Elon Musk Predicts AI Will Surpass Human Intelligence in Five Years and Calls for Industry Cooperation

[00:00] I think AI may exceed the sum of human

[00:03] intelligence

[00:05] in about in around five years.

[00:07] >> In five years,

[00:08] >> roughly five years is my guess. Um,

[00:12] so there really won't be anything that

[00:15] AI

[00:16] can't do better than humans apart from

[00:19] being human perhaps.

[00:20] >> In a more prosaic level, what will life

[00:22] be like?

[00:25] The most likely outcome

[00:28] is an age of amazing abundance uh where

[00:34] anyone can have anything they can think

[00:36] of.

[00:37] This is this is this may sound

[00:39] preposterous but well here we are in

[00:41] 2026.

[00:43] Let's see where we stand in 2036. Now of

[00:45] course there are things that could

[00:47] derail this like a a massive global

[00:50] thermonuclear war or something like

[00:51] that. Um, but assuming that there's not

[00:54] World War II,

[00:56] uh, I think I think we're headed for an

[00:59] age of amazing abundance. So, this is, I

[01:01] guess, a message of of optimism

[01:05] and, uh, excitement about the future.

[01:07] >> You're sure that humans will no longer

[01:09] be in control in 10 years time?

[01:12] >> I think it is unlikely. Um, in the same

[01:14] way that if you had a vastly more like,

[01:20] you know,

[01:22] the

[01:24] like like if if if the difference in

[01:26] intelligence between AI and humans is

[01:30] vastly greater than the difference in

[01:31] intelligence between AI and chimpanzees,

[01:34] it's hard to imagine that the

[01:36] chimpanzees would be in charge.

[01:38] >> I'm really sorry. We are essentially um

[01:42] an evolved form of chimpanzee. So it

[01:45] wasn't that long ago that we're swinging

[01:47] through the trees eating bananas um in

[01:51] the jungle and we still eat bananas.

[01:53] >> You said there was a 10 to 20% chance of

[01:55] killer robots wiping out humanity.

[01:57] >> So you sounded worried then. Now you

[02:00] seem very relaxed about it. Have you

[02:02] changed your view?

[02:03] >> I still think there's risk associated

[02:06] with AI and robots. It's not zero. Um,

[02:09] I' I guess I've come to

[02:12] my my sort of philosophical conclusion

[02:15] is to look on the bright side. Um, I

[02:19] like I I can't see any way to really

[02:21] stop this incredible momentum of AI and

[02:25] robots. Um, and and at times I sort of

[02:29] think, well, perhaps, you know, even if

[02:31] there was a stop button, we probably

[02:33] shouldn't press it. uh because

[02:38] because the most likely outcome is

[02:41] incredible abundance for for all.

[02:45] >> That's what I thought actually. It

[02:46] struck me that now you seem to have

[02:48] decided it's inevitable.

[02:51] Let's hope for the best and go for the

[02:53] ride.

[02:54] >> Yeah. I mean, yes, pretty much. Let's uh

[02:59] let's enjoy the ride is is is I think my

[03:02] philosophy at this point because I mean

[03:03] honestly this there's it's it's there

[03:05] there seems to be just an inexurable

[03:08] uh progress here in in AI and and in

[03:12] robots

[03:13] >> that um

[03:15] >> but you even if I wanted to stop it I

[03:18] don't I couldn't um

[03:20] >> do you think there's still a 10 to 20%

[03:21] chance that killer robots mean the end

[03:24] of humanity?

[03:27] That's a high P doom as they say.

[03:30] >> Well,

[03:32] in the long run, we're all dead of

[03:33] course. Um, you know, um, so I mean, so

[03:36] I mean like according to the standard

[03:40] model of physics, the end of the

[03:42] universe is um a long boring sort of

[03:47] thin gray soup of uh gradual cooling. um

[03:54] meaning the sort of the heat death of

[03:56] the universe it's called. So now if if

[03:58] the heat death of the universe is in

[04:00] fact the outcome for our reality, then

[04:03] it really is all about the journey

[04:05] because the destination is terrible.

[04:07] >> But 10 years [laughter] is not very

[04:08] Yeah, I get that. But that could be a

[04:10] long way and you're basically very long

[04:12] way.

[04:12] >> If it's 10 to 20% in 10 years, you know,

[04:15] would you go into one of your rockets if

[04:16] you thought there was a 10 to 20% chance

[04:18] of it blowing up and killing you?

[04:21] >> Yes. But [laughter]

[04:23] but let's say there's you can't do

[04:25] anything about it. Mhm.

[04:27] >> that I guess the only thing we can try

[04:29] to do is minimize the probability of the

[04:32] bad outcome. Um but trying to stop the

[04:36] intelligence in robots.

[04:38] Um

[04:40] I mean for the longest time I

[04:44] declined to participate in AI or was you

[04:47] know you know created open AI as a

[04:51] essentially a counterweight to to Google

[04:52] because at the time they were um they

[04:55] had more or less a monopoly on AI in

[04:57] creating open AI and then having

[04:59] anthropics spin out of that and now

[05:01] currently anthropics leader in AI. Um,

[05:06] I mean the these actions have actually

[05:08] resulted in

[05:10] knock-on effects that accelerated AI,

[05:13] which wasn't really my intention.

[05:16] So, it just seems like all roads lead to

[05:19] acceleration of AI.

[05:22] So, then I'm like, okay, well, you can

[05:24] just sort of be sad about it or

[05:27] join the club, I suppose.

[05:29] >> Well, you've certainly joined the club.

