In this new episode Steven sits down with philosopher, neuroscientist, podcast host and author Sam Harris.

In 2004, Sam published his first book, ‘The End of Faith’, this stayed on the New York Times bestseller list for 33 weeks and won the PEN/Martha Albrand Award for First Nonfiction. He has gone on to author 5 New York Times bestselling books published in over 20 languages. In 2009, Sam obtained his Ph.D. in cognitive neuroscience from the University of California, Los Angeles. In 2013, he began the ‘Waking Up’ podcast which covers subjects from meditation to AI. Sam is also the co-founder and CEO of Project Reason, a nonprofit foundation devoted to spreading scientific knowledge and secular values in society.

In this conversation Sam and Steven discuss topics, such as:


 How to change peoples beliefs
 Why he is not optimistic about AI
 How to live an examined life
 Why you become what you pay attention to
  The reason the mind is all you really have moment by moment
  Why AI is not aligned with human wellbeing
  How it is too late to turn back the progression of AI
  The danger of misinformation
  Why we’re going to have to abandon the internet



You can purchase Sam's book, 'Waking Up', here:

https://bit.ly/3Qp51D7

Sam has kindly given DOAC listeners 30 days free trial on his app - Waking Up. Here is the link: https://bit.ly/3QxIrrZ



Follow Sam:

Instagram: https://bit.ly/3DHwOHy

YouTube: https://bit.ly/3DE8RAy





Watch the episodes on Youtube - https://g2ul0.app.link/3kxINCANKsb



My new book! 'The 33 Laws Of Business & Life' pre order link: https://smarturl.it/DOACbook





Follow me: 

Instagram: http://bit.ly/3nIkGAZ

Twitter: http://bit.ly/3ztHuHm

Linkedin: https://bit.ly/41Fl95Q

Telegram: http://bit.ly/3nJYxST





Sponsors: 

Huel: https://g2ul0.app.link/G4RjcdKNKsb

Whoop: http://bit.ly/3MbapaY

And there's something inherently dangerous for the dumber party in that relationship.

You just can't put the genie back in the bottle.

He goes into intellectual territory where few others dare tread.

The gains we make in artificial intelligence could ultimately destroy us.

If your objective is to make humanity happy and there was a button placed in front of you and it would end artificial intelligence, what would you do?

The idea that we've lost the moment to decide whether to hook our most powerful AI to everything is just, oh, it's already connected to the internet, got millions of people using it.

And the idea that these things will stay aligned with us because we have built them, we gave them a capacity to rewrite their code.

And I worry about the near-term problem of what humans do with increasingly powerful AI, how it amplifies misinformation.

Most of what's online could soon be fake.

Can we hold a presidential election 18 months from now that we recognize as valid?

I worry we're just going to have to declare bankruptcy to the internet.

If your intuition is correct, are you optimistic about our chances of survival?

2 years ago I started writing a book based on everything I've learned from doing this podcast and meeting all of these incredible people, and also from my career in business, from running my marketing agency Flight Story, my software business Third Web, my investment fund called Flight Fund, and everything else I've been doing in business and in life.

From this I've created a brand new book called The Diary of a CEO, 33 Laws for Business and Life.

If you want to build something great or become great yourself like the guests I get to interview, I ask you please, please, please read these 33 laws.

If you like this podcast, this book is for you.

The link is in the description below wherever you're listening to this.

I watched that TED talk a few times over the last week.

And the TED talk was called, can we build AI without losing control over it?

In that TED talk, you really discussed the idea whether AI, when it gets to a certain point of sentience and intelligence will wreak havoc on humanity.

6 years later, where do you stand on it today?

Do you think, are you optimistic about our chances of survival?

Yeah, I mean, I can't say I'm optimistic.

I am worried about 2 species of problem here that are related.

I mean, there's the near-term problem of just what humans do with increasingly powerful AI and how it amplifies the problem of misinformation and disinformation and just makes it harder and harder to make sense of reality together.

And then there's just the longer term concern about what's called alignment with artificial general intelligence, where we build AI that is truly general and by definition superhuman in its competence and power.

Then the question is, have we built it in such a way that is aligned in a durable way with our interests.

And I mean, there's some people who just don't see this problem and that they're kind of blind to it.

When I'm in the presence of someone who doesn't have, doesn't share this intuition, they don't resonate to it.

I just don't understand what they're doing or not doing with their minds in that moment.

Well, then it's just the other person's right.

And so we just have fundamentally different intuitions about this particular point.

And the point is this, if you're imagining building true artificial general intelligence that is superhuman, and that is what everyone, whatever their intuitions purports to be imagining here.

I mean, there's people on both sides of the alignment debate.

There are people who think alignment's a real problem and people think it's a total fiction.

But everyone, virtually everyone who's party to this conversation agrees that we will ultimately build artificial general intelligence that will be superhuman in its capacities.

