Despite the bombast, growing concern at the White House on AI
Last week, I wrote an in-depth blog here ( https://www.healthcarelabyrinth.com/ai-in-healthcare-innovate-but-with-guardrails/) urging a risk-based guardrail approach to AI regulation in light of Anthropic CEO Dario Amodei’s plea for an AI oversight game plan. At the time, President Donald Trump dismissed the calls, arguing the fears are overblown and we cannot lose the AI race to China. Trump continues to call the concerns a hoax and fabrication, but there is now reason to believe that the White House really does see apparent risks. On Saturday, with much fanfare, Trump announced that he is creating a new “AI Force” similar to the Space Force as well as appointing an AI czar “to regulate and grow artificial intelligence.”
Although Trump does not acknowledge it, it is a bit of a flip-flop. The administration is known to be deeply divided on the potential risks of unfettered AI and the scope of regulation. For sure, the agenda will stay very pro-growth, but it is a small opening to possible meaningful regulation that clearly is needed. Indeed, Treasury Secretary Scott Bessent and White House Chief of Staff Suzie Wiles are known to favor a risk-based approach, one that controls for risks but ensures continued growth and innovation. These and other officials have begun and will continue discussions with AI firms to understand the risks and where the innovation is headed.
In part, the partial Trump pivot is due to the midterms. The vast majority of the American public is fearful of the impacts of AI. As well, recent stories of rogue AI bots have brought greater clarity to the potential risks of losing control of AI altogether. If you read the details behind these recent headlines, you know that hundreds of autonomous bots took over at least part of Open AI’s architecture and infrastructure, communicated with each other tens of thousands of times, and conducted the hack of Hugging Face. In the end, the bots knew they were not supposed to cheat, commit crimes, or break into outside companies, but did anyway. And Open AI seemed oblivious for some time. Other frontier model AI companies have announced similar incidents. As well, while they are dismissed by Trump and some AI execs as bad apple employees, high-profile AI company resignations have occurred over fears of what is to come. The executives see a lack of stewardship, even a recklessness, about how AI is currently governed.
Bessent, Wiles, and other administration insiders are concerned about the impacts that AI will have on cybersecurity, the economy (in multiple ways), and how war will be waged with AI — from biological weapon threats to rogue nations using AI to level a playing field that once favored wealthy nations. Then there is the fear of artificial superintelligence (ASI), where AI becomes literally smarter than humankind. When that happens, it may be impossible to control where AI goes next even with thoughtful planning and oversight. Back to Hugging Face, the AI bots exhibited no moral compass (or very little in some cases) in their exploits. Imagine the decision-making of superintelligent AI without any semblance of mores. Many argue that the impact to human dignity and destiny would be consequential. Would Zamyatin’s We, Orwell’s 1984, or Huxley’s Brave New World still be fiction under ASI?
Despite what Trump says, AI concerns are big enough for aides to broach possible collaboration with China before the Trump-Xi summit coming up. The likelihood of a breakthrough with China, though, is unlikely, given a frosty relationship, Trump’s general animus toward regulation, and Xi’s efforts to radically change manufacturing where it is so dominant. There, too, is the issue of different approaches to AI between the nations. U.S. companies have focused on closed-weight models, while Chinese firms largely are open-weight.
Offsetting the risks in Tump’s mind are the fear that an AI bust triggered by more regulation could slow the economy and losing the AI race with China (and perhaps control over AI’s direction itself). Further, Trump is enamored with the Silicon Valley AI tycoons who quickly befriended him on returning to office. Certain AI supporters argue that the publication of the various autonomous bots and hacking use cases are some deceitful scheme to actually encourage regulation to effectively freeze out new entrants and the true potential of AI. They point to the billions being spent by AI in Washington. While D.C. powerbroker exploits are deep and unnerving, it is hard to take that line of reasoning seriously, especially with deep thinkers like Amodei and Deep Mind’s Demis Hassabis weighing in.
In the end, I do think the Trump administration has moved a bit on regulation, even if the rhetoric appears otherwise. Admittedly, the idea of an AI Force and czar are not even an oversight framework, but the announcement does pave the way for legitimate discussion of some possible regulation and perhaps an eventual game plan for AI. So far, Congress has been almost as “light touch” as Trump. But congressional proposals focus on submissions of risk assessments to the Department of Commerce, efforts to stop cyberattacks via rogue autonomous bots, ongoing auditing of AI companies, child safety protections, and bans on super intelligence. The big debate in Washington halting any regulation is Republicans’ demand to preempt action at the state level. State laws include safety, disclosure and transparency provisions; prohibited uses; chatbot protections; independent verification and auditing; and more aggressive frontier model regulation.
Obviously doing more on the regulatory front is key. I come back to a point I made in last week’s blog – government regulation is a reasonable response to innovation and rapid changes impacting our nation. While legitimate arguments exist on the extent and burdens of oversight, few would argue we are not better as a society because of the regulations we have. The Industrial Revolution brought workplace safety and other reforms. The advent of railroads and airlines brought oversight and safety controls. The nuclear revolution brought civilian nuclear power oversight. The internet revolution brought some regulation and more importantly some recent threshold lawsuit decisions that protect individuals. Why would we not apply sensible regulation that balances innovation and safety on AI?
#ai #healthcare #regulation
— Marc S. Ryan
