In January, Stephen Wolfram — the computer-scientist, part-time epistemologist, and author of both “A New Kind of Science” and the Wolfram Language, a “universal” programming language that (among other things) informed the alien communication in the movie Arrival — wrote an exceedingly long blog post about how best to communicate with aliens. It was called “Showing Off to the Universe: Beacons for the Afterlife of Our Civilization.” In March, Tim Urban of waitbutwhy.com called him up to talk about it.
I’m very excited, this is very fun to get to talk to you. You know, you kind of created a universal language you might be able to communicate with aliens with. I don’t talk to many people who have done that.
Indeed.
But I have talked to a few people recently about the Fermi paradox, including Anders Sandberg at Oxford’s Future of Humanity Institute, who recently co-authored a really fascinating paper … about how when you run a simulation of all the possible outputs of the Drake equation, using this Monte Carlo technique, you end up with the conclusion that there’s a substantial probability that we’re alone in the Milky Way.
Oh, I did see some mention of this, but I think I saw it in New Scientist.
Yeah.
Which I’ve subscribed to for 50 years, but my main conclusion is if I read it in New Scientist, it probably won’t happen.
Okay, well, I guess that’s good because it freaked me out a little bit. I’ve written about the Fermi paradox and have always grouped expert theories into two broad camps: There are those who say, “Look, we haven’t heard from other civilizations because there probably aren’t any out there, because we would have seen some evidence otherwise” — like Nick Bostrom and a number of others … And then the second camp that says, “There are almost definitely many other advanced civilizations out there, there are just plenty of other explanations for why we wouldn’t see evidence of them.” And this paper was interesting because it kind of was saying both camps are legitimate. Not only might we be alone in the Milky Way; we might be the only intelligent life, or even the only life at all in the observable universe. This is what the simulation suggests, but at the same time the median result that the simulation spits out is that the galaxy is teeming with life. So it very much allows for both possibilities.
I think the main difficulty with all of these things is the definitional one. That is … you talk about life, you talk about intelligence; what are those things abstractly?
We know the specific example that we have historically been exposed to: life on Earth, human intelligence. The question is, when you generalize away from that, what do you get to? In my work in basic science, one of the main conclusions has been that the core element of both life and intelligence is sophisticated computation. And it turns out sophisticated computation is actually a quite ubiquitous thing.
One of the things that I’m fond of quoting is the statement “The weather has a mind of its own.” What does this mean? What is mind? What is the abstract kind of thing that’s like mind? It’s the ability to do sophisticated computation. That’s something that exists in the weather, just as it exists in our brains, just as it exists in lots of living systems. And then the question is, what’s different between the weather and its sort of mindlike thing and our human intelligence? The fundamental answer to that is our human intelligence has its particular cultural, civilizational history and the weather doesn’t.
Right.
People turn to start saying, “What is intelligence? What’s special about, for example, human intelligence?” I think this is a moment, with respect to intelligence, that’s very much like the moment 400 years ago or something, with Copernicus and so on, where we’re asking, “Okay, is our Earth at the center of the universe? Is there something extremely special about the Earth?” And there’s nothing else like it in the universe, so to speak.
We kind of have this assumption that our particular form of intelligence, with its particular history, is somehow the only, the best, the definition of intelligence, so to speak. And I just think this isn’t true.
Now when we ask, “Okay, have we seen extraterrestrial intelligence?” — okay, what might that mean?
Right — what exactly is that question asking?
Well my own suspicion is that the first form of nonhuman-aligned intelligence that we’re getting exposed to is artificial intelligence. My guess about how things will develop historically is we’ll get more and more used to the idea of AI and nonhuman intelligence and so on. Eventually, we’ll get used to the idea that there are these kind of nonhuman intelligences embodied in what’s possible in the computational universe and what we can use for AI. Then we’ll realize that actually lots of things out there in the universe are intelligence-like things. And we’ll eventually realize the resolution of the Fermi paradox is, yeah, in some sense there’s intelligence in lots of places in the universe; it just hasn’t happened to be historically aligned with our particular intelligence.
Now we might turn out to be lucky in a sense, and it might turn out that in the space of possible intelligences, there are some whose history is sufficiently similar to ours that we can kind of point to a lot of these details of human intelligence and say, “Yeah, that’s a replication of that.” But it’s hard to know what this kind of space of possible intelligences looks like — how far apart the different things in it are. And that’s my feeling about the resolution of the apparent paradoxes — there’s intelligence in lots of places in the universe, it’s just that it doesn’t happen to be in detail aligned with our idea of intelligence.
