The Future According to John Carmack

**Andrew Bosworth** (0:10)
Welcome back, everybody, to Boz To The Future. I am your host, Boz. It is great to be back for our second season, and I am thrilled today to be joined by John Carmack, Formula Oculus CTO, now heading up AGI at Keen Technologies. A man who really does not need much of an introduction. I want to say that the first programmer's name I ever knew was John Carmack when I was a young programmer, aspiring programmer growing up. I usually ask the guests to introduce themselves. What do you think is the most important thing for somebody listening to this to know about you, John?

**John Carmack** (0:46)
Well, I'm really deeply an optimist. I mean, I'm on my sort of fourth technical arc through my career after 3D video games and aerospace and virtual reality and now artificial intelligence. And I see more opportunities and more just chances for things being amazing than I ever have. I think that the present is more amazing than most people give it credit for, but the future is going to be even better.

**Andrew Bosworth** (1:11)
I couldn't agree more. And actually, you talk with these multiple careers. You gave some great kind of viral advice on Twitter this week, talking about don't focus so much on being a programmer so much as what it entails. And that is really kind of the arc for you. Like you just you, you happen to have been a phenomenal, maybe the world's greatest living programmer. But that wasn't the point. The point was the things that you were building at each stage. And you're back at it again with AGI.
Actually, I'm going to start there. The whole premise of this podcast is most podcasts kind of go very broad and not very deep. And that makes sense. You know, I kind of give the guests an easy time and you're just covering a breadth of topics and it's quite interesting. I have tried with this one to go deeper for the listeners who want to understand more. And you will talk about VR later because we got to give the people what they want. But I want to start with AI, which is where you have pivoted your considerable talents lately. Talk to me about your vision for Keen and what led you to it.

**John Carmack** (2:11)
So we're a really small company and I, you know, actually I've been trying to go out of my way to not really talk about it very much because, you know, we have funding. I've got the employees that I want. I'm not really don't have a product to sell and I don't want to be an AI pundit. So you're kind of ambushing me on this.

**Andrew Bosworth** (2:27)
Okay, we won't go too deep on this.

**John Carmack** (2:30)
Yeah, it's okay.

**Andrew Bosworth** (2:30)
We'll go to VR sooner.

**John Carmack** (2:32)
Though I'm... But yeah, it's been an interesting arc because I've been so deep in VR for the last decade basically. But things were clearly happening in the machine learning and AI space over the last several years. And really it's been quite interesting where the modern era of VR and the modern era of AI both started almost exactly a decade ago.
And it's been interesting kind of contrasting the paths that the different things have taken. But AI has clearly reached sort of a real critical mass where it is in the public consciousness, it is taking off, and there's all these practical things, all the companies that are being founded to do AI stuff. And it's almost just a Silicon Valley joke about, well, we're going to do this with machine learning and we're going to add this to it. But when you get down to it and you look back at sort of the primordial desire of artificial intelligence all the way back to like the late 50s meetings where the term was coined, and this idea that unless you're some kind of a dualist, thinking that there's a separate spirit or a soul from the body, we are biological machines, anything that our brains do, we should be able to simulate in software and hardware. And if you can do that and you can wind up doing the things that make humanity great and wonderful and then possibly run them with greater breadth or depth or speed or any of these other things, then it is transformative in a way that almost beyond almost any other technology that you can imagine. And when I started looking at this seriously a few years ago, I did really get this sense that we are on the cusp of this. And of course, people have thought that since the 50s. You know, the joke was back in that original Dartmouth meeting that they thought it's like, wow, we could make great progress in a couple summers with a few grad students. And it's been a running joke that, you know, it fails deliver over and over. But there's been an awful lot of goalposts moving there. And, you know, at the risk of being a butt of a future joke, it does feel like this time it's different, that we really are at a point where the big, big breakthrough is close. It's insight. We don't really we don't have a direct line of sight. Nobody knows what to do about it. But it feels like we're about there on the hardware capabilities for it, which is one of the grand lessons of the last decade. It didn't matter how smart you were. You weren't going to build an AGI in 1990 It just wasn't going to be possible. But it feels like we are at the point where it's possible. Today, nobody knows exactly what needs to be done. But it really does feel like it's this small handful of insights away from where we are now to this human level general intelligence. That's something that people that aren't really fairly deep in the trenches don't realize where the only difference between a decade ago when AI was kind of a joke and where it is today where people are afraid it's going to take over the world and be some extinction event, it is this handful of really simple insights in the technical levels of the way these are implemented, insights about architectures, activation functions, training procedures, and each of these is simple enough that you can write them on the back of an envelope. They're not like super deep. I mean, yes, you do read papers that have a hundred pages of really dense mathematics, but then you summarize it all down into the two sentences that actually matter. And so it is my feeling that right now, this coming decade is potentially the highest leverage point for sort of a single individual that comes up with some of these ideas potentially ever in history. And, you know, there's no guarantee of this. And the smart money is that anybody setting out to do this won't come up with those particular ideas. But the chance is there, and I don't think the chance is insignificant. And so because I'm in this position where I can afford to kind of throw the next decade of my life at this, this is kind of the play I made. Now, as you know, I was right on the edge there where it's like I could go lead VR for a decade to try and do what I thought would be a very high probability outcome of like I'm damn near positive that I could make huge differences in that in that area. While I'm only barely credibly considering the ability that I could make one of these differences here, but they have different magnitudes in the payoffs. So it's been exciting for me.