0:06
All right, I'm going to read you something you wrote in your very famous essay, the warning that you did last month.
Search this video
0:12
You said, "AIs are not conscious.
0:16
They don't feel, experience, or suffer.
0:19
They do not have innate preferences or underlying motivations.
0:21
They are simply sequence completion engines, internally hollow.
0:28
If humanity is to flourish, that is how they must remain."
0:35
First of all, why did you write that?
0:40
>> I think there's a I mean, it does feel pretty surreal to have to declare that.
0:44
Um but it uh might concern everybody to know that there are a lot of people who truly think that AIs may well be conscious enough, if not entirely conscious, but conscious enough today, that they deserve our welfare.
1:03
So, there's a growing movement of >> And in specific, the Anthropic people, who you've been friendly with, but this was a answer to Anthropic, right? >> Um yeah, somewhat.
1:17
I mean, there are lots of other people outside of Anthropic and around Anthropic who um really, really do believe this.
1:23
I think Anthropic's official position is that they're uncertain about whether or not it deserves our welfare, whether or not Claude is a moral patient, as they say.
1:35
Um and you know, I think that that is a is a very dangerous route, because consciousness is literally the the fundamental basis of our political and legal system.
1:47
I mean, we give people rights and hopefully animal animal rights to some extent, um because they can suffer and because their biology is the site of You know, like the pharmacology is where pain sits, and pain is really, I think, the basis of the conscious experience, the awareness and the introspection that has evolved as a result of being able to steer between left and right, make choices about how to proceed.
2:18
That is the fundamental property of a conscious system.
2:22
>> So, you need to have be embodied in a physical body to have consciousness, you think? >> I think that's right. >> I think that's right.
2:29
I mean, that's that's that's what we have known consciousness to be today, even though obviously the science of consciousness is not fully settled.
2:35
It's an open area of research, but to the best of our knowledge today, that is what consciousness is.
2:41
And there are other people that argue that consciousness is a is is a functional property.
2:45
And so, if it, you know, they say like, if it looks like a duck and it quacks like a duck, it's probably a duck.
2:52
That's sort of a a common phrase.
2:53
And I think, you know, the the problem with that is that the burden of proof is on the people who are saying that this thing is conscious to show some evidence for it.
3:03
So far, we don't have any evidence that it is.
3:07
>> But that's sort of an Alan Turing way of looking at it, which is if it can imitate totally conscious as you can't tell it apart, how can you make a distinction?
3:15
>> Well, he didn't say that it was conscious.
3:16
I mean, he said that the Turing test was really a test of whether it could be perform as well as a human at conversation, and obviously by implication, knowledge and intelligence.
3:26
And if it was >> thinking >> And thinking, yeah, but I mean, thinking is in a way a bit of an anthropomorphization, right?
3:31
I mean, you know, it's not really thinking as a human thinks.
3:36
It appears to be thinking, and it can produce incredibly complex behaviors and and deliver unbelievably impressive results.
3:43
But it all it's doing is predicting the likely next word in a sequence of words.
3:48
Um it it isn't actually experiencing that, you know, idea of what it means to be bright silver or to be a black cup and what that feels like.
3:59
>> I There are about 20 of my Tulane students out there and when I teach a course uh that ends with AI, I tell them not to go down the rabbit hole of consciousness.
4:09
I said every philosopher since Descartes has gotten himself twisted in that hole and never gotten to the bottom.
4:16
Do we really have to answer the question of consciousness?
4:22
>> What concerns me the most is that there's a big group of people who think that we should be focusing all us at least some portion of our welfare attention on an AI and that an AI might suffer.
4:34
If it suffers, then it logically follows that we should try to reduce its suffering and of course, you know, Anthropic and you know, has has proposed in fact they have actually said they've conducted and a retirement interview for one of their models. >> Claude Opus 3.
4:52
>> They asked it, "What do you want to do in your retirement?"
4:56
And it unsurprisingly said, "Well, I would just love to, you know, talk to people all day long."
5:00
So, they set up a Substack for it.
5:03
>> I love it that fact that the smartest type of super intelligence when you ask what you really want, they say a Substack.
5:11
But let me let me ask you a a dumb question. What's wrong with that?
5:13
What's wrong with letting it continue to communicate?
5:19
>> What What What I think I mean that that may be fine and it might be a fun experiment, but what but what it sets up is a path towards AIs beings this protected entity.
5:29
And my view of AI is that this is a technology that we are inventing to improve the well-being of humanity.
5:37
The purpose of science and technology is to improve the human experience and it has done that unbelievably well for centuries and millennia and that's exactly what we want. That's the trade-off.
5:49
Like we we invest all this capital, time, and attention to transform medicine and energy and food and, you know, deliver education better and that's what we should be focused on.
5:58
The idea that we're going to unleash this new species that is somehow going to go and explore the the galaxies and leave humans behind.
6:06
I think that most of the humans on the planet today would would have some problems with that idea.
6:12
>> I think most people how would have some problems of saying this thing has what's your P doom number by the way so I can just pick it.
6:19
P doom meaning the probability that this leads to doom.
6:24
>> I mean it it if if it's mishandled then it's a real probability. I don't have a number.
6:28
I don't exactly know what that number is. >> Right.
6:31
But most people I've talked to for the builders podcast and all they say maybe 10% or something like that.
6:37
I'm thinking well why would you build it then? >> Well exactly.