It sounds like what you’re doing is you’re zooming out really far. When you zoom out, you’re saying that even life as we define it is not actually a clear-cut thing. It’s not a binary universe where there’s lots of dead stuff and then there’s this special thing that we call life. And life has the capacity for intelligence, while dead things do not. You’re saying that life is just another system that can exhibit some kind of sophisticated computation and that we’re very Earth-life-centric when we think about this problem.
Let’s take life as an example. How do we define it? You look at the historical efforts to define life; first of all, it was “Things which can move themselves must be alive.” Okay, well then, with all kinds of steam engines and other kinds of technology — that’s not a good definition. Then there are definitions like “Life means certain kinds of organic molecules”; those turn out to be synthesizable, and they turn out to even exist in some cases in interstellar space and so on.
Then there are the definitions like decreasing entropy or self-replication. Self-replication, for example; we know it’s easy to make programs that replicate themselves. It’s even easier to make things that replicate themselves — well, certainly in the digital domain it’s particularly easy to make things that replicate themselves.
But there’s nothing tremendously special about this feature of self-replication.
But people do seem to have some intuitive sense of what it means for something to be alive, don’t they? Even if it’s a sense that changes over time?
If you present somebody with, “Here’s two things, which one of them is alive?” For us, given the history of life on Earth, it’s pretty easy to answer that question. You go look and see if it has RNA in it, see if it has cell membranes in it. Those things are, for us, the pragmatic, practical definition of life. But I don’t think we believe that the essence of life is RNA, is that specific molecule. And so as you say, one is thinking about zooming out to figure out what is the abstract definition of life.
I remember when I was a kid and the first Mars landers were going out. One of the big questions was, “Was there going to be life on Mars?” And there were little things that scooped up soil from the Martian surface and then they did various tests on them. And I think the one that was kind of the finest and best test for life was pretty much you feed the soil sugar and you see whether it metabolizes. Now I don’t think that the essence of what we might reasonably call life is that it happens to like eating sugar, so to speak. That seems a very disappointing conclusion. If that’s the thing we’re aiming for, if that’s the thing they’re concentrating on, that doesn’t seem like really a suitable, general definition that it’s likely to cover things that are lifelike out there in the universe.
You could even look at the whole thing as a continuum — it’s all just atoms and energy doing stuff. And atoms and energy do stuff in stars and planets and gas clouds, dictated by these fundamental forces, and then they also sometimes form what we call biology. And then the biology sometimes forms what we would call artificial intelligence. And if you take a big step back, it’s just all one thing. It’s atoms doing what atoms do. But I still think that leaves an open Fermi paradox, which is just asking the question, “Well, okay, but how about life like us? How about other intelligent stuff that we would recognize as biology that has something that we would call self-awareness that is …”
You’re sliding down a slippery slope there because you bring in self-awareness and you think perhaps that that’s kind of the key thing. I’ve studied a lot of simple computational systems. One of my favorite types is called cellular automata, which is sort of arrays of black and white cells that follow simple rules. And you look at what they do, and they do very complicated things. And as you start talking about them and trying to explain to somebody, “Well, this is what it does. It’s got this thing, it’s got that thing.” You pretty quickly find yourself giving a very humanlike explanation for what it’s doing: “It wants to do this. It’s trying to do that.” You kind of realize that you are, at least implicitly, attributing a certain kind of free will to the system. And when you talk about all these kinds of things — consciousness, self-awareness, free will … those are slippery things to define. But I don’t think that there’s a definition of them that is sufficiently abstract, so it would apply to these other sophisticated computational systems.
Now, imagine radio signals coming from the cosmos and ask yourself, Are they intelligent or not? What would it take for you to be convinced that they were intelligent? A little more than 100 years ago, Tesla [and] Marconi were kind of radio pioneers. And they were listening to natural radio for the first time. And I think Tesla in particular found that he could hear these weird sounds when converted from radio coming from the cosmos, so to speak. And his immediate assumption was that must be signaling by the Martians. It sounds complicated enough, it must have an intelligent origin; it must be the Martians. And it turns out what he was hearing were features of the magnetosphere and the ionosphere that are, in a sense, what we would say today are pure physics. When you listen to them, you say,What do those sound like? And one of the obvious things that they sound like is they sound like whale songs.
