Ben Buchanan served as the White House Special Advisor for Artificial Intelligence under President Biden and is now a professor at Johns Hopkins University and an adviser to Anthropic. He is the coauthor, alongside Tantum Collins, of The Bitter Struggle: Superintelligence, Superpowers, and the Fate of the World, forthcoming on November 10.
We spoke the morning after the midterms about what voters are signaling on AI, the administration’s approach to export controls, and the leadership he hopes to see from both parties heading into 2028.
This interview was transcribed and edited with the aid of AI tools.
How are you interpreting the election results with respect to AI, and where do you see room for agreement?
I think what happened here was voters saw a technological reality that they did not like: the pace of AI progress, the pace of the data center buildout, the lack of safeguards and controls on those things. And amongst many other things this election was about voters saying, “No, we’re not comfortable with this.”
So I hope what comes from this is the new Congress and the president coming together and saying, “Okay, can we hear the voters on this, and can we try to meet the voters where they are with what are very real concerns?” What that looks like is hard to say, but I do think this is a signal on behalf of the American electorate that politicians should pay very close attention to.
Having a nation that can build things and make things work is really important for competitiveness in AI. And if you look at the inputs to AI, amongst them electricity, then this is really a key part of making sure the United States is competitive. At a minimum, we need to build more power to keep prices down for consumers as demand grows. And our view in the Biden Administration was that there is a way to do that that is entirely consistent with the wishes of communities, either by involving them in the process or by building in places where people are not. We want an America that builds things with care, and I expect that this is something on which the left and right can agree.
What grade would you give this administration’s approach to export controls—and how much reflects the policy itself versus its enforcement?
I teach a lot of Gen Z, and there’s a lot of Gen Z grade inflation. But even grading on a curve, I would give them a C minus.
It’s hard for me to disentangle from the outside what is a policy decision to explicitly revoke versus what is a policy decision to not enforce, but their actions have been bad on both fronts. The decision to let the Chinese continue to have remote access to computing power doesn’t make any sense. The decision to sell H200 chips to the Chinese doesn’t make any sense. The decision to not crack down on DUV immersion tool sales to the Chinese doesn’t make any sense.
The president is fond of saying “whoever wins AI, wins.” But if you proceed from a notion that AI is going to matter a lot for national security, and if you proceed from a notion that computing power is really important for AI, I don’t understand why the administration would continue to let such computing capabilities get to the Chinese.
When I worked in the White House, we were looking at things like the scaling law and the jump in capabilities of GPT-3 and saying, “Wow, this really shows the preeminence of computing power as a driving force for AI.” What are the policy implications for that? What does this mean for geopolitical competition? It has given America a lot of leverage that we would not have had in a world in which data was the primary accelerant of AI progress. That was the origins of the export controls, and the logic still holds today.
The Information recently reported on Nvidia’s use of “soft power” to shape export-control policy. When you were in the White House, how did you distinguish the company’s technical advice from its commercial interests?
In 2022, when the export controls were starting to roll out, Nvidia was making the argument that this would be very bad for their business. This action would hurt their ability to run faster than other competitors and would ultimately hurt the United States.
And our response to that was, “No, we have total confidence in your business. And we have such confidence that the scaling laws are going to mean that AI chips, of which there’s essentially a fixed supply, will be in such extraordinary demand that things are going to go great for the AI chip industry. And in a world of a fixed supply of chips, we want it to be American buyers and not Chinese buyers of those chips.”
And I would argue that’s what happened. Nvidia at the time of the chip controls was worth probably something like $350 billion. And now they’re, depending on where the market is on any given day, worth 15 or 20 times that. So it’s been great for the United States, and it also has been great for Nvidia.
You’ve expressed admiration for Admiral Hyman Rickover’s ability to take a transformative technology and make it operationally useful. What kind of leadership do you want to see on AI in the run-up to 2028?
I do think this is going to be a moment, especially as we transition to a new administration, where we’ll need a lot of execution-oriented people inside government. And probably we will need a candidate, I hope for one on both sides of the aisle, who can say, “Well, AI really matters here for American competitiveness. It matters for government capacity. It matters for American national security and safety. We’ve got to get this right on a range of fronts. And here’s my plan to do that. Here’s my conception of this issue.”
How is the United States going to lead the world in inventing this technology? I think it goes better for democracies and humanity if this is an American technology. Again, I think that’s something that left and right can agree on.
How are we going to use this to improve government services and government capacity, including for national security? A lot of the government tech is really out of date.
How are we going to regulate these companies and make sure that citizens and government have some mechanism to control what the companies do and prevent a race to the bottom? I believe some kind of action addressing this is essential.
And then how will we set up multilateral institutions to make sure that other nations follow American leadership, in the same way that presidents of both parties during the Cold War set up multilateral institutions to lead the free world? I think we’re going to have to do that again in AI.
In some sense that’s Rickover plus-plus, because it’s touching on the political side, not just military execution. But given how powerful the technology is and where it’s going, I suspect we’re going to need that in 2028. And I also think it’s a big political opportunity for either party to present their theory of the AI case to the American people.
What do you hope readers take away from The Bitter Struggle ?
One of the things we try to do in the book is just look at what is possible and articulate it in a clear, scientifically rigorous and grounded manner so we can all start thinking: What should we do about this?
We should expect that as AI systems get more powerful, which they will, the risks will only grow. And one of the reasons we wrote this book, one of the reasons we do this work, is to suggest that policymakers need to get ahead of that—and they’re quickly running out of time.
I hope people who read this book will recognize that we are bringing out the technological facts that are driving a lot of where the world is and where the world is going to be. We’re presenting them in a fashion that’s accessible to people who might not be technologists at their core, who might be approaching this from a national security lens, and saying: Okay, what are the things that we can all do here in a bipartisan way to make the most of this for the United States and for democracies? To our mind, that’s one of the most important questions in the world right now.