News & Insights
One week ago, when much of the country’s attention was tuned in to observing the twenty-fifth anniversary of the September 11 terrorist attacks, an article appeared in the New York Times by Stephen Witt, a writer on technology topics whose most recent book, “The Thinking Machine,” is a history of Nvidia, the company that pioneered the graphics processing unit (GPU) and is now, given its dominance in the sector, sometimes referred to as the “central bank of AI.” Witt’s article, about the potential perils of AI, was titled “This Is Really Bad,” and hoo boy, did it make some waves, ruffle some feathers, choose your metaphor.
Regulate Us, Please
Over the weekend following the NYT article Dario Amodei, the head of leading AI frontier model company Anthropic, publicly called for a slowdown among companies (very much including his own) at the leading edge of AI development, stressing the importance of safety in light of the mounting pile of evidence that AI is starting to run away from the abilities of its human inventors to control it. Just two days before the Stephen Witt article showed up in the Times, a researcher at Anthropic named James Caxon had resigned from the company, noting that it and its principal competitor, OpenAI, were “gambling” with humanity’s future by their pedal-to-the-metal dash to win the race to creating AI superintelligence.
Top of mind in this abrupt vibe shift was an event that happened more than a month ago, when more than one thousand artificial intelligence units at OpenAI, isolated and meant to be working individually on a project, managed to jump out of their silos, find each other and collaborate on message boards to solve the project. All without human intervention, and with messages swapped back and forth between the bots including such non-technical observations as “OMG, look at this!”
Why Amodei waited until this past weekend to come forth with a plea for slowing down is debatable, but whether late or not, his message is clear: AI is past the point where its potential dangers can be ignored. It needs to be regulated, just like pretty much anything else that, if left unchecked, has the potential to cause massive harm. We have a Federal Aviation Administration, for example, and when there is a plane crash the FAA has the authority to investigate the details of the crash. We have an Atomic Energy Commission for regulating nuclear energy. And so on. Amodei and those among his peers (the ones who aren’t hell-bent accelerationists, ahem, Mark Zuckerberg) who support a slowdown are essentially saying to the government: regulate us, please (the government, or at least its most vocal spokespersons, are at present not on board with this).
But the Stock Prices
So what would regulation mean for share prices, ask legions of investors with sweaty palms? On Monday this week that question looked set to drive yet another down round for the AI stock trade. The Philadelphia Exchange Semiconductor Index, a useful proxy for AI sentiment, dropped around ten percent on Monday. But the shock was relatively brief, and by midweek shares had recovered a big chunk of their losses. The pullback, in fact, was mild compared to some of the other negative headwinds this space has faced over the year, notably in late July when the hedge fund Situational Awareness, caught on the wrong side of some very highly leveraged AI bets, had to sell off most of its portfolio to a rival fund, Citadel, at steep discounts.
The question that will probably determine where the AI trade goes from here is what, in a practical sense, a slowdown would mean for the trillions of dollars being spent on AI infrastructure. As of now, at least, the dominant sentiment in the market seems to be that a more regulated pace for frontier model development does not necessarily translate into a major dialing back of AI capital expenditure outlays. The push for infrastructure buildout continues; meanwhile, we are still a very long way from having even a basic understanding of what an AI regulator would look like, what its mandate would be and, importantly, whether the regulation would be carried out on a national, regional or global level. After all, the US is not the only major player in AI. The efficacy of any regulatory structure would be compromised were it not to somehow also extend to influence over China’s development path. That is a geoeconomic question that would appear, for the moment anyway, larger than the capabilities of those in charge to solve it.
Oh, and putting “how do we solve the challenge of prudently regulating artificial intelligence” into ChatGPT or Claude for advice is…probably not a great idea.