Even in the case of whales, it’s very unclear what the intelligence level or the communication level or whatever of whale songs actually is. And I think that’s a good example of where you could say, “Well, let’s talk Fermi paradox on Earth. Why isn’t there more intelligence on Earth?” Well, are we sure there isn’t? What exactly is going on with whales?
Those creatures don’t happen to build skyscrapers and electric cars and so on, but that’s a fairly recent thing in human civilization and I think we’d still count ourselves intelligent before we had all the modern skyscrapers and so on. So I think it’s a really tricky thing. It’s very clear what we mean by humanlike intelligence, at least at this point in history. Humanlike intelligence may itself evolve over time and may even be — as we get more connected to AI and more digital and so on — it may end up being almost unrecognizable at some point in the future.
Or it might become just an unimportant distinction. When I think about cellular automata, I think that when you really start boiling down the physical world, it seems like it ends up becoming indistinguishable from the computational world in a way; where at the very core, the physical world is just elementary particles instructed by the rule of the forces of physics to do something that’s based on what’s happening in their neighborhood. And it suddenly starts to feel like if you boil down a human, you end up with something that is no more alive than a sequence of cellular automata. Is that fair?
Yes, a whole lot of that’s right. But in a sense all that’s saying is we’re all fated to be part of the universe with its particular laws of physics. We, by our luck or something else, have been thrown into this particular universe in which we happen to live. And it has certain attributes. The fact is that what we understand from the structure of computation and so on is that within that physical universe, we can construct essentially virtual worlds that behave according to any other rule that one might find in the computational universe. Just like we can program a computer to do any kind of computation, we can similarly program our universe to effectively do any kind of computation.
It like our universe is just one program. Our physical universe is a subset of the far more vast computational universe.
What we realize is that in a sense, ubiquitous computation is ubiquitous; it’s kind of a blank slate. There’s nothing notable about it in this sense.
So when we think about the human condition and humans and so on, what we realize is actually many of the details of human condition are what we should really care about because they’re what’s special about us. It’s not this thing where we say, “Look, the thing that’s really special about us is we have intelligence and nothing else does.” That’s not right. The thing that’s special about us is endless details of our particular history that won’t be shared by anything else in the universe. It’s inconceivable. Probabilities are vastly staked against any repetition of the detailed history that’s led to our particular human intelligence and so on.
This leads into the main topic for this conversation, which is that you wrote a very long, very considered post on your website about how we might, theoretically, try to communicate with aliens. I want to get into that, but first, let’s discuss why we would even want to do this. Why is this mission important to try to communicate and send out time capsules and capture information about us?
I think it’s having pride about what we’ve achieved and building a monument to it. That’s as important as it is to imagine that at some point, some alien is gonna pick it up and do something with it. It’s like, Why were the pyramids built? Well, it was probably in large part as an exercise for the time, more so than that they thought this is gonna be something to transmit to the afterlife.
It’s something which is an interesting thing for our current civilization to imagine: What is the essence of what we’ve produced and how can we imagine communicating it? I’m kind of particularly amused by for example, Elon Musk’s car going into … launching his car into space, because that is so extremely aligned with the notion of grave goods from ancient Egypt or ancient China or something. It’s really charming. It’s literally this concept of grave goods, where you’re taking things from your everyday life and you’re burying them with you. In this case, you’re sending them out into the cosmos.
Capturing who we are and sending it outward is just as much an emotional thing as it is anything else. We happen to be attached to this information. We think it’s important to us. And it makes us sad to consider it just being extinguished from the record. And so maybe we don’t need a better reason than that. We would like this information to be preserved, even if we don’t end up being preserved.
And the questions you ponder in your post are what exactly we should be trying to communicate and how we could actually get that information across effectively. And you concluded that, even though you yourself have written a whole language to use to communicate, that we might not actually do better than sending cryogenically preserved humans, which, as you pointed out, is a lot like what the Egyptians thought they were doing.
Right.
And that’s because so many other approaches have problems with them, both in the content of what we might send out and the method we’re using to communicate that content. Let’s start with the first issue — what exactly is it that we would want to say to the rest of the universe if we had the chance? People have often proposed that we communicate math, since they view it as universal — but as you point out, if you’re just letting them know one plus one equals two, what are you actually communicating?
Right, because what do you learn about the originators of the message one plus one equals two? For example, we have lots of tablets, Babylonian mathematical tablets. And what do we learn from those? The thing we learn is how much math the Babylonians knew, but once you’ve got a fix on how much math the Babylonians knew, every new tablet gives you no new information, because it’s telling you a bunch of arithmetic facts, which are not special for the Babylonians, so to speak.
