Ray Kurzweil is an author, inventor, and futurist. Please support this podcast by checking out our sponsors:
- Shopify: https://shopify.com/lex to get 14-day free trial
- NetSuite: http://netsuite.com/lex to get free product tour
- Linode: https://linode.com/lex to get $100 free credit
- MasterClass: https://masterclass.com/lex to get 15% off
- Indeed: https://indeed.com/lex to get $75 credit

EPISODE LINKS:
Ray's Website: https://kurzweilai.net
Ray's Books:
The Singularity Is Nearer (pre-order): https://amzn.to/3BNXmGR
How To Create A Mind: https://amzn.to/3qqlkBw
The Singularity Is Near: https://amzn.to/3DfXP5z
The Age of Spiritual Machines: https://amzn.to/3RSjtAX
Danielle: https://amzn.to/3Bww2N7
Transcend: https://amzn.to/3RAEYGV

PODCAST INFO:
Podcast website: https://lexfridman.com/podcast
Apple Podcasts: https://apple.co/2lwqZIr
Spotify: https://spoti.fi/2nEwCF8
RSS: https://lexfridman.com/feed/podcast/
Full episodes playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4
Clips playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOeciFP3CBCIEElOJeitOr41

OUTLINE:
0:00 - Introduction
1:06 - Turing test
14:51 - Brain–computer interfaces
26:31 - Singularity
32:51 - Virtual reality
35:31 - Evolution of information processing
41:57 - Automation
51:57 - Nanotechnology
53:51 - Nuclear war
55:57 - Uploading minds
1:03:38 - How to think
1:10:08 - Digital afterlife
1:19:28 - Intelligent alien life
1:22:18 - Simulation hypothesis
1:26:31 - Mortality
1:34:10 - Meaning of life

SOCIAL:
- Twitter: https://twitter.com/lexfridman
- LinkedIn: https://www.linkedin.com/in/lexfridman
- Facebook: https://www.facebook.com/lexfridman
- Instagram: https://www.instagram.com/lexfridman
- Medium: https://medium.com/@lexfridman
- Reddit: https://reddit.com/r/lexfridman
- Support on Patreon: https://www.patreon.com/lexfridman

By the time we get to 2045, we'll be able to multiply our intelligence many millions fold, and it's just very hard to imagine what that will be like.

The following is a conversation with Ray Kurzweil, author, inventor, and futurist, who has an optimistic view of our future as a human civilization, predicting that exponentially improving technologies will take us to a point of a singularity beyond which superintelligent artificial intelligence will transform our world in nearly unimaginable ways.

years ago, in the book Singularity is Near, he predicted that the onset of the singularity will happen in the year

He still holds to this prediction and estimate.

In fact, he's working on a new book on this topic that will hopefully be out next year.

To support it, please check out our sponsors in the description.

And now, dear friends, here's Ray Kurzweil.

In your 2005 book titled The Singularity's Near, you predicted that the singularity will happen in

So now, 18 years later, do you still estimate that the singularity will happen

And maybe first, what is the singularity, the technological singularity, and when will it happen?

Singularity is where computers really change our view of what's important and change who we are.

But we're getting close to some salient things that will change who we are.

A key thing is 2029, when computers will pass the Turing test.

And there's also some controversy whether the Turing test is valid.

Most people do believe that, but there's some controversy about that.

But Stanford got very alarmed at my prediction about 2029.

So they held an international conference, you might have been aware of it, of AI experts in 1999 to assess this view.

So people gave different predictions and they took a poll.

It was really the first time that AI experts worldwide were polled on this prediction.

80% believed it would happen, but not within their lifetimes.

There's been so many advances in AI that the poll of AI experts has come down over the years.

So a year ago, something called Meticulous, which you may be aware of, assessed different types of experts on the future.

They again assessed what AI experts then felt, and they were saying 2042.

A few weeks ago, they again did another poll, and it was 2030.

So, A-experts now basically agree with me.

I haven't changed at all, I've stayed with 2029.

And A-experts now agree with me, but they didn't agree at first.

So Alan Turing formulated the Turing test and.

Right, now what he said was very little about it.

I mean the 1950 paper where he had articulated the Turing test, there's like a few lines that talk about the Turing test.

And it really wasn't very clear how to administer it.

And he said if they did it in like 15 minutes, that would be sufficient, which I don't really think is the case.

These large language models now, some people are convinced by it already.

I mean, you can talk to it and have a conversation with it, and you can actually talk to it for hours.

There's some problems with large language models, which we can talk about.

But some people are convinced by the Turing test.

Now, if somebody passes the Turing test, what are the implications of that?

Does that mean that they're sentient, that they're conscious or not?

It's not necessarily clear what the implications are.

Anyway, I believe 2029, that's 6, 7 years from now, we'll have something that passes the Turing test and a valid Turing test, meaning it goes for hours, not just a few minutes.

