The First World Takeover

Eliezer Yudkowsky19 Nov 2008 15:00 UTC
42 points
24 comments6 min readLW linkArchive

Before Robin and I move on to talking about the Future, it seems to me wise to check if we have disagreements in our view of the Past. Which might be much easier to discuss—and maybe even resolve… So...

In the beginning was the Bang. For nine billion years afterward, nothing much happened.

Stars formed, and burned for long periods or short periods depending on their structure; but “successful” stars that burned longer or brighter did not pass on their characteristics to other stars. The first replicators were yet to come.

It was the Day of the Stable Things, when your probability of seeing something was given by its probability of accidental formation times its duration. Stars last a long time; there are many helium atoms.

It was the Era of Accidents, before the dawn of optimization. You’d only expect to see something with 40 bits of optimization if you looked through a trillion samples. Something with 1000 bits’ worth of functional complexity? You wouldn’t expect to find that in the whole universe.

I would guess that, if you were going to be stuck on a desert island and you wanted to stay entertained as long as possible, then you should sooner choose to examine the complexity of the cells and biochemistry of a single Earthly butterfly, over all the stars and astrophysics in the visible universe beyond Earth.

It was the Age of Boredom.

The hallmark of the Age of Boredom was not lack of natural resources—it wasn’t that the universe was low on hydrogen—but, rather, the lack of any cumulative search. If one star burned longer or brighter, that didn’t affect the probability distribution of the next star to form. There was no search but blind search. Everything from scratch, not even looking at the neighbors of previously successful points. Not hill-climbing, not mutation and selection, not even discarding patterns already failed. Just a random sample from the same distribution, over and over again.

The Age of Boredom ended with the first replicator.

(Or the first replicator to catch on, if there were failed alternatives lost to history—but this seems unlikely, given the Fermi Paradox; a replicator should be more improbable than that, or the stars would teem with life already.)

Though it might be most dramatic to think of a single RNA strand a few dozen bases long, forming by pure accident after who-knows-how-many chances on who-knows-how-many planets, another class of hypotheses deals with catalytic hypercycles—chemicals whose presence makes it more likely for other chemicals to form, with the arrows happening to finally go around in a circle. If so, RNA would just be a crystallization of that hypercycle into a single chemical that can both take on enzymatic shapes, and store information in its sequence for easy replication.

The catalytic hypercycle is worth pondering, since it reminds us that the universe wasn’t quite drawing its random patterns from the same distribution every time—the formation of a longlived star made it more likely for a planet to form (if not another star to form), the formation of a planet made it more likely for amino acids and RNA bases to form in a pool of muck somewhere (if not more likely for planets to form).

In this flow of probability, patterns in one attractor leading to other attractors becoming stronger, there was finally born a cycle—perhaps a single strand of RNA, perhaps a crystal in clay, perhaps a catalytic hypercycle—and that was the dawn.

What makes this cycle significant? Is it the large amount of material that the catalytic hypercycle or replicating RNA strand could absorb into its pattern?

Well, but any given mountain on Primordial Earth would probably weigh vastly more than the total mass devoted to copies of the first replicator. What effect does mere mass have on optimization?

Suppose the first replicator had a probability of formation of 10-30. If that first replicator managed to make 10,000,000,000 copies of itself (I don’t know if this would be an overestimate or underestimate for a tidal pool) then this would increase your probability of encountering the replicator-pattern by a factor of 1010, the total probability going up to 10-20. (If you were observing “things” at random, that is, and not just on Earth but on all the planets with tidal pools.) So that was a kind of optimization-directed probability flow.

But vastly more important, in the scheme of things, was this—that the first replicator made copies of itself, and some of those copies were errors.

That is, it explored the neighboring regions of the search space—some of which contained better replicators—and then those replicators ended up with more probability flowing into them, which explored their neighborhoods.

Even in the Age of Boredom there were always regions of attractor-space that were the gateways to other regions of attractor-space. Stars begot planets, planets begot tidal pools. But that’s not the same as a replicator begetting a replicator—it doesn’t search a neighborhood, find something that better matches a criterion (in this case, the criterion of effective replication) and then search that neighborhood, over and over.

This did require a certain amount of raw material to act as replicator feedstock. But the significant thing was not how much material was recruited into the world of replication; the significant thing was the search, and the material just carried out that search. If, somehow, there’d been some way of doing the same search without all that raw material—if there’d just been a little beeping device that determined how well a pattern would replicate, and incremented a binary number representing “how much attention” to pay to that pattern, and then searched neighboring points in proportion to that number—well, that would have searched just the same. It’s not something that evolution can do, but if it happened, it would generate the same information.