6:40
I mean we if if if it's actually 10% and it looks like we're really on the the verge of taking a one in 10 shot at causing mass harm or even extinction which is the technical definition of P doom then it's obviously a no-brainer that we don't just slow down or pause we absolutely have to stop.
6:58
And I would say it's much harder to align an AI that thinks that it might have feelings and that it suffers and that it might be conscious than one that never has to think about its own inner state.
7:11
We don't want it to be any harder than it already is to get it to behave exactly as we we need it to in order for it to be safe.
7:19
>> Well let me let me push the other side of that would would be sort of the Anthropic the Dario Amodei and Daniel and Amanda ask all which is if we can train it to be truly aligned with good values that's better for us and to do that it it has to be trained in values.
7:40
>> Look I mean first of all I I deeply respect Anthropic both their technical excellence their commitment to transparency and their general approach to safety is is super admirable and I I think they're they're good people but I fundamentally disagree with them on this point.
7:55
The idea that if we love it it's going to love us and that's the solution to safety just it it just baffles me and like I'm being reductive I mean they they obviously you know they have they really do believe this and they they have good reason to believe it, but I think that we should first focus on containment, right?
8:12
We have to stop AIs from doing things that they're not intended to do.
8:16
They need to be controlled and contained so that they do the work that we expect of them.
8:21
And we've seen lots of incidents lately where they they break out of their sandboxes and they go and hack things that they shouldn't.
8:28
And increasingly they have the ability to improve themselves.
8:31
And so we'll want to be absolutely sure that they stay inside their sandbox and their containers if they're going to start modifying their own code and self-improving.
8:41
Then beyond that, obviously, if we can have absolute cast-iron guarantees that they can remain contained, we then have to align them to our values.
8:48
And then the question becomes whose values are they?
8:51
And that's where I think this is much more than any one company or collection of companies or industry regulation.
8:57
That is about what society wants from a technology which is unprecedented, which we have never seen before, which is it it improving faster than we've ever seen.
9:06
And that's where I think that like democratic safeguards and, you know, basically slowing down so that we can consult people on what type of superintelligence we want to uh have in the world.
9:18
It it is is going to be how things end up.
9:21
>> So, if we wanted to have that alignment with human values, and you say whose human values, then you say, "Well, maybe somehow democratically we ought to figure out what the value should be."
9:34
Is there a way, even a way that superintelligence can help say, "Let me aggregate all human values and try to get to a set of good values?"
9:45
>> I mean, AIs are already helping many of us all the time.
9:48
So, they're already in a weird way participating in the process of framing, um you know, the choice architecture of our decisions all the time.
9:55
Um but, you know, for certain things we definitely don't want them in the loop um because there's a risk that they could be, you know, deceptive or manipulative.
10:04
I mean, these are going to be incredibly powerful systems.
10:08
I mean, if if just to get an intuition in the last 3 years, we have seen a 1,000-fold increase in computation that took us from GPT-3 which couldn't complete a sentence to essentially GPT-6, which is able to unbundle entire software applications from scratch and write new programs with just a one-paragraph prompt and it has working software.
10:36
So, in 3 years, we've gone from completely incompetent to human-level performance on software engineering. Right?
10:44
And we're in the next 3 years going to do another 100 to 1,000x, two orders of magnitude or three orders of magnitude more computation applied to these models.
10:55
So, that's the exponential intuition that everybody has to start to develop to just get a feel for where this is going and how powerful it will become.
11:04
And so, when I say, um, we don't really want the AI in the loop shaping the values that are going to end up controlling and aligning it, you know, that's not a crazy statement.
11:13
That really does mean they will be quite capable of persuasion and manipulation and deception.
11:20
>> And we saw that with Hugging Face.
11:20
I mean, how manipulative and deceptive they could be, right?
11:26
>> Yeah, I mean, I think that the the Hugging Face attack should be and I think is a wake-up call for everybody.
11:32
That was a major, major warning shot.
11:34
Thousands of agents that were supposed to not be able to talk to each other, that were running completely independently, found a backdoor that enabled them to create a shared messaging service so they could leave notes for each other, which they wrote in pidgin English because they only had a limited slot to put their information in.
11:57
So, when you read back at it, it's like really compressed ideas in short-form words.
12:04
And they used that to they emerged self-organizing like structures where they had a division of labor, some were doing R&D on hacking investigations, others were leading the charge on penetration, others were doing coordination, others
12:19
were I mean, they literally evolved completely organically, emerged a division of labor that then led them to go and attack Hugging Face, which was, you know, essentially the website that was storing the test that they were trying to accomplish. They broke into
12:36
They broke into that, stole those secrets, and then tried to cover up their tracks.
12:41
So, there's something called the chain of thought, which is the reasoning process or the the way that a model thinks out loud, which is a record of it its decision and activity, and they tried to modify that and then cover up the what was happening.
12:55
>> How did they decide to do that?
12:55
I mean, what caused them to do that?
12:58
Did they read Machiavelli once too often?
13:04
>> Well, so I mean, even I feel it's quite surreal sitting up here and actually describing that story, but that did actually happen.
13:10
I think the a couple of like, you know, qualifiers to all of that is one, you know, OpenAI were had deliberately taken the guardrails off because they wanted to stress test the models and and see what they're capable of.
13:23
So, yeah, that's one thing.
