The past ten days or so seem like a different era for AI governance. The collective consciousness about the risks of AI and giant data centers seems to have been raised substantially. The proximate cause is concern about AI safety risks coming from staff at the frontier LLM developers themselves this time. But public concern about construction and placement of data centers and related potential harms have also grown simultaneously.
What’s going on? I think part of it is the view that a lot of people are not sure they actually want AI in their lives in the first place. Here’s what I feel I am hearing when I talk to people not directly involved in AI development. It goes something like this:
I don’t want this. It isn’t good for me or for my family. I didn’t ask for it. I don’t want to leave the decision how to handle this dangerous technology to the small number of people building it. And I don’t want my government to defer to those few very wealthy and powerful people.
I think a lot of people are wondering what AI means for them. And it is not sounding like a great deal right now.
It seems like it may wipe out jobs. It might cause harm to my environment. It may make it less likely my child will learn critical thinking skills. It could pose a risk to their mental health. And people inside these companies are saying it could wipe us all out in 10 years? Do the harms have to show up disproportionately in my neighborhood, again? Why would I sign up for this deal? How can I opt out? Can’t we just *not do this*???
I think this way of thinking about AI is rational, even though it surely misses the potential upsides from a profound new technology — the benefits of new drug discovery, ways to solve climate change, improvements to productivity and daily life, creation of new jobs and businesses, supports for economic growth, and on and on.
That has me thinking about whether there is a maxim that goes something like this:
The more powerful the potential of a new technology, the greater the worries many people will have about it coming into existence.
If that’s so, and AI is as powerful as its creators tell us, doesn’t that mean there should be a society-level discussion about it, not just a series of decisions made by a few people who are creating the tools themselves?
When even the frontier AI companies are saying that the dangers of AI are real, surely those in power to do something about it should pay attention. (BTW the same people, and their friends and competitors, said the same thing in 2023. Yet what exactly have they done since? Taken together, they have done precisely the opposite of what they said should be done. So take their 2026 words with giant grains of salt.)
The response from those in power in government in the United States who could regulate federally: no dice. Whether the President or the Speaker of the House, they have declined to act — instead saying we need continued self-regulation. Leave it to the companies, they say — the companies know best. The leaders who seem to be stepping up are the states, creating a running conflict between the states and the federal government.
There are both benefits and harms from these new emerging AI technologies. Wherever you stand on the balance between them, we all agree the stakes are high. At a minimum, we need people in public service who understand the technologies, how they can help us all, and the real harms they cause now and into the future–not people in positions of power who will simply defer to self-regulation. That is plainly out of step with what large numbers of regular citizens are asking for.
There is a parallel issue of data centers. There need to be processes for siting any of these data centers that give the people who live there a real say as to whether they want them. There need to be guardrails for data center placement just as there need to be guardrails on the development of the relate ld AI technologies.
There are many possible metaphors for the governance of AI circa 2026. Some compare it to nuclear power and nuclear weapons. Others call it a tsunami. Yet others look to drug development for useful comparisons.
What I favor these days is to think of the many related AI technologies the way we think about highways. We should learn from what has worked well and what has not worked. For instance, the highway system has been a good thing for our overall economy. It has led to new jobs and more efficient modes of getting goods and people from point A to point B. It has saved lives when people can get to urgent care more quickly than in the past. And so forth.
But too often we ran highways through communities with the least economic power, often communities of color in major cities. Too often these highways have caused environmental harms, destroying important habitat. And people can die on these highways when they drive too fast, out of control, or in unsafe vehicles.
Some of what we did well in building highways: we established guardrails. We required safety measures on cars, such as seatbelts. We set speed limits. We hired and empowered state troopers. We disallow people to drive drunk or while texting. These rules and constraints have not killed innovation in cars; to the contrary, there has been huge innovation in the development of automobiles and trucks. Often, well-crafted regulation can go hand in hand with world-class innovation.
If you don’t like the highways analogy, what about drug development? We don’t release major drugs for use by humans without required testing. We have agencies and experts in government who can review technologies and third party audits before new models are released. Many countries have established AI Safety Institutes. These technologies can be developed safely and with humans at the center. There is no reason to leave regulation to the companies themselves.
Our leaders quickly object: “But China…” It’s plain that our geopolitical standing in the US (or any nation) could be affected by AI development, especially the development of artificial general intelligence. But doesn’t it seem likely that those in other countries also would favor keeping the technologies in control, rather than having them cause terrible harms? Of all the possible governments in the world, surely the Chinese government is one that favors control through new technologies, not a loss of control.
The process of coming up with a new international agreement of some sort is surely not easy, but if the stakes are as high as we are told they are, isn’t this the time to work on it? We can reach a reasonable agreement with other countries to work with us in this regard.
We have the tools we need to establish guardrails for AI development. We need the political will to do so. And I believe that individuals, families and communities across America are increasingly thinking this is necessary for the good of all of us. My hunch is that Americans are not alone in this regard, either.
—
Further reading:
Verity Harding and co., AIxGEO: Reimagining the AI Arms Race.
Jeff Jarvis: They’re covering the wrong AI story
A piece I wrote in The Hill a while back on the limits of industry self-regulation.
Janet Haven and Rangit Singh, Data + Society, new collection on Governing AI Agents
Miranda Bogen et al., Center for Democracy and Technology, notes on Getting Third-Party AI Assessment Right
Discover more from John Palfrey
Subscribe to get the latest posts sent to your email.