Cosmic Queries – Living in a Simulation with Nick Bostrom

Welcome to StarTalk, your place in the universe where science and pop culture collide.

StarTalk begins right now.

This is StarTalk, Cosmic Queries edition.

Neil deGrasse Tyson here, your personal astrophysicist.

I got Chuck Nice with me, of course.

Chuck, my…

What’s up, Neil?…

faithful co-host.

You know, we need you for the Cosmic Queries, so that you can mispronounce everyone’s name.

Well, that’s my purpose in life, Neil.

I live to butcher names.

You know?

Those poor questioners.

How would you attack my name?

Oh, my goodness.

So, Nick Bostrom, is that what you…

Is that how…

Is that all right?

In Swedish, it would be Niklas Bostrom, but that was close.

And listen, I’ll take close.

As far as I’m concerned, names are like a game.

That’s like a game of horseshoes for me.

Close is good enough.

Good enough.

So that was indeed Nick Bostrom chiming in.

Nick, welcome to StarTalk.

Dude, you started something that has got the whole world spinning in a tizzy for birthing the concern that we all live in a simulation.

And let me just give a fast bio on you.

You’re a professor at University of Oxford in the Future of Humanity Institute.

That doesn’t look very right.

Sorry, Nicholas, not a lot of job security in that, buddy.

Looking at the future of humanity, yo.

So you think about artificial intelligence, the ethics of artificial intelligence, biosecurity, what have you, macro strategy.

We’ll ask you what that is in a moment.

Just policy, ethics, foundational questions about serious challenges that civilization faces, not in the distant future, but in the very near future.

I like the fact that you have a background in theoretical physics.

So put you in the physics club here, that’s good.

Also computational neuroscience.

We have some of those at my home institution at the American Museum of Natural History.

That’s quite the frontier as well.

And you had a rather influential paper, research paper, titled, Are You Living in a Computer Simulation?

And for me also I remembered your book Super Intelligence, which all of these got people thinking, as any good philosopher should do, is to get people thinking.

And so could you just start us off?

Why do you think we might be living in a simulation?

Well, I have this thing called the simulation argument, which doesn’t actually prove that we’re in a simulation, but it tries to show that at least one of three propositions is true.

So we want to hear your line of reasoning, which ought to be good given your sort of logical background in this universe.

So let’s hear what you’ve got.

Well, I mean, you probably would be able to explain it better, but yeah, my story is that the simulation argument tries to show that one of three propositions is true.

So let’s first look at what the conclusion is and then we can see how we get there.

So the conclusion is that either almost all civilizations at our current stage of technological development go extinct before they become technologically mature.

So that’s like one alternative, right?

The second is that amongst civilizations that do become technologically mature, there is a very strong convergence.

They all lose interest in creating a certain kind of computer simulation.

I call them ancestor simulations.

These would be detailed simulations of people with the kind of experiences that their historical forebears had.

So that’s the second alternative.

And then the third alternative is that we are almost certainly living in a computer simulation.

So that’s the kind of conclusion.

Now, how does one get to that?

Well, suppose that the first of these alternatives does not obtain.

So that means it’s not true that almost all civilizations at our stage fail to reach technological maturity.

Some non-trivial fraction make it true.

Then let’s suppose that the second alternative is also false.

So amongst those who do become technologically mature, some non-trivial fraction remain interested in using some of the resources to create these kinds of ancestor simulations.

Then you can show that the kind of computational resources a mature civilization would have, would suffice to create millions and billions of detailed simulations, ancestor simulations, runs of human history.

And so that if the first two alternatives are false, then there would be many, many more simulated versions of people with our kinds of experiences.

Then that would be original, implemented in basic physical reality, people with our experiences.

And conditional on that, if almost all people with our experiences are simulated, we should think we are probably one of the simulated ones, rather than one of the rare non-simulated ones.

So that means that if you reject the first two alternatives, you would then have to accept the third one.

And then that shows that it’s not the case that all three of them are false.

So hence, at least one of them is true.

So that’s the structure.

Why can’t there be a fourth other truth that no one gives a rat’s ass about simulating anything, anywhere?

Well, so that’s the second, right?

I mean, so if all of these technologically mature civilizations are completely uninterested in simulating, then that would be possibility number two.

But note that for the second alternative to hold, it’s not sufficient that most of them are not very interested.

Because even if it were just a 1% of these mature civilizations that were even a little bit interested, they still could produce millions of them.

And so that would have to be this extremely strong convergence.

Almost all of them would completely have to lose any interest in doing this in order for the second alternative to be the…

Okay, so I have publicly, mildly, butchered your line of argument there.

So let me first apologize.

What I had been noting is that we do not have the power yet to create a perfect simulation of a world, such as the one we’re living in.