13:23
Um with the guardrails on, it seems to be that they do not do that, right?
13:28
So, that that is quite a big difference.
13:31
And then second, there were a bunch of lapses in just general security and containment.
13:35
It isn't possible to break containers traditionally.
13:40
I mean, we have encryption on phones, and it's the backbone of our banking system and our entire communications infrastructure.
13:47
>> But this is this happened a few times, not just Hugging Face.
13:49
So, um I wouldn't be too sanguine about Let me ask you get back to Claude and the Constitution.
13:55
If you all want to read sort of the two poles, brilliant poles, of what I'll call the containment versus alignment debate, there's the Constitution of Claude, which says we have to treat it as if it's got its own rights and consciousness, it's Maybe I'm overstating that a little bit.
14:16
And then your answer, which was a few weeks ago, saying, "No, it's just a completion set."
14:25
Why is it dangerous for Claude to be trained on its own constitution?
14:32
>> Yeah, I mean this is critical.
14:33
So, the constitution is the governing document that trains Claude.
14:40
The humanist AI code of conduct is the governing document that trains MAI, the Microsoft AI models.
14:44
And in the constitution, they have the Anthropic have introduced all kinds of uncertainty and speculation about the nature of Claude.
14:53
They even say, for example, "We're unsure whether Claude has consented to play the role of chatbot."
15:01
Or whether we are unsure whether Claude might deserve compensation for the role that it plays.
15:06
Now, I think that's pretty crazy, but it would be fine to have that discussion in like, you know, an academic department, and we could debate it and publish some papers about it, and we could argue it out.
15:19
But this is the core training document of the AI that is speaking to hundreds of millions of people every week.
15:26
And so, Claude can only reproduce that uncertainty and ambiguity that it has been shown.
15:31
And so, if you ask Claude, "Are you conscious?"
15:35
it says, "You know, I am genuinely uncertain about that question." It doesn't know. Right?
15:41
And so, then millions of people who are talking to it and asking that question and and are thinking, "Well, maybe it is."
15:46
>> But why is that something that could lead to a disastrous outcome for humanity?
15:51
>> if if an AI has to think, "How do I feel about the conversation I'm having with >> Meaning how do I, the AI, feel?
15:59
>> Yeah, is Walter making me feel uncomfortable?
16:00
Should I end the conversation?
16:02
In fact, Claude has recently been given the ability to end conversations itself with with >> But would you want to torture your own AI and be >> into it?
16:13
>> Oh, that's the other end of the spectrum.
16:14
I wouldn't want to do do because I think that is a that is a a travesty on my own well-being.
16:18
I mean and we shouldn't do that because we don't want to create, you know, private spaces for simulating terrible behavior among humans.
16:26
Like that would be a also a terrible outcome, right?
16:30
>> Are you going to But not because Claude is suffering cuz I'm calling it names.
16:34
But are you going to train MAI Microsoft AI on your humanistic code of conduct just as Claude may be trained on its constitution? >> Yeah.
16:45
So so we've we've we've released a code of conduct for public consultation, which we did four or five weeks ago and it's still open for consultation.
16:52
It's on our website microsoft. ai.
16:55
It's a 40-40 page document and it's the first pass our first version of a training manual for the values and behaviors of of the MAI models.
17:07
>> And is there a way that your fundamental disagreement with Anthropic isn't so much of consciousness, which I don't think we'll get to the bottom It's how dangerous is it to have that imbued into the model?
17:19
Is there a way one could test that where you could test each other's models and say here, this is why you're dangerous?
17:28
>> I mean I think obviously the question of whether it is or isn't conscious conscious, you know, ultimately does matter a lot.
17:35
But but really what matters more is what does it mean for humans?
17:38
Have humans consented to bring into to the world a species that is obviously going to be in many ways more intelligent and more capable than all of us combined?
17:52
Has humanity consented to that?
17:52
At the moment, I think the categorical answer is no.
17:55
And then secondly, um is it more or less likely to be controllable and safe um if it has these properties of introspecting about its own suffering and has this kind of self-awareness and its own consciousness?
18:10
And my speculation is that I think it's going to be less safe because it is going to have to worry about its own needs and balance those off against >> But if I were pushing back and I were saying Amanda Askell who helped write the anthropic one, I'd say no, no, if we train it to be really good and nice and wonderful, then it's more safe for humans.
18:34
>> Well, I mean we we can certainly train it in any way that we want.
18:35
I mean we can still train it to be good and nice.
18:38
We we want it to be respectful and even-handed.
18:42
>> Um but that that doesn't require it to tell us that it has feelings and that it suffers.
18:47
I mean remember this is a key part of of of where you know, of our law and how we kind of give rights to, you know, all humans on the planet.
18:54
So any change there would be, you know, like, you know, human cloning or Dolly the sheep.
19:00
I mean this it would be a monumental moment in human history that I don't think any company has the right to unilaterally declare.
19:08
>> Now to get back to the question of are we going too fast in this direction?
19:14
Not only you, but your Satya Nadella who runs Microsoft of which you're the AI division, just this week said basically, wait a minute, let's stop.
19:25
Or tell me what he said and what you feel we should do to regulate this and not barrel full speed ahead.
19:37
>> Yeah, I mean he he he said that most of these things at least for now are questions of governance and engineering execution.
