The Best AI Regulatory Recommendations From Sam Altman’s Congressional Testimony
The time to regulate artificial intelligence is now
“Congress failed to meet the moment with social media”, as Senate Judiciary Chair Richard Blumenthal noted in his opening statement. It cannot risk doing the same with artificial intelligence (AI). The dark dangers of AI could be far more severe than what we’ve witnessed with social media.
With that said Congress should not act simply for the sake of acting. It must be thoughtful, collaborative with the AI industry, and supportive of AI innovation. If the government takes the approach it has with crypto, chaos could ensue.
This is why it was reassuring to see AI industry leaders like OpenAI’s Sam Altman and IBM’s Christina Montgomery voluntarily testifying before a bipartisan group of Senators on the Senate Judiciary Committee. They educated, issued warnings, and presented ideas alongside an AI academic and entrepreneur, Professor Gary Marcus. The discussions were insightful and all of the Senators appeared to agree (shocking!) that something must be done to regulate AI before it’s too late.
Not all of the recommendations were great, however. Although granted, nobody knows exactly how AI innovation will develop, let alone the unforeseen risks that may lurk in the shadows.
One thing is certain though: we must learn the lessons from social media and develop a regulatory regime now before too much of the AI toothpaste explodes from its tube. Once some of these more advanced systems are in the wild, they will be far harder to control.
The following highlights what I thought were the best AI regulatory recommendations from the congressional testimony. And don’t worry. I’ll discuss some of the worst recommendations too. Let me know if you disagree.
No Section 230 style immunity
Section 230 of the 1996 Communications Decency Act includes the following:
“No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.”
This one sentence has shielded social media companies, among others, from liability for decades. Accordingly, they’ve had little incentive to combat misinformation and other harmful content absent the public pressure over the past few years.
We cannot make this same mistake with artificial intelligence. Two important points to highlight: (i) Section 230 arguably does not apply to tools like ChatGPT today because the interactions are limited to a user and the AI tool; and (ii) use of AI does not absolve you from what would otherwise be legally problematic (e.g., using an AI tool in the hiring process that results in discriminatory hiring practices).
AI companies must have legal skin in the game. We obviously do not want to open the barn doors for every plaintiff lawyer to ride in on a herd of frivolous claims, but there must be accountability on the companies that operate in this arena. We must be cognizant of striking the right balance.
No AI moratorium
Some Senators seemed in favor of a recent initiative that included support from the likes of Elon Musk and one of the witnesses at the hearing, Professor Gary Marcus, calling for a moratorium on certain AI advancements. The idea being we need to pause the deployment of certain high-risk AI tools until we have a better idea of their capabilities.
The world, unfortunately, will not wait. It’s not practical to expect the industry to pause while China, the European Union, and other global competitors advance the technology. If America wants to be an industry leader, it cannot afford to promote moratoriums. Instead, it should work to collaborate with the industry and design initial rules of the road that make sense.
We need a new government agency dedicated to artificial intelligence
One of the best solutions that came from the hearing was the proposal to create a new government agency dedicated to AI. Given the vastness of the technology, which will likely touch all of our lives, both personally and professionally, we need a central clearinghouse and enforcement body that specializes in all things AI. We have focused public attention now. It’s time to seize the moment.
IBM’s Christina Montgomery advocated for a precision regulation approach — as opposed to regulating AI technology itself — but I don’t think that’s workable for a number of reasons. Namely, there are simply too many possibilities with AI tools. Precision regulation would focus on the uses of AI. But the number of uses is unfathomable and is growing exponentially every day.
Instead, a new AI government agency should focus on licensing the technology after reviewing its safety, much as government agencies would give a driver’s license for a car. Only this AI “car” would need not only a license, but a thorough FDA-style safety inspection before being deployed in the wild. This safety inspection would need to assess potential uses prior to deployment, and then it would need to monitor how the AI is being used post-deployment.
A new AI government agency with licensing and enforcement authority is crucial to reasonably ensuring safety and soundness. But it is nothing without clear rules of the road.
The AI rules of the road should initially be principles-based
We currently do not know enough about generative AI to regulate it with specific rules. The technology is rapidly evolving, as are the ways people are choosing to use it. A principles-based approach therefore would be ideal as it would give more flexibility as the industry continues to innovate.
Some of the principles-based rules discussed at the hearing, which Sam Altman and others seemed to agree with, were around transparency, accountability, and restricting use. There must be some level of disclosure in terms of what models generative AI is using (i.e., the data it runs on) and people should know when they are interacting with AI. I already mentioned accountability by making it clear — potentially with a federal private right of action — that AI companies could be sued for the actions of their AI. And finally, there must be informed consent to train AI on someone’s private or personal information. Otherwise, AI should be restricted from using that data.
Until we have a clearer picture of the major and specific risks we need to regulate, a principles-based approach is ideal. It sets guardrails while giving the industry the opportunity to continue innovating. It also focuses on the bigger picture of each type of AI as opposed to the many uses people can creatively conjure.
Jobs are not our biggest risk with AI — AI nuclear warfare is
I don’t mean nuclear warfare in the literal sense. Although that’s certainly a possibility in the future should AI figure out how to access and actually engage in nuclear warfare. For now, I’m more focused on the major ramifications AI could have on the hearts and minds of everyone it reaches.
Most of the Senators at this hearing were focused on job loss from AI. While that is a legitimate concern, I’m still bullish on the opportunities AI will create when compared to the opportunities it makes obsolete. Potential job loss, however, pales in comparison to the shape-shifting effect AI could have on human civilization.
Risks abound from unduly influencing democratic elections, creating novel biological agents, and providing mountains of misinformation in everything from medical and psychiatric information to politics.
By first making it clear that AI companies have liability for what they create, that government is here to collaborate and partner in the development of new AI technology, and that a new government agency will help apply principles-based rules sets, some of these bigger risks can be mitigated from the start. But what this hearing with Sam Altman and others made clear was that we desperately need to start. Congress cannot sit on the sidelines like it did while social media moved fast and broke almost everything.
As Sam Altman said, and I’m paraphrasing, artificial intelligence is like social media on steroids. We shouldn’t wait to find out what that really means.