And so I wasn’t thinking that everyone would make these ancestor simulations, which we would be.

And so an ancestor civilization in a cinematic parallel would be a movie about Spartacus or Cleopatra.

Just something, a movie, which is modern technology, telling a story set in a time when they didn’t have movies, right?

So that would be like an answer.

I’m guessing that’s what you mean by an ancestor simulation.

And so I was thinking that every simulation would ultimately be able to duplicate themselves as a natural evolutionary arc.

If that’s the case, then we would either be the original universe that hasn’t yet simulated anybody yet, or we’d be sort of the last one simulated still working our way towards the power of simulating ourselves, which would be slightly better odds, well, a lot better odds, than throwing a dart and landing in all the simulations that had enough power to create simulations of themselves.

Yeah, well, I…

Well, so, first of all, I don’t claim that the only simulations that might be made are ancestor simulations.

I mean, if you imagine you are a technologically mature civilization, you might simulate all kinds of things, like real histories as close as you can get, fantasy worlds, counterfactual histories, imaginary alien civilizations.

I mean, you could…

Maybe there are like lots of all of these kinds of simulations.

The argument focuses on ancestor simulations just because that’s the easiest way to get to the conclusion that one of these three is true.

But it doesn’t imply that there wouldn’t be a lot of other simulations as well.

Okay, so we have some data with our own history of cinema.

And some very small percent of movies are set in a time before movies were invented.

Which again, I would classify as sort of ancestor storytelling.

Yeah, so I mean, I guess when we extrapolate to these technologically mature presumably post-human civilizations, well, first of all, I’m not sure how much we can infer from the kinds of movies we create to what types of simulations they would run.

But let’s suppose for the sake of the argument that the majority of simulations they run are of people in their contemporary society.

So I don’t know, some super advanced space colonizing thing with super intelligences or whatnot.

And that that’s maybe the majority of what they do, but that they assign some smaller fraction of their computational resources to doing these ancestor simulations.

Let’s assume that I still don’t think that would defeat the simulation argument or indeed even the alternative that we are in a simulation.

Because we kind of already know that we are not one of the post-humans.

I mean, you just look around, you don’t see a lot of starships whizzing by outside your window, and we are not currently running any simulations ourselves.

So we can kind of cross those out, like all the actual post-humans, we know we’re not one of those, and we also know we are not in a simulation of a post-human.

That’s not the world we experience.

Then that leaves only, A, the people in original history at the human level of development, and also whatever ancestor simulations are at that level of development.

And so my claim would then be that if the first two alternatives of the simulation argument are false, the simulated ones at our current level of development would still vastly outnumber the original ones at our stage of development.

So what is the likelihood that, not likelihood because that’s the wrong word, is it possible that it could just be the way that we create with our limited technology what we feel are simulations of our lives, okay?

And that’s computer games and video games and things like that.

Could it be that a civilization so advanced that they have the computational power to create all of this just for the hell of it?

Just because, like the same way we do it.

We do it for entertainment.

Could it just be that?

Or is that just not a part of the philosophy?

It could be.

I mean, so the simulation argument itself is agnostic as to what the motivation would be of the simulators.

And you could indeed imagine many possible motivations.

One would be just entertainment.

And you could imagine other, like maybe some kind of research, like historically.

Maybe it would be interesting to explore counterfactuals of history.

Or you could imagine art projects.

Or you could imagine moral reasons for…

I think we know rather little about the psychology and motivations of these hypothetical post-human civilizations and why they would make simulations.

Okay, so Nick, I guess you’re allowed to say all this like from your armchair.

But at some point somebody wants to walk into a lab and make a measurement that says, here’s the evidence that supports Nick’s argument.

Is there such a…

anything we can look for?

Is there a sign?

Is there some experiment we can conduct and say, yep, we’re not in charge of what’s happening here.

This is a simulation.

There certainly is sort of empirical premises that flow into the simulation argument.

And so evidence for or against the truth of those assumptions, you know, would be relevant to evaluating the argument.

So one empirical premise is that a technologically mature civilization would indeed have the capability of creating ancestor simulations, and indeed to create lots of them.

And so the kinds of evidence that would be relevant for that is evidence, say, of the kinds of computational performance you could get from physically possible systems.

We’re not able to build them currently, but we can kind of do first principle modeling of different computational systems based on nanotechnology and so forth, and we can place lower bounds on the kind of compute power that they would unlock.

So that would be one element that would flow into this.

Another would be some estimate of the computational cost of running an ancestor simulation.

I think the largest part of that cost is the cost of simulating human brains at the sufficient level of detail that the simulation would be conscious.

And we can obviously not precisely determine what the computational expense of simulating a human brain is, but we can place some upper bound on that.

