Elon Musk on AI: humans will no longer be in control in ten years | The Economist — Transcript
Full transcript
- 0:00I think
- 0:01AI may exceed the sum of human
- 0:03intelligence
- 0:05in about in around 5 years.
- 0:07>> In 5 years?
- 0:07>> Roughly 5 years is my guess.
- 0:10Um
- 0:12So
- 0:13there really won't be anything that AI
- 0:15can
- 0:16can't do better than humans apart from
- 0:19being human perhaps.
- 0:20>> In a more prosaic level
- 0:22what will life be like?
- 0:25>> The most likely outcome
- 0:28is an age of amazing abundance
- 0:31uh where
- 0:34anyone can have anything they can think
- 0:35of.
- 0:37This is this is this may sound
- 0:38preposterous but
- 0:40well here we are in 2026.
- 0:43Let's see where we stand in 2036. Now of
- 0:45course there are things that could uh
- 0:47derail this like a a massive global
- 0:50thermonuclear war or something like
- 0:51that. Um
- 0:53but assuming that there's not World War
- 0:55III
- 0:56uh I think
- 0:58I think we're headed for an age of
- 0:59amazing abundance. So this is I guess a
- 1:01message of
- 1:03of optimism
- 1:05and uh excitement about the future.
- 1:07>> You're sure that humans will no longer
- 1:09be in control in 10 years time?
- 1:11>> I think it is unlikely um in the same
- 1:14way that
- 1:15if you had a vastly
- 1:18more like a a
- 1:20you know
- 1:22the
- 1:24like like if if if the difference in the
- 1:26intelligence between AI and humans is
- 1:30vastly greater than the difference in
- 1:31intelligence between AI and chimpanzees
- 1:34it's hard to imagine that the
- 1:36chimpanzees would be in charge.
- 1:38>> I'm really stuck by this.
- 1:41>> um
- 1:42an evolved form of chimpanzee. So it
- 1:45wasn't that long ago that we were
- 1:47swinging through the trees eating
- 1:48bananas um in the jungle. And we still
- 1:52eat bananas.
- 1:53>> You said there was a 10 to 20% chance of
- 1:55killer robots wiping out humanity. So,
- 1:58>> you sounded worried then. Now you seem
- 2:00very relaxed about it. Have you changed
- 2:02your view?
- 2:03>> I still think there's risk
- 2:05associated with AI and robots. It's not
- 2:07zero.
- 2:08Um I I guess I've come to
- 2:12my my I sort of philosophical conclusion
- 2:15is to look on the bright side.
- 2:17Um
- 2:19I like I I can't see any way to really
- 2:21stop this
- 2:23incredible momentum of AI and robots.
- 2:26Um and
- 2:27and at times I sort of think, well,
- 2:29perhaps, you know, we're even if there
- 2:31was a stop button, we probably shouldn't
- 2:33press it
- 2:34uh because
- 2:37I cuz the most likely outcome
- 2:40is
- 2:41incredible
- 2:42abundance for
- 2:44for all.
- 2:45>> That's what I thought, actually. It
- 2:46struck me that now you seem to have
- 2:48decided it's inevitable.
- 2:51Let's hope for the best and go for the
- 2:53ride.
- 2:54>> Yeah. I mean
- 2:56yes, pretty much.
- 2:58Let's uh
- 2:59let's enjoy the ride as as
- 3:01I think my philosophy at this point. Cuz
- 3:03I mean, honestly, this this
- 3:04it's it's there seems to be just an
- 3:06inexorable
- 3:08uh progress here in in AI and and in
- 3:12robots
- 3:13that um
- 3:15>> But you
- 3:15>> you know, even if I wanted to stop it, I
- 3:18I don't I couldn't.
- 3:19>> Do you think there's still a 10 to 20%
- 3:21chance that killer robots
- 3:24mean the end of humanity?
- 3:27That's a high bar to as they say.
- 3:30>> Well,
- 3:32in long run we're all dead, of course.
- 3:33Um you know, um
- 3:35So, I mean, I I I so
- 3:36>> I mean
- 3:37>> the
- 3:37let let's
- 3:38according to
- 3:40the standard model of physics, the end
- 3:42of the universe is um
- 3:44a long, boring
- 3:47sort of thin, gray soup of
- 3:51uh
- 3:52gradual cooling.
