# A superintelligence ban lacks a workable definition and reach

Markdown copy of https://realalpha.app/s/a-superintelligence-ban-lacks-a-workable-definition-and/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Dean W. Ball](https://realalpha.app/builders/dean-w-ball/) · Head of Strategic Futures, OpenAI · Primary author of the Trump administration's AI Action Plan, now Head of Strategic Futures at OpenAI.
- Host: [John Coogan](https://realalpha.app/people/john-coogan/) · Co-host, TBPN · Co-founded Soylent and Lucy; OpenAI bought his show TBPN in April 2026.
- Host: [Jordi Hays](https://realalpha.app/people/jordi-hays/) · Co-host, TBPN · Founded Branded Native; co-founded fintech Capital, which Rho bought.
- Show: TBPN
- Episode: [Marc Andreessen & Ilya Sutskever Host TBPN](https://realalpha.app/v/m2mWKU0srfY/), Oct 31, 2025
- Moment: 1:47:16 to 1:52:04 of the episode (5 min), playing on https://realalpha.app/s/a-superintelligence-ban-lacks-a-workable-definition-and/; the episode from there: https://realalpha.app/v/m2mWKU0srfY/?t=6436

## Quote

> “you need a **global treaty to ban** superintelligence. That's what they're asking for.”
> — [Dean W. Ball](https://realalpha.app/builders/dean-w-ball/) at [1:48:06](https://realalpha.app/v/m2mWKU0srfY/?t=6486) · TBPN

## Takeaways

1. A ban would need international cooperation.
2. There is no shared definition of superintelligence.
3. Broad policy goals can lead to poor implementation.

## About this segment

Dean Ball argues that a ban would require global coordination and agreement on what superintelligence means. Without either, a sweeping policy demand is hard to implement responsibly.

From “Marc Andreessen & Ilya Sutskever Host TBPN”, Oct 31, 2025. [Watch the full episode](https://realalpha.app/v/m2mWKU0srfY/)

## Transcript

From the transcript TBPN publishes with the episode.

[1:47:21](https://realalpha.app/v/m2mWKU0srfY/?t=6441) Yeah. So, I mean, the statement is very simple. It says no superintelligence until we have agreement that it's good until there's broad public buy in. So I think the first problem with this is, like, look. If you wanna have a letter signed by a bunch of important people or celebrities and whatnot that says superintelligence is something that has profound implications, and we should really think about it, and we should have groups of you know, civil society type of stuff.

[1:47:43](https://realalpha.app/v/m2mWKU0srfY/?t=6463) That's fine with me. There have been letters like that in the past, and I haven't complained about them. But when you say ban, you're making a ask about public policy. And not only that, you're actually making an ask about global public policy because, you know, the superintelligence could be trained in China, could be trained in The United Arab Emirates, Saudi Arabia. A lot of other places are amassing large amounts of compute.

[1:48:03](https://realalpha.app/v/m2mWKU0srfY/?t=6483) It's not like we're the only people that can do this. And a lot of the researchers, you know, who do this work are highly mobile. So, like, you need a global treaty to ban superintelligence. That's what they're asking for. And the problem is that not only is there not scientific consensus about the safety of such a thing, more importantly, there's not scientific consensus about what superintelligence is.

[1:48:24](https://realalpha.app/v/m2mWKU0srfY/?t=6504) Is if GPT five or let's say GPT seven Mhmm. Say that it, like, can make novel advancements in mathematics. It can come up with new theorems and whatnot, come up with new science. But say it also can't autonomously plan a wedding for me. I think that's a very plausible feature, by the way. I think it's super it's super possible that we'll get advanced, you know, genius mathematic AI before we get sort of an agent that can, like, plan a wedding for you over the course. Truly agentic AI.

[1:48:54](https://realalpha.app/v/m2mWKU0srfY/?t=6534) Yeah. Yeah. Yeah. And so, like, what you know, which one of those is superintelligence? Totally. No. No. And so how

[1:49:02](https://realalpha.app/v/m2mWKU0srfY/?t=6542) So take an even more extreme example. We were reflecting on the original OpenAI definition with which Ilya was, like, reflecting on, which is, AI, you should be able to tell it to do a job, and it should be able to do that any job at the level of a human. And Mhmm. We were sitting for three hours getting makeup done, and we were like, okay. To replace this job with AI, you need, like, the most flawless humanoid robots because this is not a digital task.

[1:49:29](https://realalpha.app/v/m2mWKU0srfY/?t=6569) You need dexterity, but also understanding the artistry of how the colors blend and mixing the different chemicals and stuff. Like, it's just like it's just like and and that's for, like, you know, just a it's just makeup. It's it's like it's not it's not as complicated as the IMO gold medal, and yet because it exists in this cross domain world, it's somehow more challenging. So, yeah, it's just a fascinating thing that we do. We have

[1:49:55](https://realalpha.app/v/m2mWKU0srfY/?t=6595) The other dynamic that's interesting right now is, like like, I feel like the industry, like, is somewhat going through a small period of disillusionment Mhmm. When, like, the big labs are, you know, or or OpenAI doing, like, Sora and announcing erotica and stuff like that and realizing, like, wait, like Mhmm. I thought we were building super intelligence, like, why are we creating Yeah. More forms of social media? And but that's at the same time that you have like peak sort of like fear and like peak pressure to make moves from a regulatory standpoint because people are afraid of you know, runaway super intelligence and, you know, losing losing their job and and Mhmm.

[1:50:35](https://realalpha.app/v/m2mWKU0srfY/?t=6635) All these things. And also, big tech companies have been doing layoffs and the CEOs Right. Wanna say, like, well, we did these, like, the as a CEO, you wanna say, well, we're getting so much efficiency out of our, you know, AI. Like, of course, we're doing layoffs. Mhmm. But in reality, it's like they just went through they've been bloated for a while. There's, you know, x is a good case study of, like, you can cut, you know, like, a lot of people and still have a functioning business. And so, I don't know. There's just such a complicated moment.

[1:51:06](https://realalpha.app/v/m2mWKU0srfY/?t=6666) Do you think there's a massive techlash coming? Yeah. Probably. It would be my guess. It's I don't know exactly what'll cause it. Maybe it'll be fears about labor market, which could be real, could be not could be caused by AI, could be not. Could also be electricity prices. I think that's that's very possible. Maybe electricity prices. Yeah. Yeah. Yeah. So, again, like, superintelligence ban, like, it's doing populist politics that are based on kind of vague emotional sentiment. Yeah. And that's, like the problem with that is that those kinds of broad statements that implicate policy tend to not lead to good places. Defund the police was one of them.

[1:51:42](https://realalpha.app/v/m2mWKU0srfY/?t=6702) Net zero by 2030 was another one, where it's like, we have no real plan for how to do this, but we're gonna articulate some broad goal that sounds good that everyone can kinda you know, you can just slide down the incentive curve of politics. And it's it's it will not lead to good places, I basically guarantee you.