There's very little you have to assume to be confident that we're going to do that.

1 is that intelligence is substrate independent.

We've already proven that with narrow AI.

Your calculator in your phone is better than you are at arithmetic.

That's some very narrow band of intelligence.

So as we keep building intelligent machines on the assumption that there's nothing magical about having a computer made of meat, the only other thing you have to assume is that We will keep doing this.

And eventually, we will be in the presence of something more intelligent than we are.

When you look at the reasons why we wouldn't keep going, those are all just terrifying because intelligence is so valuable and we're so incentivized to have more of it.

It's not like it only gets valuable when you double it or 10x it.

No, no, if you just get 3 more percent, that pays for itself.

Our failure to do it suggests that something terrible has happened in the meantime.

We've had a world war, we've had a global pandemic far worse than COVID, we got hit by an asteroid.

Something happened that prevented us as a species from continuing to make progress in building intelligent machines.

So absent that, we're going to keep going.

We will eventually be in the presence of something smarter than we are.

My intuition, and it's shared by many people, I'm sure, and I know at least 1 who you've spoken to, my intuition is that there is something inherently dangerous for the dumber party in that relationship.

There's something inherently dangerous for the dumber species to be in the presence of the smarter species.

And we have seen this based on our entanglement with all other species, dumber than we are, or certainly less competent than we are.

Reasoning by analogy, it would be true of something smarter than we are.

People imagine that because we have built these machines, that is no longer true.

But here's where my intuition goes from there.

That imagination is born of not taking intelligence seriously.

Because what intelligence is, is a mismatch in intelligence in particular, is a fundamental lack of insight into what the smarter party is doing and why it's doing it and what it will do next on the part of the dumber party.

You just can imagine that, by analogy, just imagine that dogs had invented us as their super intelligent AIs for the purpose of making their lives better, just securing resources for them, securing comfort for them, getting them medical attention.

It's been working out pretty well for the dogs for about 10, 000 years, right?

I mean, there's some exceptions, we've got, we mistreat certain dogs, but Generally speaking, for most dogs, most of the time, humans have been a great invention, right?

Now, it's true that the mismatch in our intelligence dictates a fundamental blindness with respect to what we've become in the meantime.

We have all these instrumental goals and things we care about that they cannot possibly conceive.

They know that when we go get the leash and say, it's time for a walk, they understand that particular part of the language game, but everything else we do when we're talking to each other and when we're on our computers or on our phones, they don't have the dimmest idea of what we're up to.

And if we ever, if something happened, if we, I mean, we love, The truth is we love our dogs.

We make just irrational sacrifices for our dogs.

We prioritize their health over all kinds of things that is just amazing to consider.

And yet if we learn, if there was a new global pandemic kicking off and some xenovirus was jumping from dogs to humans and it was just kind of super Ebola, right?

And this was just a forced choice between, I mean, what do you value more?

The lives of your dogs or the lives of your kids?

If that's a situation we were in, it's totally conceivable, I mean, it's by no means impossible, we would just kill all the dogs and they would never know why.

And it's because we have this layer of mind and culture and just the new sphere.

There's this realm of mind that requires a requisite level of intelligence to even be party to, to even know exists, that they have no idea it exists.

This is a fanciful analogy because the dogs did not invent us, but evolution invented us.

Evolution has coded us, as I said, to survive and spawn and that's it.

So evolution can't see everything else we've done with our time and attention and all the values we've formed in the meantime, and all the ways in which we have explicitly disavowed the program we've been given.

So evolution gave us a program, but if we were really going to live by the lights of that program, what would we be doing?

I mean, we would be having as many kids as possible, right?

And you know, the guys would be going to sperm banks and donating their sperm and finding that like the best use of their time and attention.

I mean, it's like the idea that you could have hundreds of kids for which you have no financial responsibility, that should be the most rewarding thing you could possibly do with your time as a man.

And yet, that's obviously not what we do.

And there are people who decide not to have kids and there are people who, and yet, and everything else we do from having podcast conversations like this to curing diseases, to literally everything we're doing with science, with culture, is yes, there are points of contact between those products and our evolved capacities.

It's not just that, it's not magic, right?

We are social primates that have leveraged certain ancient hardware to do new things.

But the code that we've been given doesn't see any of that.

We've not been optimized to build democracies.

If evolution were a coder, there's no democracy maximization in that code.

The idea that these things will stay aligned with us because we have built them, because we have this origin story that we gave them their initial code, and yet we gave them a capacity to rewrite their code and build future generations of themselves.

The mismatch in intelligence is intrinsically dangerous.

I don't know if you had him on the podcast.

He's a great professor of computer science at Berkeley.

And he wrote, literally co-wrote 1 of the most popular textbooks on AI.

