The professor who was considered ‘crazy and alarmist’ and has been correcting predictions about artificial intelligence – 05/11/2023 – Market

“Something unbelievable is happening in artificial intelligence right now — and it’s not entirely for the good,” wrote six months ago Gary Marcus, professor emeritus at NYU (New York University), in the United States, and one of the main voices in the debate on artificial intelligence. AI today.

In his view, the launch of ChatGPT is taking us to the “Jurassic Park moment” of machines: the possibility —as in the Steven Spielberg film— that the situation gets out of control.

“When I wrote this article, I think people thought I was crazy or an alarmist”, said Marcus, in an interview with BBC News Brasil.

But serious problems with this type of artificial intelligence began to proliferate in 2023: in March, in Belgium, a man who frequently talked to the chatbot Eliza, from the company Chai, committed suicide.

His wife maintains that contact with the program led him to take his own life. For the Belgian government, the case is “a precedent that must be taken seriously” and “the danger of using [de inteligência artificial] It is a reality that needs to be considered”.

It was a possible scenario described four months earlier by Marcus in an article for Wired magazine: “Perhaps a chatbot will hurt someone so deeply that the person will be driven to end their life? (…) In 2023, we may see our first kill by a chatbot”.

“I think these systems can be very destructive. And part of the reason for the potential for destruction is that they are unreliable. These programs can make something up and say [a um usuário] that this is a fact. And they can be used by people for that purpose too,” she says.

“The artificial intelligence systems we have now are not well controlled. It’s still not a terrible situation, but people are giving them more and more power. And we don’t know what those systems can do in a given situation.”

seven gloomy predictions

Marcus last year compiled “seven dire predictions” about systems like ChatGPT, including that the newest version of the program would be like a “bull in a china shop, reckless and difficult to control. It will make a significant number of mistakes mind-boggling, mind-boggling, in ways that are difficult to predict”.

In late March, a bizarre case drew attention. One person asked ChatGPT to name academics involved in sexual harassment.

The list mentioned an American law professor, Jonathan Turley. The show said Turley made sexually suggestive comments to a female student during a trip to Alaska and tried to touch her. The response cited as evidence a 2018 report by The Washington Post.

But none of that ever existed: not the trip, not the report or even the prosecution. It’s as if the robot had invented a slur.

OpenAI, the company behind ChatGPT, released a statement saying that the program “does not always generate accurate responses”.

For Marcus, “we have no formal guarantee that these programs will work correctly, even when they do mathematical calculations. Sometimes they are correct, sometimes they are not. Lack of control and reliability are problems that I see”.

“Your traditional calculator is guaranteed an arithmetic answer. But the big language models are not.”

He refers to the system behind ChatGPT, the LLMs (large language models), which store huge amounts of data and generate, through powerful algorithms, approximation answers based on what has already been said before by humans.

In short: an ultra-sophisticated parrot, but one that has no idea what it’s talking about and sometimes “hallucinates” —an AI term that designates an out-of-the-ordinary response, out of line with the programmers’ expectations.

“LLMs aren’t that smart, but they’re dangerous,” says Marcus, who also put the rise in hallucinatory moments on his list of “dark predictions.”

In addition to text generators, programs that manipulate images also evolve rapidly.

Recently, a photo of Pope Francis with a cool silver jacket, made with the Midjourney program, left the internet confused for a few hours: was that image real?

The episode had harmless consequences, but it was a taste of the potential to usher in a permanent gray area between facts and falsifications.

“Unless we take measures, we are close to entering a post-truth environment”, says the NYU professor.

“Which makes everything very difficult for democracy. We need sanctions for those who produce mass disinformation, require watermarks to identify where the information comes from and create new technologies to detect untruths. Just as there is an antivirus program, we need anti-disinformation software.”

Capitalism will not solve these problems

Marcus, 53 years old, is not restricted to the academic world. He sold a company to Uber and became director of an AI laboratory at the transport app giant – he left the post after just four months, at a time when the company was facing accusations of maintaining a “toxic” environment.

Asked whether the famous “move fast and break things” mantra of Silicon Valley and the unbridled competition for markets do not just create dangerous circumstances for the development of artificial intelligence, he says that “it is not possible to expect that capitalism, by itself just go solve these problems”.

He defends that companies are subject to regulation and cites the aviation market as an example that it is something necessary.

“The airline industry in the 1950s was a disaster. Planes crashed all the time. Regulation was good for the airline industry, it got the airline industry to develop a better product in the end,” he says.

“Leaving things to business doesn’t necessarily lead in the right direction. You want business to partner to build what needs to be done. But there’s a reason to have governments, right?”

Understanding with the ‘godfather of AI’

The stance of caution and distrust with enthusiasm for the rapid evolution of AI has not always gone down well.

Marcus’s skepticism has been mocked in other years by his peers (mostly in stings on Twitter), but the tide has turned: several personalities in the field have begun to adopt a different tone.

Geoffrey Hinton, called the “godfather of AI”, announced his departure from Google and said shortly afterwards that he considers the problems with artificial intelligence “perhaps more urgent than those of climate change”.

“Hinton and I have different views on some aspects of artificial intelligence. I corresponded with him a little while ago, I explained my position and he agreed with me, which is not always the case. But the main issue we agree on is control”, he says. .

“I don’t necessarily agree that [IA] is a bigger threat than climate change, but it’s hard to know. There is a lot of established data to try to estimate the risks of climate change. But with artificial intelligence we don’t even know how to calculate those risks.”

“But for me, the chance that these tools will be used to undermine democracies is essentially 100%. Now, whether there’s a chance that robots will take over the planet, we have no idea. It’s reasonable for some people to dwell on this scenario. We are building very powerful tools. We must consider these threats.”

This text was originally published here