I think the thing to realize is that in the computational universe, we in our particular history of our civilization have followed a particular path. There are an infinite number of possible paths that we could’ve followed. We understand the particular path that we’ve followed, and you could imagine some other “civilization” in quotes somewhere that followed a different path. And to them, our path would be quite as mysterious as the paths that these systems like the neural networks we’re building and so on take.
It’s connected, I think, to your ideas about computational irreducibility. In our day-to-day lives, there are a lot of examples of ongoing processes that follow a consistent pattern that allows us to use math to see how things will play out in the future. If we know a fixed growth rate on an investment, that’s computationally reducible, so it’s easy for us to plug “t = 30” into an equation and see what the investment will have grown to 30 years from now. But in the natural universe, things don’t tend to boil down to consistent, repeating patterns — they tend to be, as you say, irreducibly computational. You mentioned the weather, which is an example of something that’s not computationally reducible. You can ask the smartest person in the world what the weather will be like on a particular Thursday 100,000 years from now, but they’re not going to be able to get there with math — you’re gonna have to go through 100,000 of weather patterns, step by step, a few days at a time, without making any mistakes along the way. In theory, a powerful enough computer might be able to do that, but it would take a tremendous amount of computation.
That’s right. I think the thing to realize is at some level, computational irreducibility seems disappointing because it means that even if you have the best theory in the world, you can’t predict everything. But it also is kind of very satisfying because it means that if you look at the evolution of human civilization, or any one of a whole lot of different processes that correspond to sophisticated computations, you don’t just say, “Oh, you don’t really need to do that computation. You don’t need to live through all of that, you can just jump to the end and say this is the answer.” There’s something actually irreducibly achieved by going through those steps of evolution. By actually living through it, so to speak.
So it might seem intuitive to us that if you know enough about the Big Bang and you know exactly how the earliest particles were moving, you can start to find a rule and you can just apply that rule forward and enter the current year as the time variable and it will spit out the universe as it is now, which would include all of the details of human civilization. And your point is that because of computational irreducibility, it’s like the weather where you can’t actually do that. The only way to actually discover what’s on step x is to just go through all the steps to get there. You have to just live through it and actually see what is currently here to understand what would have ended up here. So, by communicating to aliens the specifics of what happened here on Earth — what this program, this physical universe program, has outputted in this particular location — we actually tell them something both unique to us and something they don’t know. It’s something they have no way to know without a grand universe simulator.
And when it comes to simulating the whole physical universe, we’re obviously kind of out of luck because we have to build that simulator in the physical universe and it doesn’t get to outrun itself.
Right.
The fact is, this is beginning to seem obvious to us. That conclusion is beginning to seem obvious to us. But it’s kind of interesting that for the last 300 years, that conclusion would be, “Oh my gosh, that can’t possibly be right.” Because the basic idea of exact science that came from Galileo and Newton and people like that 300 years ago was actually, you can write down a mathematical equation and just be able to jump to the answer.
So as far as what we might want to communicate to extraterrestrial life, the valuable thing we can tell them is about the particulars of us and our history. You could have the most sophisticated species, they could be controlling the energy of their whole galaxy, and we still have something they can’t get.
That’s right.
And that brings us to the second quandary here, which is about how we communicate this information to extraterrestrial life. You write about how we need to somehow form a language with whomever we’re communicating with. And once we have a common language that they can understand, we can suddenly tell them everything. We can send them the contents of Wikipedia or Wolfram Alpha, or we can send them the contents of every book, and they can, with the language, now learn everything that we would ever want them to learn about us. So the challenge is coming up with some kind of language that can work.
But what is language? What is the point of language? If you have a brain, maybe it’s a neural net, maybe it’s an actual biological brain and you’re trying to communicate something that brain knows to some other brain. Well, if these brains were exactly, precisely similarly constructed; maybe you could just do a brain dump. Move the information just bit by bit to the other brain. That probably wouldn’t work because the other brain already has a different state from different experiences that it’s had. You don’t get to just do bit-for-bit copying of information like that, brain-to-brain, so to speak. You kind of have to communicate things at a somewhat higher level. When you walk down the street, you don’t want to have to say there are ten-to-the-26th atoms that are arranged in this particular way, and they’re doing this particular thing, because what you want to do is zoom out and make this more symbolic statement about what’s going on.
Right.