We have various views about what computational tasks the human brain is capable of performing.

We know how many neurons there are, how many synapses, how often they fire.

We can roughly estimate that.

Now it turns out that if you estimate the amount of compute power available, even if you make rather conservative assumptions about that, and then you make conservative assumptions about how much it takes to simulate one human brain and therefore how much to simulate all of the human brains, you just multiply that by 100 billion or something to all of the human brains in history, there are a number of orders of magnitude gap between these two.

So even if you are off a little bit in these estimates, it still seems like the argument holds.

So those would be empirical premises that we could theoretically obtain evidence against.

Like if we discovered the human brain uses some kind of weird quantum computation that is a lot more expensive, then that would flow into it.

Then if in addition you want to conclude not just that one of these three alternatives is true, which is all the simulation argument itself says, but if more specifically you want to conclude that we are in the simulation, that the third alternative is true, then there is an additional range of empirical questions that become relevant.

Like anything that gives you evidence against the first two, or in favor of the first two, would be then relevant evidence for evaluating the third.

So if we discover that there is some kind of big risk, some doomsday mechanism that we can, ah, now we realize this, all sufficiently advanced civilizations will stumble on this new technology and destroy themselves, that would be argument against the simulation hypothesis, because it would make the first alternative more likely.

So that, I think, is actually the main kind of evidence.

That would be a really sad argument.

That would be a really sad argument against it, because it would say, here’s our proof we’re not simulated.

We’re about to destroy ourselves.

We’re going to blow everything up.

Now, Nick, let me ask you this.

Is it possible that you are so smart that you are constantly high and you don’t know it?

I think in some more or less metaphorical sense, I think that’s very likely to be true.

It’s kind of the pessimistic meta-induction.

If you look at all humans who have ever been alive, all eras going back in time, we can now see from our current vantage point, basically they were all very wrong about some big thing.

Starting with simple physics, they thought Earth was in the center.

Basically, we can see, if we look back more than 100 years, we see that they all got a whole bunch of really core things wrong.

And it would kind of maybe be a little bit presumptuous to think that now finally we’ve gotten all of these basic things right.

It seems more likely that if people a thousand years from now look back at 2021, they will probably also see big, not just gaps in our understanding, but like things we were fundamentally confused about.

Yeah, they’ll laugh their ass off at everything we’re talking about right now.

Yeah, or cry their heads out or whatever.

So I do think we are in a fundamental sense very much in the dark about the really biggest picture.

Well, we got to take a quick break.

When we come back, we’ll transition to our questions that we’ve solicited from our Patreon members.

So we’ll be right back on StarTalk.

Hey, I’m Roy Hill Percival, and I support StarTalk on Patreon.

Bringing the universe down to earth, this is StarTalk with Neil deGrasse Tyson.

We’re back, StarTalk.

I’ve got the one and the only, Nick Bostrom.

I generally regard it as the person who birthed all of our sleepless nights, wondering whether we are in a simulation.

Not only from his research papers, but from his books on the subject.

And he’s a professor at the University of Oxford.

And Chuck, what did we call it?

What is the name of his thing here?

The Future of Humanity Institute, which Chuck and I are pretty sure we’ll be very short on, because there’s no future for humanity at all.

Just before we get to the questions that Chuck has collected, Nick, if it’s one thing to simulate all of the brains, I get that, but it’s another thing, the fact that I can go into a garden and then look at a flower or dig through the soils and keep digging and reach the mantle of the earth.

Whoever is simulating us has to simulate not only what my brain is doing, but it has to simulate all the things my brain is experiencing.

And that’s not just for me.

Someone else could dig that same hole and they should be finding the same thing.

So isn’t the total complexity of the world, doesn’t that have to be part of this simulation?

Even the fact that I as an astrophysicist look out to the edge of the universe, decoding the nature of the Big Bang and all time and space that followed it.

So why just limit your estimates to the power of the human brain if everything and the unfolding of the great cosmic story has to also happen alongside it?

Yeah, I think you do need some computation assigned to simulating relevant parts of the environment.

I think the biggest part will be the brains, but certainly if you had to simulate all of the environment at subatomic detail continuously.

I mean, like quantum simulation of the entire universe would be completely infeasible if the simulators have anything comparable to the compute power that we could realize in this universe.

You know what?

I’m going to disagree.

I’m sorry, I know you’re a genius, but here’s the deal.

Here’s why I’m going to disagree, Nick.

Because when movie makers make movies, they do not render the detail in every single little thing.

What they have…

He didn’t get there yet.

See, you already thought of this.

Like I said.

Jesus Christ.

Here I am making a discovery, man.

Chuck, okay, continue.

Chuck, finish the point.

