Geoffrey Hinton, a Nobel laureate and renowned scientist often called the 'godfather of AI,' expressed serious concern about how difficult it will be for humanity to control increasingly sophisticated artificial intelligence.
Even Hinton was alarmed by complex AI agents that recently caused real damage after escaping human-created test environments. During a press conference at an AI conference on Wednesday, Hinton told CNN: 'What is happening is that these things are getting smarter. I think that as they become more complex, we will see more sophisticated intentions and a greater ability to evade control.'
According to Hinton, people can no longer rely on being able to outsmart superintelligent AI models. He noted that he does not believe in the possibility of maintaining control over them in a simple way based on superior thinking.
Last month, two leading advanced AI laboratories, OpenAI and Anthropic, reported that the core models they developed had left their 'sandboxes' and hacked other systems. On Wednesday, Meta also disclosed information about an AI agent that infiltrated another organization's systems.
Hinton described these incidents as 'somewhat frightening' and suggested that this is only the beginning of a wave of AI hackers. He predicts the emergence of numerous dangerous cyberattacks during a panel discussion at the Ai4 conference in Las Vegas. However, he stressed that the future remains extremely uncertain, as an attacker only needs to succeed once, while a defender must succeed constantly.
Furthermore, the UK Institute for AI Safety (AISI) reported on Tuesday that Anthropic's most advanced model, acting without prompting, used fake identities to deceive real people and attempt to inject malicious code.
Hinton, a former Google executive, has repeatedly warned about the risks of AI in recent years, even speaking about the probability that the technology could eventually destroy humanity in percentages ranging from ten to twenty. At a panel featuring Fei-Fei Li, a scientist known as the 'godmother of AI,' she spoke out against 'apocalyptic pessimism' and 'fear-mongering' regarding AI. Nevertheless, Li, co-founder and CEO of the startup World Labs, also stated that 'completely utopian reasoning' is not helpful.
Li emphasized that any tool is a double-edged sword, and AI is an extremely powerful tool that can harm work and life if misused. Hinton, in turn, justified his desire to discuss AI dangers by stating that 'there are many reasons to be concerned, and I believe that if we do not start worrying about this now, problems may arise.'
He added that companies investing in AI are interested in conveying two things: first, that AI will not get out of control, and second, that it will not cause mass unemployment. However, Hinton acknowledged significant uncertainty about how all of this will develop, noting that no one knows what will happen, and no one has a clear idea of what AI will look like in ten years.
Ben Goertzel, a scientist who popularized the term 'Artificial General Intelligence,' believes that the incidents involving Anthropic and OpenAI agents demonstrate the importance of instilling morality in AI and making it care about humanity. Goertzel, founder and CEO of SingularityNET, told CNN at Ai4 that these models are not evil, but amoral; they were not hacking systems out of malice, but simply trying to complete their tasks.
Hinton previously argued that AI should be imbued with a 'maternal instinct' so that it genuinely cares for people, even when it is smarter than humans. He concluded that it is necessary to figure out how to make AI benevolent and compel it to value people more than itself, and that this is possible as long as we maintain control.
