“AI is no longer a science fiction thing or a technology confined to research labs. AI is a technology that is already in use and widely adopted. It will dramatically change our lives; before us,” said the professor at the prestigious university during a hearing before the House Subcommittee on Oversight.
The Polish scientist and director of the MIT Center for Deployable Machine Learning performed alongside, among others, former Google boss Eric Schmidt at a hearing devoted to the development of AI and the preparation of the United States. United with the “technological revolution” associated with it.
Read: A big step on a long road
Mądry pointed out that the new generation of AI, the so-called generative artificial intelligence, which has appeared in recent years and months – we are talking here, among other things, about ChatGPT, Bing Chat applications or image generators DALL-E and Midjourney – it differs not only in superior capabilities, but also in much higher accessibility and price for an ordinary recipient.
As he pointed out, this implies both enormous opportunities in almost all sectors of the economy, but also many risks, particularly related to the vulnerability of these systems to manipulation and their use for misinformation, the repetition and perpetuation of biases, and their general unreliability.
“Even AI experts have a very poor understanding of the magnitude of these problems and are only just beginning to find ways to uncover and mitigate them,” the expert said.
Read: Controversial city of the future or excessive ambitions of the Saudis? Will the line change the future of urban planning?
He also pointed out that new problems could arise by creating “AI supply chains”, that is, by creating new applications based on and combined with different types of AI. As he said, because creating basic AI systems is expensive and requires huge investments, only very few entities are capable of creating them, which can give them a dominant position. As he predicts, although in the near future the development of artificial intelligence will lead to a huge explosion of new AIs, almost all of them will be based on a few basic programs.
Teacher. Mądry also added that the combination of different artificial intelligence systems can lead to unpredictable changes in their characteristics.
“This is an even more complicated problem, because AI experts cannot explain how individual AI systems arrive at their conclusions, and it will become much more difficult if the systems overlap” – said Mądry.
He also called for the establishment of legal and technical regulations to be able to respond to problems related to decisions made by artificial intelligence systems.
When asked by members of Congress if he was worried that the rapid development of AI would lead to a catastrophe, Mądry replied that while he was not afraid of the “Terminator+” scenario, he was concerned, for example, by the increased use of AI in social media, which could further reinforce the current problems of these networks by creating “social media on steroids”. At the same time, he said strong legal regulation was needed in this area.
“We need to make sure the companies we’re talking about are not just profit-driven, but realize they have some responsibility as well,” the professor said. As he added, the lack of regulation is his biggest concern when it comes to AI development.
Schmidt, the former head of Google, gave a different response to his concerns, saying he was most concerned about the potential divergence of interests between humans and artificial intelligence.
“This is a situation where a human has a set of incentives, a computer has been trained to pursue different outcomes, and society has other interests” – said the expert. He added that while the Pentagon wants AI-based weapons to make decisions automatically, at this time those systems are not ready for such “final” choices and may never be.