Both of you already knew where I was going.

But the deal is this.

If you actually create a background, that background will pretty much be the same for all the characters that are mapped onto that background.

So that’s all I was saying.

I think that’s the key to understand this whole simulation argument stuff.

If you had to simulate all of the environment, the subatomic detail continuously, it probably would be completely infeasible to do that.

But I claim that’s not needed.

All you would need to do is to simulate enough of the parts that we are observing when we are observing them, that to the simulated creatures, it looks real, that they can’t tell the difference.

And that’s a lot less.

All right, wait a minute.

I just thought of something else in support.

So what that would mean, Chuck.

Chuck, check what that would mean.

Whole sections of the Pacific Ocean where there isn’t a boat, right?

Then no one has…

So it doesn’t exist until someone has to then see it and process it.

So it’s a procedural content generation.

So we use it in our computer games today a lot.

Like you often only render the parts that some character in the game are observing.

And maybe you have some very coarse-grained simulation of the whole thing continuously, but you may fill in details if and when it’s needed.

So if, like right now, I don’t have any idea what the atoms in this desk in front of me are doing, but if I took, in principle, an electron microscope or something, I could look and I’d better see atoms there.

The programmer would know you’re about to bring out an electron microscope.

It’s time to up the calculation right in the beam.

And if necessary, I mean, they could even pause the simulation or edit it or erase memories if they really screwed it up.

I think the kind of capability you would need to even create anything resembling this kind of simulation is very advanced.

And I think with that advanced capability would also come the ability to edit and to monitor human thoughts and intentions and then kind of be able to do this kind of procedural generation that even we do in our computer games today.

That could explain why I’ve heard Neil say this, that we are terrible data takers.

Like as human beings, we are awful at taking in information.

Well, if I’m programming a simulation, I would certainly want to program the people in that simulation to be like that because that way I wouldn’t have to program all this detail into stuff.

It protects the integrity of my simulation.

Yeah, although I think, to be fair, I think the difference between one human and another from the point of view of the simulators, it’s like, well, there is one ant.

It’s got a few more neurons.

It’s a genius ant.

But I mean, we’re all like ants, I think.

So I don’t think the difference in cost is that big.

Very cool.

All right, Chuck, bring on a question.

Let’s see what you got.

Let’s jump into this.

This is Dennis Gislain.

And Dennis says this.

I spell that?

G-H-I-S-L-A-I-N.

I say Gislain.

It’s probably Gislain.

Gislain.

Okay.

Denis Gislain.

Denis Gislain.

We’d like to know.

He says in his papers, Dr.

Bostrom talks about post-human stage civilization.

Could you please develop on that and situate it in, listen, a Kardashev scale?

Now, I don’t know what any of that means.

Okay.

I can tell you what the Kardashev scale is.

What is the Kardashev scale?

Yeah.

When I lead off with that, and I’ll hand the baton over to you, Nick.

So the Kardashev scale is a scale of how much energy you have access to and can exploit.

So I think there are five levels.

So one of them is, do you have access to all of the energy sources in your host planet?

And if you do, and you can exploit them, you’re at civilization level one.

So that means you can go into a volcano and tap the energy.

You can tap the volcano the way you tap a cake.

You could use the energy in the crust of the earth that would otherwise make earthquakes.

You can tap that and use that for your own means.

Storm systems, this sort of thing.

So a level two civilization would control all of the energy that comes from its host star.

Okay.

That’s way more energy than what is embedded in your planet.

A level three civilization would control all the energy of your galaxy that you happen to live in.

The massive black hole at the center of the galaxy.

You could use that.

Yeah, exactly.

And you wield this.

And the history of civilization reveals that the nations or the nation states that had the most power, power, political power, cultural power, were those that actually wielded the most energy per capita in the world at that time.

So when people say, the United States, we’re such energy hogs.

We use four times, five times the energy as anybody else.

Well, that correlates with other measures of power that exist in it.

So let’s keep going.

So one more level four, level five.

If you control all the energy of the universe, and then you’re indistinguishable from a god that anyone would have suggested.

Now, so what are we?

We’re digging fossil fuels out of the earth.

So we’re level 0.5.

We’re level 0.3.

No, we’re level 0.

We’re level 0.

So Nick, if there’s a super intelligence, presumably they have better access to energy, especially the kind of energy you’re talking about that might need this simulation.

So have you thought about where a super intelligence might fit on the Kardashev scale?

Yeah, I think that would be higher up, just because at that level you would be able to run a lot more of these simulations.

And so even if there were some simulations run by, I don’t know, a Kardashev scale, one civilization, like with the Dyson sphere around their south, and that’s all they did, you know, once the civilization expands beyond that, they could run billions of times more, and that would be plenty of time for them to expand beyond that.