- 3:54Um I mean, it is sort of the heat death
- 3:56of the universe is called. So, now if if
- 3:58the heat death of the universe is in
- 4:00fact the outcome for our reality, then
- 4:03it really is all about the journey.
- 4:05Because the destination is terrible.
- 4:07>> years is not [laughter] very
- 4:08Yeah, yeah, I get that, but that could
- 4:09be a long way ahead.
- 4:10>> long very very very very long way.
- 4:12>> If it's 10 to 20% in 10 years, you would
- 4:15you go into one of your rockets if you
- 4:16thought there was a 10 to 20% chance of
- 4:18it blowing up and killing you?
- 4:21>> Yes, but I I would
- 4:22>> [laughter]
- 4:23>> But let's say there's you can't do
- 4:25anything about it.
- 4:26>> Okay.
- 4:27>> But I guess the only thing we can try to
- 4:29do is minimize the probability of the
- 4:31bad outcome.
- 4:33Um
- 4:34but trying to stop the intelligence in
- 4:36robots
- 4:38um
- 4:40I mean, for the longest time I
- 4:44declined to participate in AI or was,
- 4:47you know,
- 4:48you know, created OpenAI as
- 4:51essentially a counterweight to to Google
- 4:52cuz at the time they were
- 4:55had more or less a monopoly on AI. In
- 4:57creating OpenAI and then having
- 4:59Anthropic spin out of that and now
- 5:01currently Anthropic's leader in AI.
- 5:03Um
- 5:05I mean, these
- 5:07these actions have actually resulted in
- 5:10knock-on effects that accelerated AI
- 5:13which wasn't really my intention.
- 5:15So, it just seems like all roads lead to
- 5:19acceleration of AI.
- 5:22So, then I'm like, okay, well,
- 5:24you can just sort of be sad about it or
- 5:27join the club, I suppose.
- 5:29>> Well, you've certainly joined the club.
- 5:30>> Yeah.
- 5:30>> I mean, you've
- 5:31>> [laughter]
- 5:31>> you've definitely joined the club.
- 5:33>> If you can't beat them, join them, you
- 5:34know.
- 5:34>> A number of the other leaders of of
- 5:38model makers have been pushing for
- 5:40different ways to try and minimize the
- 5:42risks. Um you know, just last week Demis
- 5:44Hassabis came out with a proposal for a
- 5:47private public regulator. Do you think
- 5:50you guys should be working more actively
- 5:53to get something in place to control the
- 5:55risk, or do you think that's impossible,
- 5:57too?
- 5:57>> Yeah.
- 5:58My recommendation to Demis cuz we I
- 6:00talked to him for
- 6:01a few hours before he published his his
- 6:03piece was that um
- 6:06the most immediate thing that we could
- 6:07do is to have the leading AI companies
- 6:10um
- 6:11at least just meet uh or have a some
- 6:13sort of call
- 6:15once every few weeks
- 6:17um
- 6:17and just
- 6:19uh discuss
- 6:21say any safety and security issues.
- 6:23>> But why why haven't you done that yet? I
- 6:25mean, you don't all like each other very
- 6:27much. Yeah, but you don't you're not all
- 6:29I mean, you're also all competing,
- 6:31right? You all want to be first.
- 6:32>> but that's that's where I think um good
- 6:36the competitors can keep each
- 6:37keep each other honest. So, meaning
- 6:40meaning like if um like I think it's
- 6:42quite difficult for someone in the
- 6:44government who doesn't have a deep
- 6:45technical understanding and isn't
- 6:47driving the frontier of of AI to know
- 6:50whether something should be released or
- 6:51not.
- 6:53Um however, the competitors
- 6:56I think what can, you know, if given
- 6:58sort of a week or two to review a new
- 7:00model um can highlight issues and uh
- 7:04and have an incentive to keep the all
- 7:06all the competitors have an incentive to
- 7:07keep the others honest.
- 7:09>> So, you'd have a system where the model
- 7:10makers would check each other's work. If
- 7:12there was something that worried them,
- 7:13you'd tell the government.
- 7:14>> Yeah.
