This is a technology that exceeds human cognition.
You're working with Anthropic, you're working with OpenAI, you're working with Google, you're working with Microsoft. What don't they get about what's going on?
Yeah. I'd tell you other than Elon, don't think you're going to stop AI by just protesting a data center.
You know, if it really does come down at the end of the day to a Donald Trump and Xi Jinping, could you call them together?
Van Jones on Substack. It's a huge honor to be with you. I read the essay. I feel much more encouraged, challenged, but also much more encouraged having read it, for you to be coming off the sidelines and pulling people together to fix this important—I want to set this up properly. You know, for you to come in and take this position is so important because you've been a titan on the tech side. You started Microsoft. It's one of the, you know, probably most important companies in the history of capitalism. Then you started the Gates Foundation, which is, you know, the signature charity probably in the world. So you kind of have a double credential here: powerful on technology, powerful on philanthropy. This essay puts you right there at the crossroads of technology and public purpose. You're calling people together to say, "Look, we've got to get AI right." Why did you decide to do it now? You got the things you're scared about, the things you're excited about, things you want us to do. But just set me up. What is a concern that has Bill Gates raising his hand today saying, "Guys, look at this"?
Yeah. There's two dramatic things during the last year that compelled me to write an almost shrill essay about, wow, we are not paying enough attention to minimizing the bad or accelerating the good. The first is that the models have gotten way more powerful even faster than I expected. You know, I was stunned when I first saw ChatGPT a little over three years ago. And now at the end of last year, I saw a lot of the models, but Claude Code first, that was actually writing programs and taking the one thing I'm good at and being almost superhuman at that, and then the power through the agentic approach of designing molecules or finding flaws in software, or being so human-like that you almost feel like you have a relationship. So that's the first thing, is that the models, anybody who hasn't used the latest, they are amazing. And I know from talking to the industry that even this year we'll see big improvements.
Second is that I was kind of stunned that after many years of saying when we got to a threshold that the models were good and were a threat for cyber attacks or bioterrorism, that then we'd exercise some discretion at least against those particular capabilities. And I've been kind of stunned this year that there hasn't been, either within the industry or outside the industry, more of a clarion call that, wait a minute, this is not like any other technology. And in fact, some people saying, well, technology's always been good, don't get upset about this one.
Yeah. Let's just talk about that before we get into some of the concerns that you're raising. For you to be saying, hey, this disruption is different, you got to take that very seriously because you were the most disruptive kid on Earth when you came on. I mean, you came in and you turned the tables on everything and created, you know, you were one of the big people—the entire internet boom, most of what we associate with the last internet and technology revolution, you led. And people were terrified of you and said you were going to disrupt too much, but it all worked out pretty well. What's different? What's different about this AI boom such that the wonder kid from the last boom is saying slow down? What's different?
Yeah. So, as you say, I've always thought innovation was a net good even if it had some drawbacks. You know, when I did word processors, people'd then say, "But you know, you could write a nasty note or something." It was overall very empowering for people. What's different here is that this is a technology that exceeds human cognition: its ability to outthink us and to understand what we want and engage in things that we don't expect. Just the level of power, you know—so the job market will be deeply affected, and somebody who couldn't have designed a cyber attack will be allowed at very low cost to do that. So we've reached this absolute threshold of power that makes this, in the history of mankind, a uniquely dangerous and also possibly, you know, positive tool for us to use.
One of the things you said in the essay that got my attention was that all of the good things that could come out of this technology, all the medical breakthroughs, they go right hand in hand at the same time with all the scary stuff. If something is powerful enough to cure disease, it's also powerful enough to create some disease. And therefore, if you want the good, you've got to take the bad seriously. And right now you're saying we just don't have any construct in the private sector, the public sector, the philanthropic sector to deal with something like this. You said something about children in this essay I think people should really look at. You said there's a danger these chatbots create a—what do you call it—a greenhouse bubble in which children don't actually develop very well. Can you talk to the parents here? We're going to get to all the positive stuff in a second, but let's get the threats right. But on the threat side, what is this greenhouse effect you're concerned about for kids?
Well, even before AI is getting so powerful, the idea that kids are a little less mature, a little less going out and engaging with each other—Jonathan Haidt, Scott Galloway, a number of very powerful observers have seen this trend. And so I would say we should be very concerned that unless AI is used tastefully, that it'll take that to a whole new level. I mean, you could literally have a kid who has four or five friends, all of which are AI, and they'll never experience, okay, my friend's in a bad mood today, or my friend misunderstood me, or my friend needs help, because it'll be so accommodating that the normal adjustment we go through of developing socially, that will be completely ignored. And so I even worry about these psychosocial dependency things for adults. But I think the black-and-white part of this is what is this going to do to kids, and how do you engage parents, how do you put limits on? And that's one area where China's actually been the strongest at saying, "Hey, we're just not going to have kids have access to this."