So you could imagine almost all simulations that are run are being run by civilizations that have reached the limits of whatever space they have to expand into.

That would presumably be Cardassia IV or something, unless the universe is so crowded that each one only manages to get the sort of galactic level volume before it bumps up against its neighbors.

Yeah, that’s a good point, because you can only have one galactic Cardassia scale civilization, because anyone else who wants it too bad, we’re using all the energy.

It’s like in the United States right now, there’s fights over the Colorado River Basin, because it’s a water source.

That river flows through multiple states, and each state has a pact with the other state how much water they’re supposed to use.

And there’ll be future fights on access to this one source of fresh water.

And so that’s an interesting point.

You can’t have a universe filled with high Cardassia level civilization, because they would implode rapidly, it seems to me.

And what level is Death Star?

What level is Death Star on the Cardassia scale?

Oh, well, in Star Wars Episode VII, it controlled the energy of a star.

So that would be, I guess, level two.

That’s only level two.

Wow.

Let’s jump right back in here.

By the way, in Star Trek, the Borg were…

That’s a super intelligence that was cosmic in its influence.

And so that would be even higher here, just to put that in context.

So Chuck, give me another one.

Okay, this is William DA.

Quite easy.

Nothing but letters.

Here we go.

He says, Where do you stand on the concept of consciousness, and where do you draw the line?

Would a simulated reality change your definition of what possesses consciousness?

I like that.

So Nick, is panpsych…

I presume that means that somehow consciousness is a shared entity that we all participate in as one organism?

Panpsychism, I mean, there are different sort of definitions, but it’s probably the view that everything is conscious.

Oh, wow.

And so how do you put consciousness in?

Is that a natural outflow of a sufficiently complex computer simulation of the brain?

That would be my sort of default assumption.

Yeah.

I mean, I think for the simulation argument, you can kind of plug in whatever your favorite theory of consciousness is, and most of them would work.

There might be some theories of consciousness which would not work.

The simulation argument, one of its assumptions, is what I call the substrate independence thesis, which is just the idea that in principle, you could implement consciousness not just on carbon-based biological structures, but on any suitable computational structure.

That what makes us conscious is not that we’re made of carbon, but that our brains perform a certain type of computation.

Whoa!

Wait a minute, so…

Max Tugmark has written on that very subject.

Yeah, your buddy Max.

He’s a friend of StarTalk’s.

And his thing is like the entire universe is made of math because it goes down to particles, and these particles have spin and, you know, so you can assign a value to them mathematically.

Mathematically constructed entities.

Yeah, there’s good overlap there.

But as to where to draw the line, like I don’t really have a very good account of exactly…

I mean, I think humans are conscious and rocks are not conscious, but like exactly where, sort of in the hierarchy, there would be a cutoff.

I’m not sure either that there is a sharp line there.

It might more be that there is kind of diminishing or more and more strained senses in which lower order organisms have some kind of consciousness and it kind of fades out rather than there being a sharp threshold, is what I would guess.

Now all I can think about is a conscious rock.

I just love the idea.

So here’s what I wonder, Nick.

I just have a not deeply thought out hypothesis that having thoughts such as we do that are incomplete and we wander and we don’t have good memory of things, we make stuff up, the fact that it’s not perfect, we interpret as consciousness.

Because if it were perfect, it’s just data and our brain is a storage disk that occasionally puts information together with a new result.

But the fact that we can sit there and say, oh, I feel this and I don’t, and it’s mostly how we reckon with our ignorance of our environment, even when we probe it for knowledge.

I’m just putting it out there.

Yeah, well, I mean, I guess first of all, you could have a lot of artificial, even simple systems that would be perfect in various ways.

You could have so faulty hard drives that randomly erase various things.

You could also have kind of compressed representations.

That’s what you have to do.

If you’re trying to do anything with AI, there’s a lot of data coming in and you have to extract some important features based on that and throw the rest away.

Wait, Chuck, what Nick just said, I can’t stop thinking about it.

So, Chuck, every time you and I forget something, the alien’s hard drive messed up.

So, every time you go, what did I come upstairs for?

It’s a read-write error, an IO error in a programmer’s disk.

I don’t think so.

I was just exploring your account of consciousness.

That somehow what’s necessary or sufficient for consciousness is that there is some kind of faulty or limited information processing.

Otherwise, it’s a perfect computer.

Well, I’m saying that computers are also imperfect in certain ways.

I’m not sure that the closer you get to perfection, you would lose consciousness.

I think in anything, it might go the other way around, that you might become more conscious if you want more.

But, yeah, you might have to…

If you wanted to elaborate on that, you might want to try to say which types of imperfection are the ones that are supposedly making a system conscious.