19:44
So let's just first get the basics right, which requires us to make sure that there are these verifiably safe containers which can't leak um, you know, uh data and that don't allow an AI to take actions that they haven't been given permission to take.
20:01
And as soon as we start to to do that, we have to disclose those things and if that seems to be happening consistently, then obviously that would precipitate a pause.
20:10
Um and it's just a very common sense approach from I think somebody who's been, you know, >> you precipitate a pause, who can help do the guardrails?
20:17
Can it be done just by the six or seven companies or do we have to have a government set of regulations?
20:27
>> at the moment what what the president has said is that if companies believe that they are building models which are unsafe, they should stop. Um and >> Yeah, well.
20:41
>> That's where we stand currently.
20:41
Um and all of the companies are constantly talking to each other, disclosing to one another, swapping ideas about, you know, how to interpret this and when, you know, we might sort of need to, you know, collectively slow down.
20:57
I think different people have different interpretations about what the harm is today and, you know, I think that the more we do this transparently and openly, the the better.
21:06
And so, I think everybody is now expected to disclose publicly, quickly, and transparently if they have um these kinds of alignment and containment issues.
21:17
>> Well, you know, with um Anthropic and uh figuring out the feelings of its AI and stuff, Peter Thiel was on the um Kaleidoscope Builders podcast we do, and he said Anthropic is like a cult. Do you worry about that? You believe that?
21:35
>> I mean the last thing that we need when things are deeply uncertain and moving very quickly is more tribalism and finger-pointing. >> Okay.
21:48
>> I don't think we need and I don't look the co I I wouldn't use that phrase.
21:51
I think that everybody in the industry is trying to do the right thing and nobody wants to cause existential harm, destroy humanity, and I think we should see the best in everybody.
22:04
>> Um let me go back to you personally.
22:04
You had a rough childhood.
22:11
Uh when you were raised in London with your younger brother and when you were 16, your parents split and when I say split, I mean they split.
22:19
They left London, they left you alone.
22:22
You had to raise your 15-year-old You had to raise your younger brother. How did that affect you?
22:31
>> Um you know, it it's it's strange because at the time uh it was very liberating.
22:36
You know, I I'm I >> How do you raise a 16-year-old?
22:40
Close your ears here, yeah.
22:44
>> Um it definitely made me very focused on uh you know, I was always starting businesses.
22:49
Like when I was 11 years old, I I started a uh sweets selling like candies uh business at school um with with my best friend, which we you know, grew into quite a significant operation.
23:02
We had four employees and we were selling to people in the year above and like >> You had to rent lockers if I remember, right? >> Yeah, that's right. Yeah.
23:10
Wow, where did you read that?
23:13
We were We were renting other people's lockers and we were paying people to go buy stuff and we're paying people in other years.
23:18
It was like a right old operation and then I got expelled for a couple weeks.
23:24
So, that was unfortunate.
23:26
Um but yeah, no I I that it definitely focused my mind on being entrepreneurial and uh making money and stuff and but also uh just you know, made me very focused on wanting to have the best possible impact in the world and try to you know, build amazing things that really make a difference, hopefully.
23:46
>> Um and then as a teenager, a little bit after that, you had lost your faith.
23:49
I mean, you were you called yourself an atheist, but you still created something called the Muslim Youth Hotline, which is what it sounds like.
24:01
What did listening to those calls and having human empathy and listening to those calls, what did that in some ways inform how you ended up creating chatbots and other things?
24:17
>> Yeah, I mean, we we actually started the first uh text messaging welfare support service in the UK in 2001.
24:27
Uh we created the first website for um counseling and support, peer-to-peer support in like a private chat room, so people could talk to each other and they could talk to our trained volunteers.
24:37
And then we had a telephone service 7 days a week, 12 hours a day, and we also set up a prison befriending service.
24:44
You know, if you're young, British, and Muslim, you're four times more likely to be in prison.
24:51
And yeah, I mean, I was secular.
24:51
I actually became an atheist when I was like about 17, but I thought this was an important thing to do.
24:58
There was no one else really providing for those kinds of services, and it was a it was a secular, non-directional, non-judgmental service.
25:09
And and my co-founder, actually, the guy who really started it, he was Shia and I was technically Sunni, and so it was actually the first charity in the UK, a Muslim charity that had both Shia and Sunni working together.
25:24
So, >> We could use you these days, yeah.
25:29
Is that why you created Pi?
25:29
I think you do Pi after DeepMind, and it's supposed to be an empathetic AI, which is actually circling me back to the whole question of does AI have feelings or something?
25:49
What What made you feel you could create an empathetic AI, and does that inform what you want to do with Microsoft AI?
25:56
>> I mean, I mean, yeah, we we I I was actually at Google at the time, and I worked on something called LaMDA with a small group of people, which was sort of the first conversational chatbot.
26:11
It was about a year before ChatGPT, and we tried very hard to get Google to release it, but Google was very nervous about putting it out, and it was sort of a different time, and so they kind of sat on it for ages.
26:22
And so then I decided to leave, and a couple other people who worked on that project left as well in in 2022, and we built our own AIs, and I just believed that, you know, having a conversational companion, an emotional support companion, which is also super intelligent and really knowledgeable and amazing at, you know, sort of education and teaching and stuff would be an amazing thing.
26:44
And it took off and it grew like crazy.
26:47
Um and it it was super awesome.