So I’ve spent a large part of my life building this thing they call Wolfram Language, which is a symbolic representation of things in the world and things that humans care about doing in the world. And in practice, Wolfram Language, which powers Wolfram Alpha, which powers things like Siri and so on; it’s in practice used by lots of people to … It’s been used to discover and invent lots and lots of things over the past 30 years where basically it’s humans are describing in Wolfram Language at a high level what ultimately can get automatically done by computers.
It’s actually rather easy to capture all possible computations but at sort of the lowest level, but what we want to capture is at a high level, the kinds of computations that humans in the present time and civilization, present time in history and so on, are interested in. And so, the goal there is to have a precise representation of lots of things that we humans care about doing. And one of the things that’s important, then, is that that’s something that can readily transport those complete ideas and goals and so on because it’s a precise, symbolic language that you can execute the same way. It just has particular rules for executing it.
So if we decided we want to communicate using the Wolfram Language, how would we physically transmit that to them? You’re working with the Arch Mission, where they want to send information on these little quartz discs. But how would an intelligent life know something is on the disc and be able to read it?
Those are really good questions. I mean the fact is that even when you say, “Know something is there.” What does “know” mean in this sense? If you have a big arrow that says, “Look here. This is the thing you should care about.” I think that’s honestly pretty hopeless. So let’s imagine that we have something that we can represent that’s sort of the crystallization of our civilization and put it out in the cosmos and it gets found by some other thing that we might recognize as sort of an intelligence-like thing. And then that intelligence-like thing goes on and it … What would it be? If we were to see what was happening, what would make us say, “Oh, great, they got it. They understood what was going on.”
Well if a similar kind of species was so smart that they saw this disc, they said this looks like something that must have been sent from … This must have been intentionally created. And they’re advanced enough to figure out how to read what’s on the disc. And then one day they’re putting on Shakespeare plays on their version of Broadway because they have all the texts on the disc. They have absorbed it, they’re listening to our music, they’re arguing about our philosophy, they’re debating our history. Wouldn’t we say that succeeded?
Well, okay, so here’s sort of a counter to that; which is we’ve been transmitting radio for 100 years, roughly. Somewhere out there in the universe the Lucy show is — it’s been what, 50-something years, so it’s 50 light-years away. There’s probably some cloud of ionized gas where atoms in it are jumping around at least a tiny bit to correspond to the pixels or the scan lines or something in the I Love Lucy show or some such other thing.
Now, the question is does that mean they got that soap opera, so to speak? Or is that just a physical process where the radio wave goes and it impinges on this plasma and it’s doing this particular thing? Well, it’s a little hard to know. That’s, again, a slippery slope because you could say, “Well, this thing that is … It’s a human effect on the natural world and the human effect on the natural world has the natural world respond in some complicated way to that human effect.” Does that mean the natural world understood the human’s input? Or is it just that we say, “Well, let’s take it apart. It’s just a natural process, it’s just operating going into physics.”
You could do the same for us — for this conversation.
Yes, somebody could look from the outside and say, “Are these guys communicating? Or is it just the sound waves are turning into bit patterns and they’re coming out of the other end and then that’s affecting neurons in this brain and electrochemistry is going on and so on. It’s all just physics. There’s nothing happening here. No communication is happening, it’s just physics.” How do we distinguish that from something where we’re talking about it now?
Now the one place where there is perhaps some distinction is, you do have to reach this threshold of sophisticated computation when all that’s happening is pure parroting of what goes on. The sound wave is propagating through the air, and it’s just the compression wave is at this place and it moves to that place — that’s a sufficiently simple process that I don’t think one would say reaches this threshold of universal computation. You would say, “No, no. There isn’t anything. That’s not intelligence by any definition.”
But there’s plenty that does meet that threshold. A lots of things in the natural world, in the computational universe, reach that level of equivalent computational sophistication. And you might think there’s this whole hierarchy of computational sophistication … There are these systems down here that are almost smart and then there are these systems that are a bit smarter and a bit smarter, and then eventually you reach our intelligence and then you go beyond that and you get to superintelligence and hyperintelligence. And there’d be some whole infinite hierarchy of levels of intelligence. But I don’t think that’s the way it is. I think that’s just one level and it’s not very hard to reach it.
And I think that’s the thing when you say, “Is there intelligence out there in the universe?” The answer is yes, there’s lots of it.
*A version of this article appears in the March 19, 2018, issue of New York Magazine. Subscribe Now!