[05:30] >> Yeah, [laughter]

[05:31] >> you definitely joined the club. If you

[05:33] can't beat him, join them. You know,

[05:35] >> a number of the other leaders of of

[05:38] model makers have been pushing for

[05:40] different ways to try and minimize the

[05:42] risks. Um, you know, just last week,

[05:44] Demises came out with a proposal for a

[05:47] private public regulator. Do you think

[05:50] you guys should be working more actively

[05:53] >> to get something in place to control the

[05:56] risks or do you think that's impossible

[05:57] too? Yeah, my recommendation to Dennis

[06:00] because we I talked to him for a few

[06:02] hours before he published his his piece

[06:04] was that um the most immediate thing

[06:07] that we could do is to have the leading

[06:08] AI companies um at least just meet uh or

[06:13] have a some sort of call once every few

[06:16] weeks um and just

[06:19] uh discuss

[06:22] any safety and security issues.

[06:23] >> But why why haven't you done that yet? I

[06:25] mean you don't all like each other very

[06:27] much. Yeah. But you're not all you're

[06:28] not all I mean you're also all competing

[06:31] right? You all want to be first.

[06:32] >> Yes. But that's that's where I think

[06:35] >> um the competitors can keep it keep each

[06:38] other honest. So meaning meaning like if

[06:41] um like I think it's quite difficult for

[06:43] someone in the government who doesn't

[06:44] have a deep technical understanding and

[06:46] isn't driving the frontier of of AI to

[06:50] know whether something should be

[06:51] released or not. Um however the

[06:54] competitors I think will can you know if

[06:58] given sort of a week or two to review a

[07:00] new model um can highlight issues and uh

[07:04] and have an incentive to keep all all

[07:06] the competitors have an incentive to

[07:07] keep the others honest.

[07:09] >> So you'd have a system where the model

[07:10] makers would check each other's work. If

[07:12] there was something that worried them,

[07:13] you'd tell the government. Yeah. The

[07:15] government would then think, okay,

[07:16] >> if there's something that was worrisome

[07:17] and that company was not doing anything

[07:19] to address that risk, then that would be

[07:22] the the moment for government to step in

[07:24] and take action, which is literally what

[07:26] just what

[07:28] >> does all this have to happen?

[07:31] >> Well, I think it would be good to do

[07:32] that I think

[07:35] immediately really.

[07:37] >> Yeah, that's what people keep telling

[07:38] me. we we have to get this some kind of

[07:39] system in the next 6 months because of

[07:42] the pace at which the models are

[07:43] improving.

[07:45] >> Yeah, 6 months is a long time. Um,

[07:49] >> so you think shorter than that?

[07:52] >> Well, I mean there there's so many AI

[07:55] breakthrough announcements, sometimes

[07:57] multiple per day. I mean, how many were

[08:00] there last week?

[08:01] >> A lot.

[08:01] >> I lost track, you know.

[08:02] >> Yeah, a lot.

[08:04] Um,

[08:06] >> so, so you you basically, you know,

[08:08] whether it's Demis' proposal or others,

[08:10] you need to get together and get this

[08:11] set up kind of now. It sounds like

[08:13] you're telling me.

[08:14] >> Yeah. I I I would recommend that we at

[08:17] least have some sort of informal g um

[08:21] you know weekly or bi-weekly um call and

[08:26] and and that like I said just that that

[08:28] there's that that

[08:30] um there's maybe a week or two weeks of

[08:33] of early access by competitors and this

[08:37] is why I think the incentives work is

[08:38] because if if you know competing

[08:40] companies I think are likely to

[08:43] understand if something is a security

[08:44] risk

[08:45] and they're not going to be shy about

[08:47] highlighting that that their

[08:49] competitor's models should be delayed in

[08:51] release.

[08:52] >> You guys don't much like each other and

[08:54] you don't much trust each other. You're

[08:55] insulting each other on on social media

[08:57] constantly. So, do you do you think you

[09:00] can do that with people that you dislike

[09:02] that that where there seems to be such

[09:03] kind of personal vitrial between some of

[09:05] these leaders?

[09:07] >> Um, yes. Well, I'm I'm not a fan of Sam

[09:11] album because if you started a nonprofit

[09:13] that was meant to be an open-source AI

[09:16] company and um you know owned by the

[09:21] world uh and it somehow got turned into

[09:23] an $800 billion for-profit company with

[09:26] closed source. I think you'd be like

[09:28] well wait a second that's uh the exact

[09:30] opposite of what I donated the money

[09:32] for. That that's my issue. I think it's

[09:34] a legitimate one. Um, and then Daario I

[09:38] think also is not a fan of Sam Alman

[09:40] because he felt uh, in fact the reason

[09:42] the anthropic team left OpenAI is

[09:44] because they they didn't trust Sam Alman

[09:46] otherwise anthropic wouldn't exist. They

[09:49] would still be at OpenAI. Um,

[09:52] I I think Dario is a very principled

[09:55] person. He cares about things a lot. uh

[09:58] cares about the future of the world. And

[10:00] I I think everyone I've met at Anthropic

[10:02] so far has been I'm like no one set off

[10:06] my evil detector. You know, they always

[10:07] seem seem like they're well-intentioned.

[10:10] Um and uh the road to hell is I think

[10:13] mostly paved with bad intentions. There

[10:16] are a few well-intentioned paving stones

[10:17] in there, so we we don't want to be

[10:19] complacent. Um but uh you at the end of

[10:23] the day, if we have to talk, we'll talk.

[10:24] I mean, set aside our personal

[10:26] differences for the good of the world

[10:28] type of thing.