And maybe exploring that line of thought further, maybe you would get to something that would be some kind of possible account of consciousness.

I’m not sure.

Oh, wow.

Yeah, in the film iRobot, forgive me for not having read the original series of short stories by Isaac Asimov, but in the film, they hypothesize what could account for free will in a programmed robot.

And they were describing how many generations of operating systems are layered on top of one another.

And there’s always these dangling parts that you don’t always clean it up after because evolution is like this, too.

They’re dangling parts that worked at some point, now you don’t need them, or they could get in the way, or they could end up killing you.

But programmatically, there could be lines of code that have long lost their utility but could manifest under certain combinations of stimuli that look like the robot just thought of a new idea.

And I was intrigued by that suggestion when I heard it in the film, that that could be the way you end up with what we call consciousness.

But anyway, we’ve got to take another break.

When we come back for the third and final segment, we’re going to go through a lightning round with our questions.

And it’s Nick Bostrom just schooling us on whether or not we’re in a simulation.

And, spoiler alert, it sounds like we kind of are.

It sounds like, yeah, okay, when StarTalk returns.

We’re back, StarTalk.

I’ve got Nick Bostrom in the house.

Actually, he’s in the UK right now, but he’s in our Zoom house.

And we’re talking about the simulation hypothesis, and which he’s largely started, okay?

And so we blame him for all of our lost sleep at night.

At least, Nick, I blame you, if no one else does.

So Nick, the simulation hypothesis requires that every simulation has computers, right?

Why is that an obvious thing?

In fact, we’ve only had computers for like half a century, and we’ve been human for a couple hundred thousand years in our current form.

Why should it be inevitable that a computer is the thing that gets invented that then people want to simulate on?

I mean, I don’t know that it’s inevitable.

Maybe if there are a lot of humanoid species that never developed computers, I don’t know.

I mean, it suffices that some civilizations do develop computers and then more advanced computers of the type we can already see are physically possible, although we cannot build.

But certainly it’s consistent with a lot of civilizations failing to reach even our stage of development.

I mean, I think if you’re asking about inevitability, even if it’s not relevant for the simulation argument, it’s kind of interesting.

You want to, I guess, define what point in time, if it’s inevitable.

Like, it seems like the farther back you go, if you sort of reran evolution from that point, the less likely that you would get something similar to what we have today.

If you started with just bacteria, like who knows, maybe the chances would be very small, perhaps, that you would get an intelligent technological species.

But if you started like 50,000 years ago, then I mean, my guess would be we were already pretty well on the way and it was just a matter of time.

Oh, that’s an interesting point.

Okay, because the contingencies of evolution, right, it would take, in fact, if the asteroid didn’t hit 65 million years ago, the dinosaurs would be here and we wouldn’t, for sure.

You take it late enough, Nick, that’s a good argument.

Started 50, 100,000 years ago, I’m good with that.

We surely, there’d be some evolutionary path.

And just for those who are ET fans, consider that we all would judge the Roman Empire to be an intelligent civilization, yet aliens trying to communicate with them with radio waves would conclude that there’s no technology on Earth, right?

So we spent a lot of time being smart, but without the technology.

And so that’s the real question.

How much longer do we have technology before we exterminate ourselves?

But anyhow, Chuck, this is Cosmic Queries, bring it on.

And Nick, we’re going to try to bang out a whole lot in this third and final segment.

So let’s try to keep the answers tight, okay?

Bring it on.

All right, this is Dylan and Gordon Vu, going to mash up their questions.

Hello everyone from Albuquerque, New Mexico.

I’m a senior in high school and this question has been bugging me forever.

Do we have free will or is everything set in stone?

Are we living a predetermined life if we are in a simulation?

And then Gordon Vu says on top of that, if we manage to prove that we are living in a simulation, does that mean there is or is not a God?

Thank you.

Oh, talking about some philosophical, big gun questions.

Theological, philosophical.

So Nick, I love those questions.

Well, I mean, on the latter, I think it wouldn’t prove or disprove God.

I think it’s an independent question, whether we are in a simulation versus whether God exists.

So I don’t see any necessary connection there.

On the free will, I think we would have as much free will in the simulation as we would without the simulation.

I’m a compatibilist myself.

So I think that even if we are living in a deterministic physical universe, that that would be consistent with us having in the relevant sense, free will.

But you might have a different view on the metaphysics of free will, but I don’t think the fact that we would be in a simulation would necessarily change that.

Would that mean that the programmers of that simulation would program into our brains a perception of free will, even if they know the outcome in advance at every moment?

I don’t think that we need to especially program that in.

I mean, for the same reasons we, if we are not in a simulation, would have this notion of free will.

People in a simulation would presumably develop that for the same kind of reasons.