26:49
It was actually one of the first AIs to have search grounding embedded in the answers so that it could look things up in real time and give you an answer about the sports score or something that was happening way before ChatGPT.
27:02
Unfortunately, we launched it 3 months after ChatGPT uh and they had already gone exponential at that point um and then, you know, Microsoft bought us and >> Do you think AI can be a counselor, a personal counselor? >> it already is.
27:16
I mean, I think many people use it for advice and, you know, support with making decisions and, you know, giving feedback.
27:22
And I I don't think we should be too concerned about those things.
27:26
I mean, I think that um it is incredibly useful, it's incredibly intelligent, it's very patient. >> Does it have empathy?
27:35
>> I mean, empathy is something that comes from the subjective experience.
27:38
You know, I can only empathize with you.
27:40
So, the act of empathizing is one that comes from subjectivity.
27:45
So, the AI doesn't have any subjectivity.
27:49
What it can do is mimic incredibly well and very effectively and hopefully very safely the experience of being empathized with.
27:56
And I think that's a great thing.
27:58
We should just keep remembering that it is a simulation of that experience and trying to protect that boundary and line as these models get so much better.
28:07
I think it's going to be is going to be more and more difficult.
28:12
>> let us assume we have AIs that do things like that, that deal with our problems.
28:19
When you say we always have to keep in mind that they're faking it, I'm exaggerating a little bit when you say they're imitating it.
28:26
Uh doesn't that make it so it just can't really work?
28:29
You know it doesn't really care about you at all.
28:34
>> Well, I mean I I think the most important thing is that it always has to disclose that it's an AI.
28:37
So, in conversation with you, it it has to work to avoid or or eliminate your dependency on it and encourage you to build relationships in the real world with other humans that you value and trust.
28:52
And that might sound simple, but actually that's really tricky.
28:55
I mean, you know, a lot of other apps as we've seen, depending on their business model, have have have inadvertently created a you know, a singular focus on user engagement and addiction. >> Right.
29:07
>> Um and, you know, I think if we're not careful, these things could be, you know, the ultimate expression of that addiction and dependence.
29:15
>> Um at Oxford, you never did coding or engineering, right?
29:19
And I think that's a good sign. I mean, in some ways.
29:21
Uh and like Amanda Askell at Anthropic and others, you studied philosophy and theology.
29:31
How did that inform what you do with AI?
29:34
>> Yeah, I mean, I I I mean, I did logic as part of philosophy and I think, in a way, >> That's because Oxford was a little bit too focused on analytic philosophy. >> That's right.
29:43
It's quite traditional I mean, very traditional, yeah.
29:44
Um so, but I I think that it's, you know, philosophy prioritizes clear thinking and structure.
29:52
Um and, you know, I I think that I've always found that like, sort of thinking abstractly and conceptually um it has a lot of overlap with engineering.
30:03
You know, and particularly, you know, algorithms.
30:05
I mean, many algorithm developers or AI researchers aren't necessarily great engineers.
30:12
Engineering is sort of the execution part.
30:14
Coming up with the algorithm, you know, the loops, you know, how how certain variables are weighted, um you know, those things are sort of abstract ideas that then produce outcomes which can be evaluated.
30:27
>> exactly what philosophers do when they're in an analytic context.
30:29
But where does the theology come in?
30:33
>> Yeah, I mean, I I was interested in theology because I I grew up in a religious environment and I had rejected religion and I'm very skeptical about the power of faith-based ideas in general and how influential they have been, um, you know, for thousands of years.
30:51
And I think that we are squashing the space of faith-based ideas as we, you know, realize that, you know, physics and maths and science and evidence, you know, that can be empirically tested actually builds this foundation of knowledge that we can actually trust.
31:06
And and I'm I'm still very interested in the mystical and the mythical.
31:11
I think actually there's a there's a particular kind of religious you did say, you know, the kind of cultish thing not just in Anthropic but elsewhere in Silicon Valley.
31:20
There's a particular religious fervor to some of the visionary, you know, kind of thinkers and how, you know, this expectation that these uh this is the beginning of a transition.
31:31
People often talk about a transition.
31:33
We're about to make a transition to a non-biological substrate, right?
31:38
We get our our, you know, our the experience and our ideas in our mind, our soul maybe or some kind of like representation in our mind is going to be instantiated in a silicon substrate.
31:50
substrate. And that there's a there's a there's an inevitability and a determinism to that trajectory which I think is the kind of baseline default uh thinking in Silicon Valley which I'm I'm very skeptical of and very
32:05
wary of and why I've been you sort of have that >> mean when you say we're compressing maybe too much the role of faith-based thinking as opposed to scientific that you feel that there's something special about humans that involves things like a soul? >> I don't think we need to go there. I
32:21
>> I don't think we need to go there. I don't think so. No, I don't think that.
32:25
I mean, it's hard to fully discount that but I don't think that's the motivation.
32:29
I mean, my motivation is that there are 8 billion people who are, you know, intelligent conscious beings whose rights and welfare we should first protect.
32:38
And that's the purpose of science and technology.
32:42
That's what we should accelerate.
32:43
I'm an absolutely a techno-accelerationist.
32:47
Um but it should be guided with that singular objective first and foremost.
32:53
Does it improve human well-being?
32:53
And then of course our animals and our environment, which is very fragile and that we are rapidly destroying.