You know, one thing you said in the essay that I thought was touching was that kids getting the skills to adjust is tough. And you know, as a kid, I wasn't that socially skilled. You said you weren't that socially skilled. But you said your mom actually played a big role in helping you. What did your mom do that helped you, because you have a bunch of parents listening now. We got their attention. What did your mom do for you that, had you been stuck with a chatbot, would have never happened?
Well, both socializing with people my own age—being willing to take me, invite people over, making it fun for kids to come over. If I did well with the kid, she'd kind of praise that. If I sort of disappeared, she'd say that's not appropriate. Also with adults—when they'd have parties or any adults coming over, I remember thinking, "Oh no, I'm going to have to talk to these adults, and they're going to say, 'Oh, you've grown so much' or whatever." But over time, actually, that part of my social skill developed faster than talking with my peers. And it ended up being very helpful to me where people would choose to mentor me because I could talk to them about the crazy ideas I had. And so now I give my mom immense credit for taking an introverted kid who could have just sat at the computer and giving me enough social skills that when I needed to hire people and promote Microsoft products, that I had some capability.
My mom was the same with me. I was a shy kid. I wrote a lot of poems and drew a lot of cartoons, and I would carry my books to the sports games. But she worked on me, and I just think it's important—parenting gets more important in the AI age, not less. But speaking of the AI age and the AI race, look, you know, you're in such a unique position because from the foundation point of view, which is the philanthropic side, you're working with all the hyperscalers. You're working with Anthropic, you're working with OpenAI, you're working with Google, you're working with Microsoft. So here you are on the foundation side trying to make the world better, and you are in direct partnership with the people who are leading this revolution. What don't they get? Talk about your relationship with the existing AI leadership.
Yeah, I'd say other than Elon, I have a great relationship with all the AI leaders: people like Demis, or Sundar who's the CEO, Mustafa, Microsoft, Satya, Sam Altman, really helped me understand their roadmap. Dario Amodei and their team—if you look at the statements they've made, they share a lot of these concerns. Maybe one is a little more black and white than others, but even when OpenAI was creating its product, they had various rules like, well, if we get to certain dangerous thresholds, maybe we'll have to hold back. Unfortunately, because we have many of these people competing with each other both in the United States and primarily in China, none of them individually can hold back. And so they're all kind of waiting for people outside the industry to come in and say, "Wait a minute, a model that can make new molecules needs to be monitored to make sure it's not a tool for bioterrorism."
And so I think actually as we make this concrete, I don't think we'll see a lot of resistance. They've said they want to engage government, but here we are. And I'd say it's the sophistication on the other side where I would say the awareness of these downsides is not as great as it needs to be in either political party. I'm not saying one party's got it wrong; it's both. As you see graduates who use AI, they know how powerful it is, and they're concerned about the disproportionate effect of the job market changes on them. So you have them being upset, you have people attacking data centers. But stopping every data center would not lessen the challenges we face. We need solutions very specific to the job market or bioterrorism that I don't think the industry will resist. If we said completely okay, they would, but I think there's a solution that...
Yeah, and what you're suggesting is a level of coordination between the public sector and the private sector, from the grassroots all the way to global, that really hasn't been achieved yet by humanity. One thing you said in the essay, and this is not just like the nuclear arms race where you're only regulating one technology in the weapons space—AI is going to hit literally every single part of the economy. And so this is kind of why you're suggesting and really calling for people to come together to get this thing solved. But it does seem to me that there's this sort of trap that the AI companies are in, and that they have to keep racing ahead of each other. America's got to beat China, Anthropic's got to beat OpenAI, and once you're in this kind of competitive race-forward dynamic, isn't it very hard to tap the brakes? I mean, is that why you're saying that governments have to come together to get this done?
Yes. Expecting an industry to self-regulate, including issues like psychosocial effects that—they're not social scientists, psychologists. We need to bring in all of academia: economists, social scientists to weigh in on these things. And some of them aren't using AI, some of them have heard, "Oh, they're always going to make lots of mistakes." And so my memo is sort of a shrill call that, hey, anyone who says this is like past technologies, they are wrong, and we will be sorry if we don't bring all these groups in. And of course, in the end, it's the government who's going to set the rules and say, "Let's minimize cyber attacks with these guardrails." I don't really understand when people say they're going to do voluntary reviews. It means nothing unless you set the criteria of what you're looking for, and you say what you would do if those criteria aren't met.
Yeah. One of the things I've been seeing is that when I'm with the AI leaders, they say we need the government to regulate, but when I'm with the government leaders, they're like, we don't know how to regulate, this is all new stuff.
