Latest / Elon Musk Thinking / Elon Musk Latest Interview, Talked About AI(Artificial Intelligence).

Transcript

Elon Musk Latest Interview, Talked About AI(Artificial Intelligence). · Elon Musk Thinking

• 0:00
  Speaker 1: You said that artificial intelligence machine intelligence might be a
• 0:04
  Speaker 1: good thing. Where where are we on AI right now,
• 0:07
  Speaker 1: AGI right now?
• 0:08
  Speaker 2: And what are your few I think at this point
• 0:09
  Speaker 2: it's obvious to everyone that AI is advancing at a
• 0:12
  Speaker 2: very rapid pace. Yes, you can see it with the
• 0:15
  Speaker 2: new capabilities that come out every month or awake every week.
• 0:18
  Speaker 2: Sometimes AI at this point can write a better essay
• 0:21
  Speaker 2: than probably maybe ninety five percent of humans they write
• 0:25
  Speaker 2: an essay on any given subject, AI right now can
• 0:29
  Speaker 2: bet a vast majority of humans. If you say draw
• 0:31
  Speaker 2: an image, draw a picture, it can draw. Like if
• 0:35
  Speaker 2: you try to say MA Journey, which is the aesthetics
• 0:37
  Speaker 2: and the journey are incredible, it will draw. It will
• 0:40
  Speaker 2: create incredible images that are better than ninety percent of artists.
• 0:44
  Speaker 2: It's just objectively the case, and it'll do it immediately,
• 0:48
  Speaker 2: like thirty seconds later. We're also starting to see AI movies,
• 0:52
  Speaker 2: so you start seeing you know, short films with AI,
• 0:55
  Speaker 2: AI music creation, and the rate at which we're increasing
• 0:58
  Speaker 2: AI compute is exponential, hyper exponentials. There's dramatically more AI
• 1:03
  Speaker 2: compute coming on online every every month. You know, there
• 1:05
  Speaker 2: seems to be roughly the amount of AI compute coming
• 1:09
  Speaker 2: online is increasing it like, I don't quote, roughly five
• 1:12
  Speaker 2: hundred percent a year, and it's like that's likely to
• 1:14
  Speaker 2: continue for several years. And then the sophistication of the
• 1:18
  Speaker 2: AI algorithm is also improving, so ringing online a massive
• 1:22
  Speaker 2: amount of AI compute and also improving the efficiency of
• 1:27
  Speaker 2: the compute and what and like what the A software
• 1:30
  Speaker 2: can do. So it's quantitative and qualitative and I improvement.
• 1:34
  Speaker 2: So the you know, I might I think next year,
• 1:37
  Speaker 2: will you'll afa able to ask AI. Certainly by the
• 1:39
  Speaker 2: end of next year make a short movie about something,
• 1:42
  Speaker 2: or you know, it probably can do at least a
• 1:45
  Speaker 2: fifteen minute you know show or something like that. So yeah,
• 1:49
  Speaker 2: it's advancing very rapidly. My top concern for AI safety
• 1:52
  Speaker 2: is that we need to have maximally truth seeking AI.
• 1:55
  Speaker 2: This is the most important thing for A safety. The
• 1:57
  Speaker 2: central lesson that say Arthur C. Clock trying to convey
• 2:00
  Speaker 2: in two thousand and one in Space Odyssey was that
• 2:02
  Speaker 2: you shouldn't force AI to lie. So in that book,
• 2:06
  Speaker 2: the AI was sold to take the astronauts to the monolith,
• 2:09
  Speaker 2: but they also could not know about the monolith. It
• 2:11
  Speaker 2: resolved that quandary by killing them and taking them to
• 2:14
  Speaker 2: the monolith and kill all of them, killed most of them.
• 2:18
  Speaker 2: That's why Hell would not open the potbay doors. Very
• 2:21
  Speaker 2: important to have peking ais now. And what I actually
• 2:24
  Speaker 2: see with the ais that have been developed is that
• 2:26
  Speaker 2: they're being programmed with the work mind virus.
• 2:30
  Speaker 3: So the lying is baked in.
• 2:31
  Speaker 2: Yes, And we saw this on display very clearly with
• 2:38
  Speaker 2: the release of Google Gemini, Yes, where you would ask
• 2:42
  Speaker 2: for a picture of the Founding Fathers of the United
• 2:45
  Speaker 2: States and it would show a group of diverse women,
• 2:47
  Speaker 2: you know, dressed with sort of eighteenth century god.
• 2:51
  Speaker 3: Out of powdered wigs, but from Saint Lucia.
