Computer scientist Eliezer Yudkowsky, who has predicted “we are all going to die” if a superhumanly smart artificial intelligence is created under the current circumstances, said he nevertheless understands the challenges lawmakers face in regulating the system.
“The analogy I sometimes use is that AI is like nuclear weapons, if nuclear weapons spit out gold, and spit out more and more gold as you made them bigger, until finally they reached a threshold nobody could calculate in advance, and then exploded the entire world,” Yudkowsky, a co-founder of the Machine Intelligence Research Institute, wrote in an email to CBC News.
“That’s an unusually difficult situation to regulate.”
It’s this regulatory conundrum — AI may produce many benefits but left unchecked could negatively impact society and perhaps pose a significant threat to humankind — that some U.S. politicians faced at a U.S. Senate Judiciary Committee hearing on Tuesday.
‘Cause significant harm’
The hearing featured testimony from Sam Altman, the head of the artificial intelligence company OpenAI, that makes ChatGPT, who advocated for a series of regulations to confront the risk of increasingly powerful AI systems that even he acknowledged could “cause significant harm to the world.”
But Congress has yet to regulate some of the Big Tech companies like Meta and Google. And the issues Altman and others raised at the hearing on AI also illustrated the challenges facing regulating that industry, some experts say.
“I think that one of the biggest problems with AI regulations is [defining] what is AI, anyway,” said Matthew O’Shaughnessy, visiting fellow of technology and international affairs at the Carnegie Endowment for International Peace.
O’Shaughnessy said some of the panellists and senators at the hearing, which included the testimony of IBM’s chief privacy and trust officer Christina Montgomery and AI expert Gary Marcus, a professor emeritus at New York University, were talking about AI as a really broad concept while others were talking about it in a very narrow way.
“Kind of the core problem is that AI is this ‘know it when you see it’ concept, that’s constantly evolving. That’s really hard [to put into] a legal definition.”
To open the hearing on artificial intelligence, Democratic Sen. Richard Blumenthal played a recorded statement created entirely by OpenAI’s ChatGPT and AI voice cloning software trained on his own speeches to mimic his voice. OpenAI CEO Sam Altman also testified at the hearing, calling on the government to regulate artificial intelligence.
In Canada, the government has proposed the Artificial Intelligence and Data Act to “protect Canadians” and “ensure the development of responsible AI.” But there are no current federal proposals in the U.S.
Concerns have been raised about ChatGPT, a chatbot tool that answers questions with convincingly human-like responses, and the ability of the latest crop of generative AI tools to mislead people, spread falsehoods, violate copyright protections and upend some jobs.
Meanwhile, others, like Geoffrey Hinton, known as the “godfather of AI,” along with Yudkowsky, have expressed fears that unchecked AI could wipe out humanity.
Yudkowsky fears that an all-too powerful superhuman intelligent AI would, as he wrote in Time in March, “not do what we want, and does not care for us nor for sentient life in general.”
“The likely result of humanity facing down an opposed superhuman intelligence is a total loss.”
Yudkowsky said that, at the hearing, Altman and IBM’s Montgomery were playing coy about AI’s worst-case scenarios raised separately by himself and Hinton.
“The actual danger is that everyone on Earth dies,” he said. “Sam Altman knows that; it seems he’s decided that Congress can’t be trusted with the information. And, honestly, I’m a bit sympathetic to that decision, and so probably are most individual Congresspersons even if they’re not allowed to say that out loud.”
‘Daunting’ task ahead
Computer scientist Mark Nitzberg, the executive director of the Center for Human-Compatible Artificial Intelligence, said policymakers certainly have a “daunting” task ahead of them in terms of regulation.
“How is it that we have this system that no one understands how it works, everyone agrees that it’s very powerful and there are absolutely no regulations at all controlling any of that,” he said.
A major problem is that while AI is a highly capable system in many ways, it can be random, make things up, and, no one really understands the principles by which it operates, Nitzberg said.
“This is not the case for any other engineered system that we have rules about.”
AI is different from gene editing or climate science, where the science is worked out, and there’s still work to be done on the politics like regulations and general agreements, he said.
“You’re forced to do the politics before the science is delivered.”
Most testing and post-mortem analysis for engineered systems, for example in car and aircraft safety, depend on systems behaving predictably: performing the same way when put in the same situation, Nitzberg said.
But large language models give different responses to the same prompt twice in a row. So different kind of testing and monitoring methodologies will need to be created, he said.
Bart Selman, a computer science professor at Cornell University and director of the Intelligent Information Systems Institute, said regulations can take years, and that even if you get input from stakeholders, they don’t deal with some of the real problems.
Some critics have suggested that Altman’s call for regulations could actually be self-serving. In an interview with ABC’s Start Here podcast, Gizmodo technology reporter Thomas Germain pointed out that it’s not unusual for the tech industry to ask to be regulated.
Canadian-British artificial intelligence pioneer Geoffrey Hinton says he left Google because of recent discoveries about AI that made him realize it poses a threat to humanity. CBC chief correspondent Adrienne Arseualt talks to the ‘godfather of AI’ about the risks involved and if there’s any way to avoid them.
“Some of the biggest proponents of privacy laws are Microsoft and Google and Meta, in fact, because it gives tech companies a huge advantage if there are laws that they can comply with,” he said. “That way, if something goes wrong, they can just say, ‘Oh well, we were following the rules. It’s the government’s fault for not passing better regulation.”
At the hearing, Altman proposed the formation of a U.S. or global agency that would license the most powerful AI systems and have the authority to “take that licence away” and ensure compliance with safety standards.
Nitzberg noted that on some chat forums, some have suggested there will be “precious few who have the resources and the connections to get a license and therefore, [Altman’s] assuring his own regulatory capture.”
However, Nitzberg said he doesn’t view Altman as a cynic who is only saying that AI is dangerous in order to “get on the good side of people so that he can build a larger empire.”
“He was talking about the dangers of AI back in 2016.”
Stifle innovation
Meanwhile, there are other concerns about AI regulation: that too much government interference could stifle innovation.
“There is no reason why private sector actors can’t develop principles for safe AI practices or create their own AI governing bodies,” James Broughel, a Senior Fellow at the Competitive Enterprise Institute, wrote in Forbes last month.
“The problem with creating new federal agencies or adding new regulatory programs and staff is they inevitably create new constituencies, including of bureaucrats, academics and corporations who use government power to sway public policy toward their own interests.”
That’s why IBM’s Montgomery urged the Senate committee to adopt a “precision regulation” approach to AI to govern the deployment of AI in specific use-cases, and not regulate the technology itself.
This, she said, “strikes an appropriate balance between protecting Americans from potential harms and preserving an environment where innovation can flourish would involve.”
O’Shaughnessy agreed that too much regulation is a real concern and that policymakers need to be careful that they regulate AI intelligently.
“At the same time, though, these AI systems are very powerful. They have very real and immediate negative impacts on people and society today,” he said. “And it’s important that we put meaningful and intelligent regulation on it.”
O’Shaughnessy said it was important that the hearings revealed some bipartisan support for some kind of regulation.
“But it’s one thing for them to support that idea at a high level. It’s a very different one for them to support an actual policy once it’s more clear what the tradeoffs are, what it looks like . So it’s too early to say there will actually be momentum for regulation.”
More Stories
Fair share: the right office solution can take finding the right partner
Ontario faces crew shortages, aircraft issues in fight against wildfires | Globalnews.ca
Refugee attends open house at Downtown Eastside affordable housing facility – BC | Globalnews.ca