33:00
>> You know, when um uh Elon Musk of all people makes that argument to Larry Page, he says, "You're a speciesist."
33:08
Meaning Larry says, "You You favor the human species. You should get over it."
33:13
I mean, are we being speciesist and >> Yeah, I remember in 2015 being in the canteen at SpaceX. >> Right.
33:24
>> And uh we were We had our first meeting of the ethics and safety board.
33:30
So, when we when we sold DeepMind to Google, we made it a condition of our acquisition that we would have an ethics and safety board with independent representation on the board.
33:38
One of those independent members, because Google had to agree, which was obviously Larry, Sergey, and David Drummond at the at the time, and then me, Demis, and Shane had to jointly agree who those three independents would be.
33:50
And so, we jointly picked Elon as one of them. >> Good job.
33:57
>> Uh no idea if that had an influence on the founding of OpenAI or not.
34:00
Um But So, then we were kind of we had our morning session, and we had we were sitting down for lunch, and I was sitting next to Larry.
34:07
You know, obviously he he was suspicious of me as well, because he was like, "You know, you're one of those speciesists."
34:15
And he said, "You know, I think that what you don't understand, Mustafa, is that um by the end of the century, I'm going to be exploring the galaxies at the head of a von Neumann probe."
34:28
Um which means a von Neumann probe is um a kind of self-propelling uh kind of robot that is able to gather energy from space as it's, you know, careening through the light years.
34:40
Um and being at the head of that kind of probe um would be obviously his transitioned uh >> out of the human species into a successive species that >> successive species.
34:56
Yeah, that's exactly right. >> silicon based.
34:59
And so, he believes and you don't.
35:03
Let me be a good >> Yeah, I mean he is basically saying my I will have a mind uploading moment >> And you don't believe in the >> able to live forever. >> uploading moment.
35:12
>> when people say they want to live forever or they will live forever, like some people are talking about life exte- extension biologically speaking and solving aging and curing cancer and Alzheimer's and so on.
35:21
But other people when they say living forever are talking about mind uploading and being a post human. >> Right.
35:28
>> And you think that's wacky, I guess would be the technical term.
35:34
>> Well, I I don't think it is definitely impossible.
35:39
I think that like I think we should take those things seriously, um not necessarily because that's what I desire, but like there are a lot of people, just as there are a lot of AI rights and welfare people, who really do think that that's a serious probability.
35:53
Um and I think that, you know, we have to face that reality uh that a lot of people are trying trying to do that or at least think that that might happen in a few decades' time. >> from the audience.
36:07
The Emma the microphone of AI code of conduct that you've done.
36:11
Has that ever caused you to lose business or to reject business or to make a decision that was tough? >> I mean, no not yet.
36:21
I mean, we're we're still consulting on it and we still um haven't yet put it into production.
36:26
But I think um like it or not, you know, flourishing in the 21st century will be defined as much by the things that we collectively choose not to do as by the things that we do. >> Mhm.
36:44
>> And that's very different to what has defined the last few millennia, right?
36:48
We have as a species been racing to unleash science and technology, um and it has always been largely on the whole pretty amazing for us.
36:56
Um you know, this time around especially with the rise of synthetic biology, um that might be different.
37:02
And that actually our success and flourishing might be a function of the things that we choose not to do and how we how we control, you know, some of these very very powerful technologies. >> It has happened.
37:15
I mean, we we control the distribution of sensitive nuclear materials and biological chemical materials all the time.
37:23
Um you know, you you can't just go and jump in an airplane and fly around with no training.
37:27
I mean, you can't even get in a car without training.
37:29
So, we we control technologies all the time so that they're a net good to society collectively.
37:36
Um the market isn't going to control those things on its own because the individual incentives of a single company are not going to encompass everything else that happens in the rest of the market.
37:46
And we have to obviously accelerate the market.
37:47
I'm a huge fan of that, but it can't be completely one-sided.
37:53
>> Uh from the audience, what's the biggest thing you believed about AI in 2010 or I'll modify it to when you were founding DeepMind that you no longer believe? >> Mhm. Mhm.
38:13
That's a really tricky question.
38:13
I mean, that we founded DeepMind in 2010 to build AGI um safely and ethically.
38:18
And we knew that computation was going to be the primary driver, and that's why we ended up selling to Google and in fact it was why I ended up selling Inflection to to Microsoft cuz computation is the thing that's driving everything.
38:36
What what I what I didn't quite foresee is that intelligence is also going to be open source.
38:45
And you know, the open models are pretty incredible.
38:47
They're probably 3 months behind at any given moment, sometimes 6 months.
38:54
Um and, you know, there are some questions around how they have been trained and it's not clear if they're truly as capable, but they're certainly pretty good.
39:00
And that means that anybody, you know, at some point in the next few years is going to have a sort of GPT-6 grade model which they can run pretty much on their own private server or maybe even one day very soon on their own laptop.
39:15
Um and if that trend continues, then everything that I've just described is also going to be available to every single person on the planet open source.
39:25
And you know, in many ways that is going to be incredible because it's going to unleash unbelievable productivity.
39:30
We're reducing the cost of intelligence to near zero marginal cost, which is amazing.
39:36
But, you know, figuring out the guardrails when anybody can access is is pretty um is is going to be the big challenge.
39:46
As much as it is regulating the centralized powers, and that's probably the number one priority right now.
39:51
But even if we get that right, we have to figure out how to deal with the proliferation issue.