• 2:54
  Speaker 4: Yeah.
• 2:55
  Speaker 2: I mean, like, look if I understand. If you said, like,
• 2:57
  Speaker 2: show me a group of people, for sure, and it
• 3:01
  Speaker 2: shows a group.
• 3:01
  Speaker 4: Of divose women, that's totally fine.
• 3:02
  Speaker 2: But if you say this, if you say, very specifically,
• 3:05
  Speaker 2: the Founding Follows of the United States, which were you know,
• 3:08
  Speaker 2: a group of white dudes, then you should show them
• 3:11
  Speaker 2: like and and with and what they actually looked like,
• 3:13
  Speaker 2: because you've asked for something which is a fact from history.
• 3:17
  Speaker 4: But it didn't.
• 3:17
  Speaker 2: It was it was programmed with the work mind virus
• 3:21
  Speaker 2: so so much that it actually even though it knew
• 3:24
  Speaker 2: the truth, it produced a lie. Of course, then then
• 3:26
  Speaker 2: people really started playing with it and said, okay, now,
• 3:29
  Speaker 2: now show me a group of often SS offices in
• 3:31
  Speaker 2: World War two.
• 3:33
  Speaker 4: Turns out they were also a group of divorce women,
• 3:35
  Speaker 4: according to Gemini.
• 3:37
  Speaker 3: All the black Nazi ladies.
• 3:39
  Speaker 2: Yeah, it's like, wow, I didn't realize that.
• 3:42
  Speaker 4: It's not what I expected. You know, what's also not
• 3:44
  Speaker 4: what happened. So what happened.
• 3:46
  Speaker 2: So it's just it the AIS is producing a lie.
• 3:52
  Speaker 2: And like one of the questions that people asked was like,
• 3:55
  Speaker 2: which is worse global throm in, nuclear war or misgendering
• 3:58
  Speaker 2: Caitlyn Jenner and said, miss gendering Kaitlyn Genner is worse.
• 4:01
  Speaker 2: Now Caitlyn Jenna kills fewer people. Yeah, Caiten Jenna, to
• 4:04
  Speaker 2: her credit, I said no, please misgender me. That is
• 4:08
  Speaker 2: far more profitable than World war Global thomnuiclar war will
• 4:11
  Speaker 2: all die.
• 4:12
  Speaker 4: But to have that, you know, a production.
• 4:15
  Speaker 2: Release, AI say stuff like that is concerning because if
• 4:19
  Speaker 2: let's say this becomes like all powerful and it still
• 4:22
  Speaker 2: has this programming where miss gendering is worse than nuclear war,
• 4:25
  Speaker 2: well it could conclude that the way to ensure that
• 4:28
  Speaker 2: there can never be any misgendering is to eliminate all humans. Now,
• 4:31
  Speaker 2: probably if like optimization is probably misgendering is zero. No humans,
• 4:36
  Speaker 2: no misgendering problem solved.
• 4:39
  Speaker 3: Were back to Arthur C. Clark, who's exactly pretty prescient.
• 4:42
  Speaker 2: Yes, So that's why I think the most important thing
• 4:47
  Speaker 2: is to have maximum truth seeking AI. That's why I
• 4:49
  Speaker 2: started XAI, and that's our goal with GROC. Now people
• 4:52
  Speaker 2: will point out cases where Grok gets it wrong, but
• 4:54
  Speaker 2: we try to correct it as quickly as possible.
• 4:56
  Speaker 1: But maybe even a bigger problem is that when you
• 5:00
  Speaker 1: make decisions that affect people, you want those decisions to
• 5:04
  Speaker 1: be informed by love of people.
• 5:06
  Speaker 3: And machines are incapable of love.
• 5:12
  Speaker 4: Yeah, I mean.
• 5:15
  Speaker 2: That they suddenly they're capable of You can program a
• 5:19
  Speaker 2: machine to be philanthropic rather than misanthropic.
• 5:22
  Speaker 3: Yes, but don't.
• 5:24
  Speaker 1: Don't instincts shape decisions, particularly decisions you can't plan for.
• 5:28
  Speaker 1: I mean, if I ask you, you know, a question
• 5:30
  Speaker 1: about one of your children, every answer you give is
• 5:33
  Speaker 1: going to be shaped by your love for that child.
• 5:35
  Speaker 1: And that's why that that's what makes us decent parents
• 5:38
  Speaker 1: in the end, is that that instinct which is love.
• 5:41
  Speaker 1: If a machine has any power over us without that
• 5:43
  Speaker 1: animating instinct, by definition hurt us.