What is your formulation of the Turing test?

You've proposed a very difficult version of the Turing test, so what does that look like?

Basically, it's just to assess it over several hours, and also have a human judge that's fairly sophisticated on what computers can do and can't do.

If you take somebody who's not that sophisticated, or even a average engineer, they may not really assess various aspects of it.

So you really want the human to challenge the system.

On its ability to do things like common sense reasoning, perhaps.

That's actually a key problem with large language models.

They don't do these kinds of tests that would involve assessing chains of reasoning.

If you talk to them, they actually can talk to you pretty well, and you can be convinced by it.

But it's somebody that would really convince you that it's a human, whatever that takes.

Maybe it would take days or weeks, but it would really convince you that it's human.

Large language models can appear that way.

You can read conversations and they appear pretty good.

You can ask, well, how many legs did 10 elephants have?

And they'll tell you, well, okay, each elephant has 4 legs and it's 10 elephants, so it's 40 legs, and you go, okay, that's pretty good.

And they don't seem to understand the question.

I mean, how advanced a human do you want it to be?

But we do expect a human to be able to do multi-chain reasoning, to be able to take a few facts and put them together.

And we see that in a lot of polls, that people don't do that perfectly at all.

So it's not very well-defined, but it's something where it really would convince you that it's a human.

Is your intuition that large language models will not be solely the kind of system that passes the Turing test in 2029?

No, I think it will be a large language model, but they have to go beyond what they're doing now.

And another key issue is if somebody actually passes the Turing test validly, I would believe they're conscious.

So okay, we can pass the Turing test, but we don't really believe that it's conscious.

But if it really passes the Turing test, I would believe that it's conscious.

But I don't believe that of large language models today.

If it appears to be conscious, that's as good as being conscious, at least for you, in some sense.

I mean, consciousness is not something that's scientific.

I mean, I believe you're conscious, but it's really just a belief, and we believe that about other humans that at least appear to be conscious.

When you go outside of shared human assumption, like are animals conscious?

Some people believe they're not conscious.

And would a machine that acts just like a human be conscious?

I mean, I believe it would be, but that's really a philosophical belief.

I can't take an entity and prove that it's conscious.

There's nothing that you can do that would be, that would indicate that.

You can say it, multiple people can experience a piece of art as beautiful, but you can't prove it.

But it's also an extremely important issue.

I mean, imagine if you had something where nobody's conscious, the world may as well not exist.

And so some people like say Marvin Rinsky said, well, consciousness is not logical, it's not scientific, and therefore we should dismiss it, and any talk about consciousness is just not to be believed.

But when he actually engaged with somebody who was conscious, he actually acted as if they were conscious.

He acted as if consciousness does matter.

But that's true of a lot of people as well.

But I would say it's not a scientific issue.

Some people believe that anything that makes a decision is conscious.

Its level of consciousness is low, not very interesting, but that's a consciousness.

So a computer that makes a more interesting decision still not at human levels, but it's also conscious and at a higher level than your light switch.

There's many different views of what consciousness is.

So if a system passes the Turing test, it's not scientific, but in issues of philosophy, things like ethics start to enter the picture.

Do you think there would be, we would start contending as a human species about the ethics of turning off such a machine?

But I'm talking about 2029, that's not that many years from now.

It has a different, I mean a computer that's conscious has a little bit different connotations than a human.

We're in an entity that does not last forever.

Now, actually, a significant portion of humans still exist and are therefore still conscious.

But anybody who is over a certain age doesn't exist anymore.

That wouldn't be true of a computer program.

You could completely turn it off and a copy of it could be stored and you could recreate it.

And so it has a different type of validity.

You could eliminate its memory and have it go over again.

I mean, it has a different kind of connotation than humans do.

Well, perhaps it can do the same thing with humans.

It's just that we don't know how to do that yet.

It's possible that we figure out all of these things on the machine first.

But that doesn't mean the machine isn't conscious.

I mean, if you look at the way people react, say 3CPO or other machines that are conscious in movies, they don't actually present how it's conscious, but we see that they are a machine, And people will believe that they are conscious.

And they'll actually worry about it if they get into trouble and so on.

So 2029 is going to be the first year when a major thing happens.

And that will shake our civilization to start to consider the role of AI in this world?

I mean, there's this 1 guy at Google who claimed that the machine was conscious.

because most people have not taken that position.

I mean, I've used different things like this and they don't appear to me to be conscious.

As we eliminate various problems of these large language models, more and more people will accept that they're conscious.

So when we get to 2029, I think a large fraction of people will believe that they're conscious.

I believe it will actually happen gradually, and it's already started to happen.

And so that takes us 1 step closer to the singularity.

Another step then is in the 2030s, when we can

actually connect our neocortex, which is where we do our thinking, to computers.

And I mean, just as this actually gains a lot to being connected to computers that will amplify its abilities.

