Artificial intelligence pioneer Jeffrey Hinton has resigned from Google and joined the growing group of experts concerned about the dangers of artificial intelligence.

The 75-year-old British-Canadian computer scientist known as ‘Spiritual father Artificial intelligence, he fears the potential dangers of his life’s work. Now that he’s gone from Google, he can talk about it, he told Tis the New York Times.

Hinton is at the cradle of the technology now used by AI systems like ChatGPT: computer systems that constantly have to process a large amount of data and detect patterns in it. That process – the so-called Deep learning – It can no longer be ignored in the field of work.

Inspired by the brain, Hinton optimistically led his colleagues to the technique in the 1980s. With success: revolutionizing current AI. For example, one of Hinton’s former students is the co-founder of OpenAI, the owner of ChatGPT. He and two colleagues were also awarded the Alan Turing Award, the Nobel Prize for Computer Scientists, for his work in 2018.

Now, half a century after receiving his PhD from the University of Edinburgh, Hinton’s optimism has given way to great anxiety. He fears that competition among tech companies will lead to the development of AI systems so much that they are moving too quickly.

One of his immediate concerns is that the Internet might be inundated with computer-generated nonsense: generated text, images, and videos that are indistinguishable from the real thing for the average user. Hinton believes that AI could radically change or even take over jobs, such as those of translators or paralegals.

In the fairly long term, it takes into account that systems can write their own programs and execute them effectively. In this context, he referred to autonomous weapons, such as killer robots.

Experts recently sounded the alarm: “Stop developing artificial intelligence”:

“The idea that these things could become more intelligent than humans was believed by very few people,” Hinton explained to the newspaper. “Most people think that would go too far.” He himself thought that it would take at least another 30-50 years. “It should be clear that I no longer think so.”

Frank van Harmelen, a professor of artificial intelligence at VU, says when asked that he’s not surprised by Hinton’s troubling letter. “What surprises me is that he has always been optimistic about technology and now expresses himself in this way, as one of the fathers of technology. This is an exciting turn in a discussion that has been very lively and active among colleagues for some time,” says Van Harmelen, who received his Ph.D. Edinburgh years later than Hinton himself.

see profbreakSimilar to “Predictions about Autonomous Weapons as a ‘nice drinks-table talking point’, but he shares some of Hinton’s concerns”. You don’t have to warn about science fiction scenarios. The way technologies are being used now, there is really a lot of potential for negative consequences in social terms.”

One is misinformation, says Van Harmelen. Sometimes these systems become hooded hoodies Masked speakers called. You can make them say anything and do it very convincingly. They can generate compelling text, images and sounds. And a combination of that.”

Another problem is that the AI ​​itself “hallucinates” things together when the systems guess the answer. For example, he recently asked an AI system about eight books he wrote himself.

I only wrote six of them, but the answer was six books I wrote and then two more I could have written, with titles very authoritative in the range of my experience. If I put this technique at the disposal of ordinary citizens, and those of whom I do not know how to separate meaning from nonsense, it quickly leads to shrapnel”.

Hinton, who categorically does not want his departure from Google to be interpreted as criticism of the tech giant, is calling for artificial intelligence legislation. This call has had a great response among scientists, says van Harmelen, who refers to the so-called artificial intelligence law that the European Commission is working on. AI systems must be classified and subject to rules for each category.

Van Harmelen concludes, “We really need to work in the sciences to advance the science around these systems as well as the building. Even the people who are building these things now don’t understand what they can do.” “It’s like we got on a plane that’s going faster and faster, while we have to figure out how flight works. I’ve never said this before, but it’s going way too fast now.”