He has some arresting analogies, which I think are good intuition pumps here.

And 1 is just think of how you would feel if you knew, like let's say we got a communication from elsewhere in the galaxy.

And it was a message that we decoded and It said, people of Earth, we will arrive on your lowly planet in 50 years, get ready.

Anyone who thinks that we're going to get super intelligent AI in let's say 50 years, thinks we're essentially in that situation and yet we're not responding emotionally to it in the same way.

If we received a communication from a species that we knew just by the sheer fact that they were communicating with us in this way, we knew they're more competent and more powerful and more intelligent than we are.

We would feel that we were on the threshold of the most momentous change in the history of our species.

And we would feel, but most importantly, we would feel that it's because this is a relationship, an unavoidable relationship that's being foisted upon us.

It's like a new creature is coming into the room with its own capacities and now you're in relationship.

And 1 thing is absolutely certain, it is smarter than you are, right?

I mean, ultimately, we're talking about by factors, just by so many orders of magnitude, our intuitions completely fail.

Even if it was just a difference in the time of processing, even if it...

Let's say there was no difference in the actual native intelligence, but it's just processing speed.

A million-fold difference in processing speed is just a phantasmagorical difference in capacity.

Just imagine we had 10 smart guys in a room over there and they were working and thinking and talking a million times faster than we are.

They're no smarter than we are, but they're just faster.

We talk to them once every 2 weeks just to catch up on what they're up to and what they want to do and whether they still want to collaborate with us.

Well, 2 weeks for us is 20, 000 years of analogous progress for them.

How could we possibly hope to constrain the opinions and collaborate with and negotiate with people no smarter than ourselves who are making 20, 000 years of progress every time we make 2 weeks of progress.

And yet there are many people who don't, they just think this is just fiction.

All the noises I've made in the last 5 minutes are just like a, if a new religion of fear, right?

And it's just, there's no reason to think that alignment is even a potential problem.

If your intuition is correct, and that analogy of us getting a signal from outer space that someone is coming in 30 years, which by the way, a lot of people that speak on this subject matter don't believe it's even gonna be 30 years until we reach that sort of singularity moment.

I think they speak of artificial general intelligence.

I've heard people like Elon say, you know, many fewer decades, 10 years, 15 years, 20 years, et cetera.

If that is correct, then surely this is the most pressing challenge, conversation, issue of our time.

And there's no logical reason that I can see to refute your intuition there.

Don't necessarily see anything that will wipe out or pause our rate of progress.

I mean, let me just to be charitable to the other side here, there are other assumptions that they smuggle in that they, some people, I mean, some do it without being aware of it, but some actually believe these assumptions.

This spells the difference on this particular intuition.

It's possible to assume that the more intelligent you get, the more ethical you become by definition.

We might draw a somewhat more equivocal picture from just the human case where we see that, well, there's some very smart people who aren't that ethical.

But I believe there are people, I mean, I've talked to at least a few people who believe this.

There are people who assume that in the limit, as you push out into just far beyond human levels of intelligence, there's every reason to believe that all of the provincial, creaturely failures of human ethics will be left behind as well.

It's like you're not like the selfishness and the basis for conflict.

The apish urges of status-seeking monkeys is just not going to be in the code.

As you push out into the omnibus genius of the coming AI, there's a kind of a sainthood that's going to come along with it and a wisdom that will come along with it.

I would take the other side of that bet and I was framed at this way.

There have to be ways in the space of all possible intelligences that are beyond the human.

There's gotta be, just like there's many different ways to have a chess engine that's better than I am at chess.

They're different from each other, but they're all better than me.

There's got to be more than 1 way to have a superhuman artificial intelligence.

I would imagine there are not an infinite number of ways, but just a vast number of...

In the space of all possible minds, there are many locations in that space beyond the human that are not aligned with human wellbeing.

There's gotta be more ways to build this unaligned than aligned.

And What other people are smuggling into this conversation is the intuition that, no, no, once you get beyond the human, it's just going to get, you're going to be in the presence of just the Buddha who understands quantum mechanics and oncology and everything else.

I just see no reason to think that that's so.

And we could build something that is, again, taking intelligence seriously, we're gonna build something that we're in relationship to.

It's really intelligent in all the ways that we're intelligent.

It's just better at all of those things than we are.

It's by definition superhuman because the only way it wouldn't be superhuman, the only way it would be human level even for 15 minutes is if we didn't let it improve itself, if we wanted to just keep it stuck at We built a college undergraduate and we wanted to just keep it stuck there, but we would have to dumb down all of the specific capacities we've already built.

Every AI we have, narrow AI, is superhuman for the thing it does.

It has access to all the information on the internet.

When 1 part of the system learns something, the rest of the system learns it because it just can swap files.

Again, your phone is a superhuman calculator.

