○penAI CEO Sam Altman appealed to members of Congress under oath to “regulate artificial intelligence.”
Altman, whose ChatGPT tool is at the forefront of generative AI technology, testified before the Senate Judiciary Committee for the first time at a hearing on Tuesday. While Altman said he was ultimately optimistic that innovation would benefit people on a large scale, lawmakers created parameters for AI creators to avoid “grave harm to the world.” I repeated my earlier assertion that it should.
“We think it could be the press moment,” Altman said. “To make that happen, we have to work together.”
In addition to Altman, two other AI experts, Gary Marcus, professor of psychology and neuroscience at New York University, and Christina Montgomery, chief privacy trust officer at IBM, also testified before the panel. Joined the. The three Witnesses, both at the federal and global levels, took slightly different approaches to AI governance.
“We built a machine like the pottery bull: powerful, reckless and difficult to control,” Marcus said. To address this, he proposed a model for regulators like the Food and Drug Administration, in which creators would have to prove the safety of his AI and show why the benefits outweigh the possible harms. made it
But the senators who led the question were more skeptical about the rapidly evolving AI industry, comparing its potential impact not to the printing press but to several other technological innovations, most notably the atomic bomb.
read more: Pausing AI development is not enough.everything must be shut down
Senator Richard Blumenthal (D-Connecticut), chair of the group’s subcommittee on privacy, technology, and law, made his wariness about AI clear by responding: not a bull. “
The session lasted nearly three hours, and the senators’ questions touched on a wide range of concerns about AI, from copyright issues to military uses. Here are some key takeaways from the minutes.
Agreement on risk
The hearing wasn’t as militant as many other high-profile exchanges between lawmakers and tech executives, but largely because of advanced conversational AI such as ChatGPT, OpenAI’s chatbot. Because Witnesses acknowledged the dangers of free growth and use of such tools. Senators asked experts why OpenAI decided to release its AI to the public before fully evaluating its safety, especially how OpenAI created the current version of GPT-4. I didn’t ask some of the thorny questions I asked. .
Senator Dick Durbin (D, Illinois) said early on that he can’t remember when representatives of private sector groups petitioned for regulation.
Altman and senators alike expressed concern that AI could “go all the wrong way.”
When Senator Josh Hawley (R-Missouri) cited a study that says, for example, Large Language Models (LLMs) such as ChatGPT can accurately predict public opinion from media diets, the bad guys said that We use technology to fine-tune responses and manipulate people to change their opinion on specific topics. Altman said one of his biggest concerns was what could be called “one-on-one interactive disinformation,” and that regulation of the subject would be “very smart.”
Marcus said the job impact could be different than previous technological disruptions and that Montgomery supports regulating AI based on the riskiest uses, such as before and after elections. added.
read more: The AI arms race has begun.start to worry
When asked about his biggest fear about AI, Altman was candid about the risks of his job.
“My biggest fear is that we, this sector, our technology, our industry, are doing serious harm to the world. I think that can happen in many ways,” Altman said.. He didn’t elaborate, but critics’ warnings range from spreading misinformation and prejudice to utter destruction of living life. “If this technology doesn’t work out, I think it could go very wrong, and we want to be very vocal about it,” Altman continued. “We want to work with the government to prevent that from happening.”
Concerns over AI prompted hundreds of high-tech luminaries, including Elon Musk, to ask AI labs to suspend training superpowered systems for six months in March, citing the risks they pose to “society and humanity.” signed an open letter requesting And earlier this month, Jeffrey Hinton, dubbed the “godfather” of AI, resigned from his position at Google, saying he regretted his job and warning of the technology’s dangers.
Specific regulatory recommendations
Altman laid out a general three-pronged plan for how Congress could regulate AI creators.
First, he supported the creation of a federal agency that could grant licenses to create AI models that exceed certain thresholds of competence, and even revoke licenses if the models do not meet government-mandated safety guidelines. .
The idea was not new to MPs. At least four senators, both Democrats and Republicans, mentioned or supported the idea of creating a new oversight body during questioning.
Second, Altman said the government should create safety standards for high-performance AI models (such as prohibiting self-replication of models) and that models must pass, such as verifying the model’s ability to generate accurate information. Said that specific functional tests should be written. Prevent dangerous content from being generated.
And third, it called on lawmakers to require independent audits by experts independent of authors or government to ensure that AI tools operate within legislative guidelines.
read more: Why Microsoft’s Satya Nadella Doesn’t Think It’s Time to Quit AI
Both Marcus and Montgomery advocated demanding radical transparency from AI creators, for example, to always know when users are interacting with chatbots. Marcus then explained the idea of “nutritional labels” that describe the components or datasets the creators used to train their models. Altman specifically avoided including transparency considerations in regulatory recommendations.
European lawmakers are pushing to regulate AI applications, and the EU is deciding whether to classify general-purpose AI technology (on which tools such as ChatGPT are based) as “high-risk.” Technology would then be subject to the most stringent levels of regulation, and many big tech companies, including Google and OpenAI’s biggest investor Microsoft, argued that such a classification would stifle innovation. , has lobbied against such classification.
Avoid Similar Social Media Issues
Senators at the hearing vowed to learn from past mistakes on data privacy and misinformation on social networks like Facebook and Twitter.
“Parliament failed to capture this moment on social media,” Blumenthal said. “We have an obligation to do it with AI before threats and risks become real.”
read more: “Don’t look up” thinking that could destroy us with AI
Faced with the unknown future of AI technology, the nearly 10 lawmakers present at the hearing posed questions on a wide range of issues. Each highlighted different areas of concern regarding the impact of AI.
Senator Marsha Blackburn (R-Tn.), for example, asked about compensation for musicians and artists who train models and use their work to produce similar works in their style and voice. Senator Alex Padilla (D-California) asked about the issue of language inclusivity and bringing the same technology to people across cultures. Senator Amy Klobuchar (D-Minnesota) asks about protecting local news agencies, and Senator Lindsey Graham (R-SC) discusses how AI could impact military drones and change warfare. I asked if it was sexual. Other topics include assessing the risks of her AI industry, which is concentrated in the power of a few corporations, and ensuring the safety of children who use tools.
Altman, Marcus and Montgomery all said they were willing to continue working with the government to find answers to these questions, and Blumenthal said it was the first in a series of committee hearings. said no.
“I feel like there’s a real, real participation here,” he said.
Other must-reads from TIME