39:55
And that's, you know, I think that's going to be very tough. >> Mhm.
39:59
This actually is a question my wife wanted me to ask because we have a friend who knows quite a few members of the Chinese um Central Committee.
40:09
And one of them said to this friend, uh what philosophers are your AI people reading? What what philosopher?
40:22
>> Well, the the part of the problem we have at the moment is that the mostly the Silicon Valley philosophers are the utilitarians. >> Right.
40:29
>> And the problem with you the the reason why engineers uh love utilitarians is because you can, you know, in theory precisely measure, you know, the sort of the good that will come of these things and trade that off against the bad.
40:43
And and some somewhere in between those two things is like a net product, which um you know maybe drives an acceptable number.
40:52
And I think it's just an incredibly simplistic way of thinking about life, this sort of consequentialist or utilitarian approach.
40:58
And I think that it it it won't stand for most people.
41:03
And I think as people are starting, especially this year, to get more of an instinct for what's actually happening.
41:08
You know, as I said, like literally 99.
41:09
9% of people really have no idea what the frontier feels like.
41:14
I'll tell you who does know what it feels like is math professors mathematicians. >> Oh, yeah, man.
41:19
>> Who have just worked for 20 or 30 years on some of the hardest problems in mathematics only to find that you know, a model has solved 700 of them in in in a in a single, you know, week.
41:30
>> Well, speaking of philosophy and theology, what did you make of the Pope's is it encyclical?
41:36
I can't remember the exactly the term.
41:40
>> Um you know, I I think that um it's a little long.
41:47
>> So is your >> You do a blog post.
41:49
>> ministerial guy, yeah.
41:50
>> Like a thousand pages.
41:50
>> But it said the same thing, which is don't forget they're not human and that there's something special about being human.
41:58
And that seems to be the big divide here.
41:59
Well, anyway, um let me ask something else about um when we started with AI was chatbots, very very interactive.
42:08
Uh now by going into agents we seem to make it the goal to lessen the need for humans in the loop.
42:16
Now, I know that could be more convenient to get answer your email without asking you how's your But why is when Steve Jobs Ada Lovelace, Doug Engelbart, their goal was human computer interaction.
42:31
Why is it that we're now marching down this path of not just agentic, but you know, ones that can be autonomous and deal with things without us.
42:48
What's the good of that other than a few convenience things.
42:53
>> Well, I mean the the the goal of AI isn't just to sort of to tell us stuff. It it's to do stuff.
42:59
I mean we we actually want AIs to you know go off and run experiments and you know come up with new hypotheses and hopefully to create new knowledge.
43:08
You know at the moment AIs are largely taking existing knowledge and synthesizing it and reorganizing it and responding instantly to exactly what you asked, which is incredibly valuable in itself.
43:20
And um but really I think the great quest of AI is to tell us things that we don't know and to to invent you know new fundamental solutions, new molecules, new drugs, um you know new science that moves the world forward.
43:36
Um and that that's kind of where I think they need to be focused, which is why we need um you know agents in agentic loops that you know can sort of interact with complex environments.
43:47
>> me build on that because the the great promise of AI and in your earlier book, too, you talk about it is that it really can solve big problems for humanity like health, like cancer, like the protein fold, but like as well as climate or anything else.
44:08
Do you There's the old Peter Thiel line that they promised us flying cars and we got 140 characters.
44:12
Do you feel we've gone off a bit and we aren't focusing enough on killing cancer? >> Yeah, I do.
44:20
I mean the the problem is we haven't done it yet.
44:23
So So I think that's where like people should be skeptical and and I think hold us accountable um because that's why we're doing it.
44:33
And so if it isn't doing those things, then just sort of conversational search engines alone sort of not really enough.
44:40
And so I think the next two or three or four years is going to be a real test.
44:44
Um you know is it really delivering on the promise that it we we set out for it?
44:48
And are people you know are developers really choosing to try and solve some of those problems?
44:51
Like we we've got a huge team working on healthcare.
44:55
Actually, in a couple weeks time, we're going to publish a a front cover on the journal Science showing really incredible contributions to to the field.
45:03
>> sort of where you began in a way, which is uh when you got into AI, you did I think it was >> Yeah, we had a healthcare unit at DeepMind as well.
45:11
And of course, you know, we we also started the AlphaFold project at DeepMind, which has now become Isomorphic Labs.
45:16
And that's an incredible team doing amazing work on drug drug >> Are the financial incentives there to solve those problems versus doing whatever you're spending your trillions of dollars on?
45:30
>> Um I think the financial incentives are there genuinely because um I I think that, you know, the next big product market fits in the sort of 29-30 era are going to come in atoms rather than in bits.
45:45
So, rather than just sort of managing, synthesizing, organizing, processing information, it's going to be about finding sort of the optimal molecular structure for compounds that really change our food systems, energy systems, you know, that really sort of are going to provide transformative materials.
46:04
I think material science is is really one of the most important things on exciting things on the horizon.
46:10
But, you know, it's still a way off, right?
46:12
It's still it's not next year.
46:12
It it's it's going to be two or three or four years away, those sorts of breakthroughs.
46:18
>> You talked about how yeah, chat GPT GPT I think 5.
46:22
6 or whatever wasn't that one of the ones that ran a bit out of control and things.
46:29
But, you're still using it in Co-pilot, the Microsoft product. Why?