I mean, if this did not have any connection, it would be pretty stupid.

It could not answer any of your questions.

If you're just listening to this, by the way, Ray's holding up the all-powerful smartphone.

So We're gonna do that directly from our brains.

These already have amplified our intelligence.

I'm already much smarter than I would otherwise be if I didn't have this.

Because I remember my first book, The Age of Intelligent machines, there was no way to get information from computers.

I actually would go to a library, find a book, find the page that had an information I wanted, and I'd go to the copier.

And my most significant information tool was a roll of quarters where I could feed the copier.

So we're already greatly advanced that we have these things.

First of all, I constantly put it down and I don't remember where I put it.

I've actually never lost it, but you have to find it and then you have to turn it on.

It would actually be quite useful if someone would just listen to your conversation and say, oh, that's so-and-so actress and tell you what you're talking about.

So going from active to passive where it just permeates your whole life.

The way your brain does when you're awake.

Now, that's something that could actually just about be done today, where we'd listen to your conversation, understand what you're saying, understand what you're not missing, and give you that information.

But another step is to actually go inside your brain.

And there are some prototypes where you can connect your brain.

They actually don't have the amount of bandwidth that we need.

They can work, but they work fairly slowly.

So if it actually would connect to your neocortex, and the neocortex, which I described in How to Create a Mind, The neocortex is actually, it has different levels.

And as you go up the levels, it's kind of like a pyramid.

And that's the level where you wanna connect these brain extenders.

So I believe that will happen in the 2030s.

So just the way this is greatly amplified by being connected to the cloud.

We can connect our own brain to the cloud and just do what we can do by using this machine.

Do you think it would look like the brain-computer interface of Neuralink?

It doesn't have the bandwidth that we need.

Right, but I think, I mean, they're gonna get permission for this because there are a lot of people who absolutely need it because they can't communicate.

I know a couple of people like that who have ideas and they cannot move their muscles and so on, they can't communicate.

So for them, this would be very valuable, But we could all use it.

Basically it'd be, turn this into something that would be like we have a phone, but it would be in our minds, it would be kind of instantaneous.

And maybe communication between 2 people would not require this low bandwidth mechanism of language.

We don't know what that would be, although we do know that computers can share information like language instantly.

They can share many, many books in a second, so we could do that as well.

If you look at what our brain does, it actually can manipulate different parameters.

So we talk about these large language models.

I mean, I had written that it requires a certain amount of information in order to be effective, and that we would not see AI really being effective until it got to that level.

And we had large language models that were like 10 billion bytes, didn't work very well.

They finally got to a hundred billion bytes and now they work fairly well.

If you say lambda has a hundred billion bytes, what does that mean?

Well, what if you had something that had 1 byte, 1 parameter?

Maybe you want to tell whether or not something's an elephant or not.

And so you put in something that would detect its trunk.

If it doesn't have a trunk, it's not an elephant.

Really wouldn't be able to tell what a trunk is, but anyway.

And maybe other things other than elephants have trunks.

I'm not sure which animals have trunks, but you know, That's how you define a trunk.

So these things have 100 billion parameters.

So they're able to deal with very complex issues.

Human beings actually have a little bit more than that, but they're getting to the point where they can emulate humans.

If we were able to connect this to our neocortex, we would basically add more of these abilities to make distinctions and it could ultimately be much smarter and also be attached to information that we feel is reliable.

So you think that there will be a merger in the 30s, an increasing amount of merging between the human brain and the AI brain?

And the AI brain is really an emulation of human beings.

Because human beings act the same way, and this is basically to amplify them.

It's a little bit clumsy to interact with, but it definitely, it's way beyond what we had 15 years ago.

But the implementation becomes different, just like a bird versus the airplane.

Even though the AI brain is an emulation, it starts adding features we might not otherwise have, like ability to consume a huge amount of information quickly.

Like look up thousands of Wikipedia articles in 1 take.

And we can get, for example, to issues like simulated biology, where it can simulate many different things at once.

We already had 1 example of simulated biology, which is the Moderna vaccine.

And that's gonna be now the way in which we create medications.

But they were able to simulate what each example of an mRNA would do to a human being, and they were able to simulate that quite reliably.

And we actually simulated billions of different mRNA sequences.

And they found the ones that were the best and they created the vaccine.

And they did, and talked about doing that quickly, They did that in 2 days.

Now, how long would a human being take to simulate billions of different mRNA sequences?

I don't know that we could do it at all, but it would take many years.

1 of the reasons that people didn't like vaccines is because it was done too quickly.

And they actually included the time it took to test it out, which is 10 months, so they figured, okay, it took 10 months to create this.

And we also will be able to ultimately do the tests in a few days as well.

Well, because we can simulate how the body will respond to

See all Lex Fridman transcripts on Youtube

Ray Kurzweil: Singularity, Superintelligence, and Immortality | Lex Fridman Podcast #321