There are two things that Yoshua Bengio and Geoffrey Hinton have in common. The first, which are considered «the sponsors of AI», for having been pioneers in everything that refers to this technology. The second, that both have become deeply critical and even speak of human extinction. Are they exaggerating or really the threat that exists is so great?
In the case of Bengio, he has returned to his fears again. Erre continues that, either measures are taken as soon as possible, or technology could be a very serious problem for the survival of the human being. To such an extent, he says, that has no qualms about comparing the situation of artificial intelligence with 2001: an odyssey in spacea science fiction classic.
The AI already resemble science fiction, says Yoshua Bengio
Shortly after you are familiar with classical science fiction, the name of Hal 9000 will sound to you. It was the protagonist of the film and novel 2001: an odyssey in spacethat Stanley Kubrick immortalized for the big screen. For many, it is one of the best or best film in its kind. But now it also seems to have become something else: a warning.
This is how Yoshua Bengio considers it, for many one of the greatest experts that exist in artificial intelligence, and also one of his greatest detractors for a long time. The expert has been talking about the risks that have tools such as Chatgpt developed with hardly any controls, and now sees similarities between them and the mythical work of 1968.
In broad strokes, Hal 9000 was what is now understood as an artificial superintelligence. That is, he had more capacities than the astronauts with whom he was traveling. The problem began when their orders contradicted themselves and ended up seeing its crew as an inconvenience to meet its objectives. What many fear of AI, go.
What Bengio says in an interview with The Wall Street Journal is precisely that: A machine with great power but without human values can become an existential risk. Follow according to what objectives too strict can end up becoming a threat to human beings, as in the Kubrick movie.
The mysterious objectives of artificial intelligence
The problem of AI, says Bengio, is that anyone can ask him to do something, what he calls «goals», but it is impossible to specify the steps he uses to do it. These would be the «subopjectives.» That is why you do not have to adjust to the final expectations. Or what is the same, be more harmful than beneficial, if you don’t have the necessary care.
And it would not be enough to train the AI with the appropriate instructions? For Bengio, it’s not so easy. He admits that there are already mandates with respect to «safety and morality.» But even so technology is unpredictable, as they tend to demonstrate their hallucinations. In the case of Chatgpt, no matter how much Openai strives to try to avoid them, they always return.
Know How we work in NoticiasVE.
Tags: Artificial intelligence