The government should not “abandon” its responsibilities and leave the future path of artificial intelligence only to Big Tech, said Aleksander Mądry, the Cadence Design Systems Professor of Computing at MIT and director of the MIT Center for Deployable Machine Learning, at a Congressional panel on Wednesday.
Instead, Mądry said, the government should be asking questions about the purpose and explaining the algorithms used by corporations, as a prelude to regulation, which he described as “an important tool” in ensuring that AI is in line with those social purpose. If the government does not start asking questions, “I am very worried” about the future of AI, Mądry said in response to a question from Rep. Gerald Connolly.
Mądry, a leading expert on intelligence and AI, testifies at a hearing titled “Advances in AI: Are We Ready for a Tech Revolution?” before the House Subcommittee on Cybersecurity, Information Technology, and Government Innovation, a panel of the House Committee on Government Reform and Oversight. Other witnesses at the hearing were former Google CEO Eric Schmidt, IBM Vice President Scott Crowder, and Center for AI and Digital Policy Senior Research Director Merve Hickok.
In his opening remarks, Subcommittee Chair Rep. Nancy Mace the book “The Age of AI: And Our Human Future” by Schmidt, Henry Kissinger, and Dan Huttenlocher, the dean of the MIT Schwarzman College of Computing. He also called attention to it in a March 3 op-ed in The Wall Street Journal of the three authors summarizing the book while discussing ChatGPT. Mace said his formal opening remarks were written entirely by ChatGPT.
In his prepared speech, Mądry raised three general points. First, he noted that AI is “no longer a thing of science fiction” or confined to research laboratories. It is in the world, where it can bring enormous benefits but also poses risks.
Second, he says AI exposes us to “interactions that go against our intuition.” He says that because AI tools like ChatGPT mimic human communication, people tend to unquestioningly believe what such large language models do. In the worst case, warns Mądry, people’s analytical skills atrophy. He also said it would be a mistake to regulate AI as if it were human — for example, by asking the AI to explain its reasoning and assuming the resulting answers are credible.
Finally, he says too little attention is being paid to the problems that will result from the nature of the AI ”supply chain” — the way AI systems are built on top of each other. At the base are general systems like ChatGPT, which can only be developed by a few companies because they are very expensive and complex to build. Layered on top of such systems are many AI systems designed to handle a specific task, such as knowing who a company should hire.
Mądry said this layering raised several “policy-related” concerns. First, the entire AI system is subject to any weaknesses or biases in the large system at its base, and depends on the work of a few, large companies. Second, the interaction of AI systems is not well understood from a technical perspective, which makes it harder to predict or explain AI results, and makes the tools difficult to “audit.” Finally, the mix of AI tools makes it difficult to know who is responsible when a problem results — who should be legally responsible and who should address the concern.
In written material submitted to the subcommittee, Mądry concluded, “AI technology is not particularly well-suited for deployment through complex supply chains,” even if that’s exactly how it’s being deployed.
Mądry concluded her testimony by calling on Congress to examine AI issues and be ready to act. “We are at an inflection point in terms of what AI will bring to the future. Seizing this opportunity means discussing the role of AI, what exactly we want it to do for us, and how to ensure it will benefit all of us. It’s going to be a difficult conversation but we need to have it, and have it now,” he told the subcommittee.
The testimony of all the witnesses at the hearing and a video of the hearing, which lasted about two hours, is available online.