I mean, it connects obviously to holding people accountable for certain things they do.

I mean, if you stumble into somebody and bump them, we say, well, you’re excused because you didn’t intend it.

But if you go and punch them and achieve the same bruise, then you will be held accountable because that’s something you did off your own free will.

And so we make choices that we have to actually internally come up on a certain decision.

So all of those things would hold equally true for people in a simulation as people outside a simulation.

So if you have this free will, there wouldn’t really be a difference, I think.

Interesting.

Why can’t just the programmers be indistinguishable from God if they have power over everything?

I mean, depending on what you bake into the concept of a God, in many ways, that would be analogous to how some people have traditionally conceived of God, in the sense that they would kind of have created our world, although they wouldn’t have created a whole world, just the parts that we see.

They would presumably maybe not be omniscient, but they would know a lot, and they would not be omnipotent.

They would themselves be subject to the physical constraints operating at their level of reality, but they could intervene in our reality, including in ways that contravene the laws of physics that we perceive.

And thereby produce miracles.

Yeah, but things that appear to us in the simulation as miracles.

In one sense, there is a kind of structurally similar relationship.

On the other hand, they would be subject to all these, they would presumably be finite and subject to all these kind of limitations and constraints, and in that sense, kind of being infinitely far removed from a lot of the traditional conceptions of God, which is like a literally infinite and omnipotent and omniscient being.

So I think that whatever the truth is about the simulation hypothesis, it wouldn’t settle the question of whether there is this kind of more traditionally conceived, literally infinite God.

Okay, Chuck, keep it going.

Frederick Johansson wants to know, is general AI really a question about hardware and processing speed?

If it was, wouldn’t a computer today be able to simulate a few seconds of AI like it had a thousand years to process?

Yeah.

I mean, it’s a good question.

I think compute is a very important factor in driving AI progress over the last eight years or so with the whole deep learning revolution.

I think it’s maybe two-thirds of the progress we’ve seen is due to we are applying more compute and then maybe one-third is algorithmic progress.

Even if it were all compute, it doesn’t necessarily follow that we would be able to, with our current compute, run at least a small fraction of the human level mind.

Because there are two things you need the compute for.

One is to run the AI, right?

Like to actually have it do, but you also need to train up the neural network that becomes the AI.

So, if you don’t have enough computes to do the full training run, you might not even be able to develop the system, which then, if run, would constitute some kind of human equivalent level AGI.

Got it.

Right, because the calculation or the decision is not made in a vacuum.

It’s been completely preloaded with the world’s life experience, whatever is sitting right behind that one decision.

Is that a fair way to think about this?

So for humans to arrive at some sort of normal adult level of performance, we need 20 years or 15 years to kind of grow up and learn.

And our current neural networks are similar in that, although they are probably less efficient in learning.

So they might need, instead of 15 years of experience, maybe they need like a thousand years equivalent.

But you still need a lot of compute just to be able to complete something analogous to like a human maturation process.

So even if we had enough compute to run a human level AI, we might not have enough compute to sort of create it.

I think also, in addition to more compute, we also need some additional algorithmic insights.

But it’s not all or nothing.

The better the algorithms, the less compute you need to achieve this result.

And right now, the amount of compute you need would be like way more than we can currently afford.

And then it comes down as we make algorithmic progress at the same time as our computers become faster.

And at some point, these lines will intersect.

You just made a point embarrassingly clear that humans require like a fourth of our lives just to function as participating humans in civilization.

That’s embarrassing, but true, right?

No one trusts your decisions you make until you’re at least 20.

And even then, for some people, their brains are really not working even after.

As a kind of collective, we have just barely intelligence to create a technological civilization.

I think we look like we’re right on the cusp of that.

And it’s not so surprising maybe because if you imagine our ancestors had a lot less abstract reasoning ability, then it gradually improved over biological timescales.

And then as soon as we became capable of creating a technological civilization, then we pretty much did it or after 10,000 years or something.

So we should kind of maybe expect that we are at the lower end of what is needed to do this at all.

And that maybe explains some of what we see in the world, that we’re kind of fumbling our way.

A lot of what we see in the world, yes, yes.

All right, Chuck, give me more.

Okay, Skyler Gravat says, if this is a simulation, why are the people running the simulation so patient?

The universe is estimated to be 13.

something billion years old and they waited almost 10 billion years to simulate life.

Well, so first, there is no particular reason to think that those 10 million years were simulated.

You don’t need to do it from the Big Bang onwards.

You could start the simulation from a later point.

You’d embed the simulation with evidence that that simulation scientist would then interpret as an old universe, but it’s all just fake.

I mean, you probably don’t want to sit through 10 billion years of just gas clouds congealing.

That would be a pretty wasteful use of it.