Human brains routinely outthink the evolution of whole species, species whose net weights of biological material outweigh a human brain a million times over—the gun against a lion’s paws. It’s not the amount of raw material, it’s the search.

In the evolution of replicators, the raw material happens to carry out the search—but don’t think that the key thing is how much gets produced, how much gets consumed. The raw material is just a way of keeping score. True, even in principle, you do need some negentropy and some matter to perform the computation. But the same search could theoretically be performed with much less material—examining fewer copies of a pattern, to draw the same conclusions, using more efficient updating on the evidence. Replicators happen to use the number of copies produced of themselves, as a way of keeping score.

But what really matters isn’t the production, it’s the search.

If, after the first primitive replicators had managed to produce a few tons of themselves, you deleted all those tons of biological material, and substituted a few dozen cells here and there from the future—a single algae, a single bacterium—to say nothing of a whole multicellular C. elegans earthworm with a 302-neuron brain—then Time would leap forward by billions of years, even if the total mass of Life had just apparently shrunk. The search would have leapt ahead, and production would recover from the apparent “setback” in a handful of easy doublings.

The first replicator was the first great break in History—the first Black Swan that would have been unimaginable by any surface analogy. No extrapolation of previous trends could have spotted it—you’d have had to dive down into causal modeling, in enough detail to visualize the unprecedented search.

Not that I’m saying I would have guessed, without benefit of hindsight—if somehow I’d been there as a disembodied and unreflective spirit, knowing only the previous universe as my guide—having no highfalutin’ concepts of “intelligence” or “natural selection” because those things didn’t exist in my environment, and I had no mental mirror in which to see myself—and indeed, who should have guessed it with short of godlike intelligence? When all the previous history of the universe contained no break in History that sharp? The replicator was the first Black Swan.

Maybe I, seeing the first replicator as a disembodied unreflective spirit, would have said, “Wow, what an amazing notion—some of the things I see won’t form with high probability, or last for long times—they’ll be things that are good at copying themselves, instead. It’s the new, third reason for seeing a lot of something!” But would I have been imaginative enough to see the way to amoebas, to birds, to humans? Or would I have just expected it to hit the walls of the tidal pool and stop?

Try telling a disembodied spirit who had watched the whole history of the universe up to that point about the birds and the bees, and they would think you were absolutely and entirely out to lunch. For nothing remotely like that would have been found anywhere else in the universe—and it would obviously take an exponential and ridiculous amount of time to accidentally form a pattern like that, no matter how good it was at replicating itself once formed—and as for it happening many times over in a connected ecology, when the first replicator in the tidal pool took such a long time to happen—why, that would just be madness. The Absurdity Heuristic would come into play. Okay, it’s neat that a little molecule can replicate itself—but this notion of a “squirrel” is insanity. So far beyond a Black Swan that you can’t even call it a swan anymore.

That first replicator took over the world—in what sense? Earth’s crust, Earth’s magma, far outweighs its mass of Life. But Robin and I both suspect, I think, that the fate of the universe, and all those distant stars that outweigh us, will end up shaped by Life. So that the universe ends up hanging quite heavily on the existence of that first replicator, and not on the counterfactual states of any particular other molecules nearby… In that sense, a small handful of atoms once seized the reins of Destiny.

How? How did the first replicating pattern take over the world? Why didn’t all those other molecules get an equal vote in the process?

Well, that initial replicating pattern was doing some kind of search—some kind of optimization—and nothing else in the Universe was even trying. Really it was evolution that took over the world, not the first replicating pattern per se—you don’t see many copies of it around any more. But still, once upon a time the thread of Destiny was seized and concentrated and spun out from a small handful of atoms.

The first replicator did not set in motion a clever optimization process. Life didn’t even have sex yet, or DNA to store information at very high fidelity. But the rest of the Universe had zip. In the kingdom of blind chance, the myopic optimization process is king.

Issues of “sharing improvements” or “trading improvements” wouldn’t even arise—there were no partners from outside. All the agents, all the actors of our modern world, are descended from that first replicator, and none from the mountains and hills.

And that was the story of the First World Takeover, when a shift in the structure of optimization—namely, moving from no optimization whatsoever, to natural selection—produced a stark discontinuity with previous trends; and squeezed the flow of the whole universe’s destiny through the needle’s eye of a single place and time and pattern.

That’s Life.

Eliezer Yudkowsky19 Nov 2008 15:00 UTC
42 points
24 comments6 min readLW linkArchive