- 7:15>> The government would then take
- 7:16>> If if there's something else was
- 7:17worrying them and that company was not
- 7:18doing anything to address that risk,
- 7:20then that would be the
- 7:23the moment for government to step in
- 7:24and take action, which is literally what
- 7:26just what occurred with my model.
- 7:28>> this all have to happen?
- 7:31>> Well, I think it'd be good to do that.
- 7:33I think
- 7:35immediately, really.
- 7:37>> Yeah, that's what people keep telling
- 7:38me. We we have to get this some kind of
- 7:39system in the next 6 months because of
- 7:42the pace at which the models are
- 7:43improving.
- 7:45>> Yeah, 6 months is a long time.
- 7:47Um
- 7:49>> So, you think shorter than that?
- 7:51>> Well,
- 7:53I mean there's there's so many AI
- 7:55breakthrough announcements.
- 7:56Sometimes multiple per day.
- 7:59I mean, how many were there last week?
- 8:00>> A lot.
- 8:01>> I lost track, you know.
- 8:02>> Yeah, a lot.
- 8:04>> Um
- 8:06>> So, so you you basically, you know,
- 8:08whether it's Demis's proposal or others,
- 8:10you need to get together and get this
- 8:11set up kind of now. It sounds like
- 8:13you're telling me.
- 8:14>> Yeah, I I I would recommend that we at
- 8:17least have some sort of informal um
- 8:21you know,
- 8:21weekly or bi-weekly
- 8:23um
- 8:24call and
- 8:26and and that like so just that that
- 8:27there's that that
- 8:30um there's maybe
- 8:31a week or 2 weeks of of early access by
- 8:35competitors. And and this is why I think
- 8:37the incentives work is because if if,
- 8:39you know, competing companies I think
- 8:41are likely to
- 8:43understand if something is a security
- 8:44risk. And they're they're not going to
- 8:46be shy about highlighting that that
- 8:49their their competitors' models should
- 8:50be delayed in release.
- 8:52>> You guys don't much like each other and
- 8:54you don't much trust each other. You're
- 8:55insulting each other on on social media
- 8:57constantly. So, do you do you think you
- 9:00can do that with people that you dislike
- 9:02that that where there seems to be such
- 9:03kind of personal vitriol between some of
- 9:05these leaders?
- 9:07>> Um
- 9:09Yes, well, I'm not I'm not a fan of Sam
- 9:11Altman because if you started a
- 9:12nonprofit that was meant to be an
- 9:14open-source
- 9:16AI company
- 9:18and um
- 9:20you know,
- 9:20owned by the world and it somehow got
- 9:22turned into an $800 billion for-profit
- 9:25company
- 9:26with closed source, I think you'd be
- 9:27like, "Well, wait a second. That's
- 9:30the exact opposite of what I donated the
- 9:31money for." That that's my issue. I
- 9:34think it's a legitimate one.
- 9:36Um
- 9:37And then Dario I think also is not a fan
- 9:39of Sam Altman because he felt in fact
- 9:42the reason the Anthropic team left
- 9:44OpenAI is because they didn't trust Sam
- 9:45Altman.
- 9:47Otherwise Anthropic
- 9:48wouldn't exist. They would still be at
- 9:49OpenAI.
- 9:51Um
- 9:52I I think Dario is a very principled
- 9:55person and he and he cares about things
- 9:56a lot.
- 9:57Uh he cares about the future of of the
- 9:59world and I I think everyone I've met at
- 10:02Anthropic so far has been
- 10:04I'm on Like no one set off my evil
- 10:06detector. You know, they always seem
- 10:07seem like they're well-intentioned.
- 10:10Um
- 10:11and uh
- 10:12and the road to hell is I think mostly
- 10:14paved with bad intentions. There are a
- 10:16few well-intentioned paving stones in
- 10:17there, so we we don't want to be
- 10:19complacent.
- 10:20Um
- 10:22but uh
- 10:23you know, at the end of the day if we
- 10:23have to talk, we'll talk. I mean
- 10:25set aside our personal differences for
- 10:27the good of the world type of thing.
About this transcript
This page contains the full transcript of Elon Musk on AI: humans will no longer be in control in ten years | The Economist by The Economist, generated from the public captions YouTube serves with the video. The transcript has 1,788 words across 333 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.