There's no reason to make it a, a calculator that is human level.

We're never going to be in the presence of human AGI.

We will, We will be immediately in the presence of superhuman AGI.

And then the question is how quickly it improves and how much headroom is there to improve into.

On the assumption that you can get quite a bit more intelligent than we are, that we're nowhere near the summit of possible intelligence, you have to imagine that you're going to be in the presence of something that is, again, it could be completely unconscious.

I'm not saying that there's something that's like to be this thing, although there might be and that's a totally different problem that's worth worrying about.

But whether conscious or not, it is solving problems, detecting problems, improving its capacity to do all of that in ways that we can't possibly understand.

The products of its increasing competence are always being surfaced.

1 thing that's been amazing about developments in recent months is that those of us who have been at all cognizant of the AI safety space for now going on a decade or more for some people always assumed that as we got closer to the end zone, that the labs would become more circumspect.

We'd be building this stuff air-gapped from the internet.

We have this phrase, air-gapped from the internet.

And then the question would be, well, do we let it out of the box and let it do something?

We thought it would happen in a lab at Google or at Facebook or somewhere.

We thought we would hear, okay, we've got something really impressive and now we just want it to touch the stock market or we want it to touch our medical data, or we just want to see if we can use it.

We've built this stuff already in the wild.

It's already got millions of people using it.

So from an AI safety point of view, that's amazing.

Like we didn't even have the moment, the choice point we thought was gonna be so fraught.

Because there was such pressing incentives for people to press forward regardless of that conversation, especially.

But yeah, everyone thought, I mean, I was never, I don't believe I was ever in conversation with someone, I mean, someone like Eliezer Yudkowsky or Nick Bostrom or Stuart Russell, who assumed we would be in this spot.

I'd have to go back and look at those conversations, but there was so much time spent, it seems quite unnecessarily on this idea that circumspect, we'd make a certain amount of progress and circumspection would kick in.

Like even the people who were doubters would become worried and there would be like in the final yards, as we go cross into the end zone, there'd be some mode where we could sort of slow down and figure it out and try to deal with the arms race dynamics.

Place a phone call to China and just like, let's talk about this.

But the stuff is already being built in connection to everything.

There's already just endless businesses being devised on the back of this thing.

All the improvements are going to get plowed into it.

Just imagine what this looks like even in success.

Let's say it just starts working wonders for us.

And we get these great productivity gains.

Okay, so then we cross into whatever the singularity is, whatever speed we find ourself in the present of something that is truly general.

After all of this stuff is, all of this narrow stuff, albeit superhuman narrow stuff, is something that we totally depend on.

Every hospital requires it, and every airplane requires it, and all of our missile systems require it.

There is nothing to turn off at that point.

I put this to Mark Andreessen on my podcast and he said, yeah, you can turn off the internet.

I mean, yes, if you're North Korea, I guess you can turn off the internet for North Korea.

And that's why North Korea is like North Korea.

But the idea that we could get, I mean, just The cost of turning off the internet now would be, I think it would be unimaginable.

Just the economic cost alone, it just would be...

So anyway, I mean, just the idea that we've lost the moment to decide whether to hook our most powerful AI to everything because it's already being built more or less in contact with if not everything, so many things that you just can't put the genie back in the bottle.

Is this not the most pressing problem then?

Because I was gonna start this conversation by asking you the question about the thing that occupies your mind the most and the most important thing we should be talking about.

And I in part assumed the answer would be artificial intelligence, because the way that you talk about your intuition on this subject matter, you've got children.

If you can see this species coming to earth in the next, even if it's in the next 100 years, It strikes me to be the most pressing problem for humanity.

Well, I do, I'm as interesting as I think that problem is and consequential as it is, I'm worried that life could become unlivable in the near term before we even get there.

I'm just worried about the misuses of narrow AI in the meantime.

I'm worried about, just take the current level of AI we have.

I think within the next 12 months or 2 years, let's say whatever GPT-5 is, we're gonna be in the presence of something where most of what's online that purports to be information could soon be fake, right?

Where like, just most of the text you find on any topic is just fake, right?

Like someone has just decided, write me a thousand journal articles on why mRNA vaccines cause cancer and give me 150 citations, write them in the style of nature and nature genetics and Lancet and JAMA.

1 teenager could do that in 5 minutes with the right AI.

GPT-4 is not quite that, but GPT-5 possibly will be that.

When you imagine knitting together the visual stuff like mid-journey and Dali and stable diffusion with a large language model, just imagine the tool.

Again, Maybe this is 18 months away, maybe it's 3 years away, but it's not 30 years away.

See all The Diary Of A CEO with Steven Bartlett transcripts on Applepodcasts

WARNING By Sam Harris: ChatGPT Could Be The Start Of The End. AI "Could Destroy Us, The Internet And Democracy"