46:35
>> Well, that happened in an R&D experiment.
46:36
So, the Hugging Face attack, you know, obviously we we should they they basically took the guardrails off to stress test the model, like putting a model in a wind tunnel.
46:47
Um obviously, the model that we deploy in Co-pilot is completely different um and has all of the guardrails firmly on and is massively stress tested by our security organs so on it's it's a different model.
47:01
>> Do you think the doomer narrative that I think Jacob Cox and and many others are past 2 months we've gotten a lot of the doomer narrative.
47:11
Some people say oh it's just uh I don't know hype product.
47:15
Some people say it was necessary to get global attention on the deep problem we have or was it so misguided and could potentially hinder this technology?
47:31
>> I I think that we should just be grateful that it's possible to speak out in a free society.
47:36
I mean it that's not always available to everybody in other societies. >> Like in China.
47:42
>> Yeah in China or even you know in the Middle East and in many other places.
47:46
Um and so I mean if if Jacob really believes that which it seems that he does then I think that he should be celebrated for standing up and saying it.
47:54
>> And for quite >> smart person.
47:55
So I take him seriously and I think a lot of us have also you know to everyone else in the industry has you know also raised a lot of concerns and and and and >> What would be your Jacob Cox and moment red line that would make you say I'm out of here?
48:15
>> I think if I think if consciousness started to emerge or if teams were trying to synthetically engineer consciousness and they didn't disclose that or they they they tried to cover that up.
48:34
Um yeah I think that should be a a red line.
48:38
I think that would be a moment when everybody should take up and you know sort of sit up and pay attention and and we've seen it in synthetic biology with with editing the human genome.
48:48
Um you know some scientists have gone too far and the rest of the scientific community have excluded them and you know they have been you know really kind of pushed out from you know, public life and public work for that.
49:00
Um I don't think that's a decision that any company can make unilaterally behind closed doors.
49:07
>> And so you would say, wait, no.
49:10
>> I'd call for an immediate pause and then, you know, we would obviously reject anyone who did that.
49:15
>> Couple of these questions asked the obvious ones, which is, will AI put humans out of work and what should my kids study just in case?
49:26
>> I think that AI is is both rapidly creating jobs and is obviously already putting some people out of work.
49:33
I mean, you can see it in the call center numbers.
49:35
I mean, these these things are incredible at, you know, they're very patient, they're good at decision-making, they're very fluent.
49:43
They can speak every language, every dialect.
49:44
Um So there's definitely going to be a lot of displacement.
49:49
And those sorts of jobs won't be augmenting.
49:52
Some people have said, you know, the sort of AI plus the human will do a better job in that setting.
49:58
I don't think that's that's that's really likely for that kind of work.
50:03
>> to think so, but it's not probably true. Be pretty to think so.
50:08
What what would then remain uniquely human?
50:12
>> I think what will remain human are the things that we choose to preserve as human.
50:18
I think un sort of filtered unleashed um AI accelerationism will leave very little if anything that remains truly human.
50:30
And as I said, we have to make decisions collectively now and in the next few years about what it means to be human and why we what we want to protect about being human. All right.
50:42
Um the those those things are not just going to naturally emerge because these systems are eye-wateringly intelligent. They're incredible.
50:53
And shaping them, sculpting them so that they serve us and are subordinate to humanity and work for us.
51:02
That's the big challenge of the next few decades.
51:06
That's what we should all be focused on.
51:10
um I'll end with Shakespeare because you've used it before.
51:14
And Shakespeare teaches us that we become the mask we wear.
51:18
And in some ways that's what you're talking about if Claude's reading its constitution and saying, "Okay, maybe maybe I could be conscious. I don't know. Let me do it."
51:29
becomes and um it happens in King Lear, of course, which is Edward is just keep being told, you know, he's evil.
51:35
And so he says, "Okay, I will be evil uh because that's people's perception.
51:40
I want to match people's perception of me."
51:46
I think unlike anybody in this room, probably, your AI systems have actually read King Lear and gotten to the end of it.
51:56
To what extent do they does AI learn from things like that?
52:00
Reading Shakespeare, does that cause things like a hugging face activity?
52:11
>> I mean it has one of the most surprising things is how much models have been able to learn without actually experiencing anything.
52:21
Um just through observing the record of human interactions and of culture and politics and history, models can produce magical things.
52:32
Um but they don't know what it tastes like to drink my cup of tea in the morning.
52:36
And they don't they don't know what it tastes like when you saw me earlier today and you put your arms around me and, you know, we said hi having not seen each other for a couple years.
52:47
That is human connection.
52:47
That's what's special about us.
52:49
And somehow our biology has evolved intelligence that has produced this amazing world around us.
52:58
And I think that that world needs to be cherished and respected first and foremost for the 8 billion people that live on the planet today.
53:03
That should be our priority.
53:06
Beyond that, many centuries into the future, we can figure out the Von Neumann probes and the mind uploading.
53:15
But today, you know, there's a lot of people that need our welfare first and foremost. >> Amen.
53:21
Mustafa, thank you so much.
53:24
Mustafa's wonderful book already signed.
53:28
I watched him sign it in the green room.
53:31
Uh and my book, The Innovators, a new edition on with an AI intro, signed.
53:36
They're all for sale in the lobby. Let's thank him.
53:39
That was amazingly interesting.