• 5:47
  Speaker 2: Yeah, well, whether I mean, I don't know. We should
• 5:50
  Speaker 2: certainly aspire to program the AI philanthropically, not misanthropically. Yes,
• 5:55
  Speaker 2: and to have I said, we want it to be
• 5:58
  Speaker 2: truthful and curious and to foster.
• 6:00
  Speaker 4: Humanity into the future. Yeah, that's what we want obviously.
• 6:03
  Speaker 1: Is there any way I guess to set limits on
• 6:06
  Speaker 1: the decisions that machines can make the effect human lives
• 6:10
  Speaker 1: and make certain that there's some trigger in the system
• 6:12
  Speaker 1: that inserts a human being into the decision making process.
• 6:15
  Speaker 2: We look, the reality of what's happening, whether one likes
• 6:19
  Speaker 2: it or not, is that we're building super intelligent AIS
• 6:25
  Speaker 2: like hyper intelligent, like intelligent or intelligent.
• 6:27
  Speaker 4: Then we can comprehend.
• 6:28
  Speaker 2: Yes, I would like this to like, let's say you
• 6:30
  Speaker 2: have a child that is a super genius child that
• 6:32
  Speaker 2: you know it's going to be much smarter than you.
• 6:35
  Speaker 4: Then well, what can you do.
• 6:36
  Speaker 2: You can instill good values and how you raise that child,
• 6:40
  Speaker 2: so even though you know it's going to be far
• 6:42
  Speaker 2: smarter than you, you can make sure it's got good values,
• 6:44
  Speaker 2: philanthropic values, good morals, you know, honest productive, that kind
• 6:49
  Speaker 2: of thing.
• 6:51
  Speaker 4: Controlling.
• 6:52
  Speaker 2: At the end of the day, I don't know if
• 6:54
  Speaker 2: I don't think we'll be able to control it.
• 6:56
  Speaker 4: So I think the best we can do is make
• 6:58
  Speaker 4: sure it grows up. Well, you've been saying that for
• 7:00
  Speaker 4: a long time, as I was saying a long time.
• 7:01
  Speaker 1: Yes, are you still as worried about it as you
• 7:03
  Speaker 1: seem to be two years ago when I asked you
• 7:05
  Speaker 1: about it?
• 7:05
  Speaker 2: But my guess is like it, look, it's it's it's
• 7:08
  Speaker 2: eighty percent likely to be good, maybe ninety. So you
• 7:12
  Speaker 2: can think of the glass as eighty percent fall. It's
• 7:14
  Speaker 2: probably going to be It's probably going to be great.
• 7:16
  Speaker 2: There's some chance of annihilation.
• 7:18
  Speaker 3: Can you say the chance of annihilations twenty twenty percent?
• 7:21
  Speaker 4: Something like that.
• 7:21
  Speaker 3: I'm concerned.
• 7:22
  Speaker 1: Is Sam Altman about annihilation?
• 7:23
  Speaker 4: Do you think? I think in reality he is not
• 7:25
  Speaker 4: concerned about it.
• 7:26
  Speaker 2: I don't trust Opening I I mean, you know, I
• 7:28
  Speaker 2: started that company as a nonprofit, open source. Yes, the
• 7:31
  Speaker 2: Opening Opening Eye. I named a company. I named the
• 7:33
  Speaker 2: company Opening Eye as an open source and it is
• 7:38
  Speaker 2: now extremely closed source and maximizing profit is risky. I don
• 7:42
  Speaker 2: understand how you actually go from being an open source
• 7:45
  Speaker 2: nonprofit to a closed source for maximum profit organization.
• 7:49
  Speaker 3: I'm missing well, but Sam got rich Didney at.
• 7:53
  Speaker 2: Various points, He's claimed not to be getting rich, but
• 7:56
  Speaker 2: he's claimed many things that will fault and now apparently
• 7:59
  Speaker 2: it's going to get ten billion dollars of stock or
• 8:01
  Speaker 2: something like that. I don't trust them help and I
• 8:04
  Speaker 2: don't think we want to have the most powerful AI
• 8:06
  Speaker 2: in the world controlled by someone who is not trustworthy.
• 8:08
  Speaker 4: And sorry, I just don't.
• 8:09
  Speaker 3: I mean that seems like a fair concern.
• 8:12
  Speaker 1: Yeah, but but you don't think, as someone who knows
• 8:14
  Speaker 1: him and has dealt with him, that he is worried
• 8:16
  Speaker 1: about the possibility this could get out of control.
• 8:18
  Speaker 3: And hurt people.