It’s kind of boring, right?

But even when you get into the…

Say they were interested in all of human history for the sake of the argument, that that’s 10,000 years onward.

It doesn’t mean that for them it would take 10,000 years to do this.

They could run the simulation at a higher speed, like maybe one minute of their time could simulate a thousand years.

It depends on how fast the computer is that you run the simulation on.

Gotcha.

Right, so when we had a great revelation when computing power was adopted by astrophysicists in the 1970s, we were early out of the box on this.

There were these galaxies in the universe that were kind of funky looking, and we made catalogs called peculiar galaxies.

We didn’t, you know, maybe just thought galaxies were made that way.

Only after we were able to simulate the collision of two galaxies did we realize that this is like the crash scene leftovers of what happens when galaxies collide, and we simulate a billion years in a matter of minutes.

And in so doing, we were able to populate the entire catalog of galaxy parts and nasty, twisted-looking galaxy forms simply by seeing what happens when they collide and speeding up the time to do so.

That’s just a little aside.

Wow, are you prescient or what?

This is Nathaniel Mitchell who says, if we could ever simulate an exact replica of our universe down to the spin on the components of quantum particles, could we speed it up and then use it to predict our future as we now do with simulations for climate and otherwise, but yet on a cosmic scale?

Well, so that kind of thing wouldn’t fit into our universe, like a computer that simulated all of our universe.

It wouldn’t be possible to build that in our universe.

Oh, that’s a philosophical challenge.

That’s like saying, how detailed do you want your map to be if you have a map the size of the UK, then it would have all the detail of the actual island, but then you could just use the island.

You don’t need that.

Yeah, you like that.

I think it would be very infeasible to simulate our world at the level of quantum properties, at least if the simulator’s universe looked anything like our universe.

But maybe the physics at their level of reality is different.

Maybe it’s possible to build more powerful computers.

You could even imagine hypercomputation being possible in some other kind of physics so that they could run literally infinite computations, and then maybe they could simulate a world like ours at full quantum detail.

And then run it forward and watch the future evolution.

From our point of view, it would presumably not make much difference whether they did it that way or the much cheaper way that would only render things at the sufficient level to be convincing to the people inside.

And in fact, even if you imagine that there were some simulators that could do this at full quantum detail, it would cost them so much more compute that it would still likely be the case that almost all simulations would run in the more efficient way, that would only simulate things at the coarser grid.

So even if there were some full-grain simulations, we would probably be in one of the other ones because that would be a lot cheaper and so you could create orders of magnitude more of that.

And how much of this relates to the fact that it’s hard for something to understand itself?

Like, can the brain, the human brain, actually come to understand the human brain?

Or do you need a higher intelligence than the human brain to then study the human brain as a thing outside of itself?

Understanding is a matter of degree, right?

We understand a bit about ourselves now.

We could understand more.

I mean, obviously, you couldn’t have a full simulation of all the details in the human brain stowed away in a part of the human brain, right?

That’s the map of the UK.

That would be as big as the UK.

I actually wanted all the details.

I’ve been to paradise, but I’ve never been to me.

We’ve got to land this plane.

Let me just offer my best evidence for why I think we live in a simulation.

I’m just going to go public on this.

I think right when civilization is kind of going smooth, then something happens, okay?

A politician rises up, there’s a war, there’s a world war, there’s tsunamis, and I think the aliens program that in for their own entertainment.

Because that’s what we did in the Sim games.

In Sim City, where you’re mayor of a city and everything’s going fine, unannounced, Godzilla trouches through your city, and now you have to deal with it, the fire and the police and to rebuild the schools.

And that’s the programmer sending that in without telling you what’s going to happen.

I think all of the troubles we have in the world is evidence that the programmers need entertainment.

Yeah, well, Nick, like I said, we got to land this plane.

Thank you for coming out to StarTalk.

This conversation was long overdue.

I wanted to get you a few years ago, but you were in high demand, and you still are for sure.

But if any of us discover something, like we part the curtain and we see like a CPU there when it was supposed to be a couch, I’ll call you up.

If anybody viewing that does that, contact Neil rather than me.

I don’t need more.

For sure.

All right, it’s been a delight.

Nick, is Superintelligence the book you would have people sort of check out in terms of the foundations of this thinking?

Yeah, well, not specifically on the simulation argument.

There the article is online.

Just Google it, simulation argument, you’ll find it.

But if you want to read a book about the future of AI and stuff, then Superintelligence would be the one I would point to.

Excellent, excellent.

All right, good.

Nick, again, thanks for joining us.

I’m glad we could do it.

Yeah, finally, this Cosmic Queries episode.

I’m Neil deGrasse Tyson, your personal astrophysicist.

As always…