they agree on one thing: leaving super-powerful AI entirely without rules is a bad idea.
At the same time, there are also serious sceptics.
Some specialists consider talk of the "end of humanity" greatly exaggerated and remind us that current systems still have many limitations.
Others suspect that it suits large AI companies to keep talk of strict regulation going, because it could make life harder for smaller competitors.
That is why a very serious debate is under way around the topic, and there is no consensus in the scientific community.
But the very fact is noteworthy: just a few years ago the question sounded like:
"What will AI be capable of?"
Now more and more often people ask another question:
"And will we manage to learn to control it before it becomes too powerful?"