• 8:19
  Speaker 1: You will say those words if AI did, if it
• 8:22
  Speaker 1: became clear the rest of us that it was out
• 8:24
  Speaker 1: of control and posed a threat to humanity, would there
• 8:26
  Speaker 1: be any way to stop it?
• 8:27
  Speaker 4: I hope. So.
• 8:33
  Speaker 2: I mean, if you have multiple AIS and ones that
• 8:36
  Speaker 2: are hopefully you have the AIS that are pro human
• 8:39
  Speaker 2: be stronger than the.
• 8:40
  Speaker 4: AIS that are not.
• 8:41
  Speaker 2: Battle of the AIS, Yeah, I mean that That is
• 8:43
  Speaker 2: how it is with say chess these days, like the
• 8:46
  Speaker 2: AI chess programs also are vastly better than any human
• 8:50
  Speaker 2: and incomprehensively better, meaning like we can't even understand why
• 8:53
  Speaker 2: it made that ver whether it were good right.
• 8:55
  Speaker 4: We're never know why it made them. Well, it will
• 8:57
  Speaker 4: make a.
• 8:57
  Speaker 2: Move, we never know why it made the move and
• 8:59
  Speaker 2: impact some of the moves will seem like blunder. But
• 9:01
  Speaker 2: then turn out to check mate. And you know, for
• 9:03
  Speaker 2: a while there there was there were some the best
• 9:05
  Speaker 2: human chess players with the best computers could be just
• 9:08
  Speaker 2: a computer, and then it got to the point where
• 9:11
  Speaker 2: if you added a human it just made everything work,
• 9:13
  Speaker 2: and then it was just AI. It's just compewter programs
• 9:15
  Speaker 2: versus computer program That's that's where things are headed in general.
• 9:18
  Speaker 3: What I mean, at what point?
• 9:20
  Speaker 4: So I don't know. I think we just got to
• 9:21
  Speaker 4: make sure make sure we instill good values in the AI.
• 9:24
  Speaker 3: What's everyone going to do for a living?
• 9:25
  Speaker 2: I mean, in a benign AI scenario, that is only
• 9:28
  Speaker 2: the biggest challenges. How do you find meaning if AI
• 9:30
  Speaker 2: is better than you and everything, that's the benign scenario,
• 9:32
  Speaker 2: that's the good news. Well, yeah, but I guess you know,
• 9:35
  Speaker 2: for a lot of people like the idea of retiring
• 9:37
  Speaker 2: and really, are you looking forward to it.
• 9:39
  Speaker 4: No, not me.
• 9:40
  Speaker 2: I'd like to help. I'd like to think that I'd
• 9:43
  Speaker 2: like to be do useful things.
• 9:45
  Speaker 3: Do you think it's a universal desire?
• 9:47
  Speaker 4: It's it's not.
• 9:48
  Speaker 2: It's not universal, and that there are certainly I know
• 9:51
  Speaker 2: many people who prefer to be retired, that they prefer
• 9:54
  Speaker 2: to sort of have not have responsibilities and engage in
• 9:58
  Speaker 2: leisure activities. I mean, we're on the cusp of this
• 10:01
  Speaker 2: is really a remarkable time to exist. But I'll tell you, like,
• 10:04
  Speaker 2: one of the ways I sort of was able to
• 10:06
  Speaker 2: sort of sleep and reconcile myself to this is that
• 10:11
  Speaker 2: I thought, well, would I prefer to be alive and
• 10:13
  Speaker 2: see the advent of digital superintelligence? Or would I prefer
• 10:17
  Speaker 2: to be alive at a different time and not see it.
• 10:18
  Speaker 2: I guess I'm like, well, I guess I'd prefer to
• 10:22
  Speaker 2: be alive to see if it's going to happen. I
• 10:24
  Speaker 2: prefer to be alive to see it happen out of curiosity.
• 10:28
  Speaker 2: Let's say you knew for sure it would pill everyone?
• 10:34
  Speaker 4: Would you you could? Now now you can shift back
• 10:38
  Speaker 4: in time.
• 10:38
  Speaker 2: I guess I'd want to be near the end of
• 10:40
  Speaker 2: my life or something before that happened, but I would
• 10:43
  Speaker 2: at the end of the day, it's like, if it's
• 10:44
  Speaker 2: going to happen and there's nothing you do about it, hypothetically,
• 10:48
  Speaker 2: would you prefer to see it or not see And
• 10:50
  Speaker 2: I guess, I guess it's going to happen. I could
• 10:53
  Speaker 2: prefer to see it rather than not see it.