Timeless Identity

Eliezer Yudkowsky3 Jun 2008 8:16 UTC
61 points
249 comments14 min readLW linkArchive

Followup to: No Individual Particles, Identity Isn’t In Specific Atoms, Timeless Physics, Timeless Causality

People have asked me, “What practical good does it do to discuss quantum physics or consciousness or zombies or personal identity? I mean, what’s the application for me in real life?”

Before the end of today’s post, we shall see a real-world application with practical consequences, for you, yes, you in today’s world. It is built upon many prerequisites and deep foundations; you will not be able to tell others what you have seen, though you may (or may not) want desperately to tell them. (Short of having them read the last several months of OB.)

In No Individual Particles we saw that the intuitive conception of reality as little billiard balls bopping around, is entirely and absolutely wrong; the basic ontological reality, to the best of anyone’s present knowledge, is a joint configuration space. These configurations have mathematical identities like “A particle here, a particle there”, rather than “particle 1 here, particle 2 there” and the difference is experimentally testable. What might appear to be a little billiard ball, like an electron caught in a trap, is actually a multiplicative factor in a wavefunction that happens to approximately factor. The factorization of 18 includes two factors of 3, not one factor of 3, but this doesn’t mean the two 3s have separate individual identities—quantum mechanics is sort of like that. (If that didn’t make any sense to you, sorry; you need to have followed the series on quantum physics.)

In Identity Isn’t In Specific Atoms, we took this counterintuitive truth of physical ontology, and proceeded to kick hell out of an intuitive concept of personal identity that depends on being made of the “same atoms”—the intuition that you are the same person, if you are made out of the same pieces. But because the brain doesn’t repeat its exact state (let alone the whole universe), the joint configuration space which underlies you, is nonoverlapping from one fraction of a second to the next. Or even from one Planck interval to the next. I.e., “you” of now and “you” of one second later do not have in common any ontologically basic elements with a shared persistent identity.

Just from standard quantum mechanics, we can see immediately that some of the standard thought-experiments used to pump intuitions in philosophical discussions of identity, are physical nonsense. For example, there is a thought experiment that runs like this:

“The Scanner here on Earth will destroy my brain and body, while recording the exact states of all my cells. It will then transmit this information by radio. Travelling at the speed of light, the message will take three minutes to reach the Replicator on Mars. This will then create, out of new matter, a brain and body exactly like mine. It will be in this body that I shall wake up.”

This is Derek Parfit in the excellent Reasons and Persons, p. 199—note that Parfit is describing thought experiments, not necessarily endorsing them.

There is an argument which Parfit describes (but does not himself endorse), and which I have seen many people spontaneously invent, which says (not a quote):

Ah, but suppose an improved Scanner were invented, which scanned you non-destructively, but still transmitted the same information to Mars . Now, clearly, in this case, you, the original have simply stayed on Earth, and the person on Mars is only a copy. Therefore this teleporter is actually murder and birth, not travel at all—it destroys the original, and constructs a copy!

Well, but who says that if we build an exact copy of you, one version is the privileged original and the other is just a copy? Are you under the impression that one of these bodies is constructed out of the original atoms—that it has some kind of physical continuity the other does not possess? But there is no such thing as a particular atom, so the original-ness or new-ness of the person can’t depend on the original-ness or new-ness of the atoms.

(If you are now saying, “No, you can’t distinguish two electrons yet, but that doesn’t mean they’re the same entity—” then you have not been following the series on quantum mechanics, or you need to reread it. Physics does not work the way you think it does. There are no little billiard balls bouncing around down there.)

If you further realize that, as a matter of fact, you are splitting all the time due to ordinary decoherence, then you are much more likely to look at this thought experiment and say: “There is no copy; there are two originals.”

Intuitively, in your imagination, it might seem that one billiard ball stays in the same place on Earth, and another billiard ball has popped into place on Mars; so one is the “original”, and the other is the “copy”. But at a fundamental level, things are not made out of billiard balls.

A sentient brain constructed to atomic precision, and copied with atomic precision, could undergo a quantum evolution along with its “copy”, such that, afterward, there would exist no fact of the matter as to which of the two brains was the “original”. In some Feynman diagrams they would exchange places, in some Feynman diagrams not. The two entire brains would be, in aggregate, identical particles with no individual identities.

Parfit, having discussed the teleportation thought experiment, counters the intuitions of physical continuity with a different set of thought experiments:

“Consider another range of possible cases: the Physical Spectrum. These cases involve all of the different possible degrees of physical continuity...

“In a case close to the near end, scientists would replace 1% of the cells in my brain and body with exact duplicates. In the case in the middle of the spectrum, they would replace 50%. In a case near the far end, they would replace 99%, leaving only 1% of my original brain and body. At the far end, the ‘replacement’ would involve the complete destruction of my brain and body, and the creation out of new organic matter of a Replica of me.”

(Reasons and Persons, p. 234.)

Parfit uses this to argue against the intuition of physical continuity pumped by the first experiment: if your identity depends on physical continuity, where is the exact threshold at which you cease to be “you”?

By the way, although I’m criticizing Parfit’s reasoning here, I really liked Parfit’s discussion of personal identity. It really surprised me. I was expecting a rehash of the same arguments I’ve seen on transhumanist mailing lists over the last decade or more. Parfit gets much further than I’ve seen the mailing lists get. This is a sad verdict for the mailing lists. And as for Reasons and Persons, it well deserves its fame.

But although Parfit executed his arguments competently and with great philosophical skill, those two particular arguments (Parfit has lots more!) are doomed by physics.

There just is no such thing as “new organic matter” that has a persistent identity apart from “old organic matter”. No fact of the matter exists, as to which electron is which, in your body on Earth or your body on Mars. No fact of the matter exists, as to how many electrons in your body have been “replaced” or “left in the same place”. So both thought experiments are physical nonsense.

Parfit seems to be enunciating his own opinion here (not Devil’s advocating) when he says:

“There are two kinds of sameness, or identity. I and my Replica are qualitatively identical, or exactly alike. But we may not be numerically identical, one and the same person. Similarly, two white billiard balls are not numerically but may be qualitatively identical. If I paint one of these balls red, it will cease to be qualitatively identical with itself as it was. But the red ball that I later see and the white ball that I painted red are numerically identical. They are one and the same ball.” (p. 201.)

In the human imagination, the way we have evolved to imagine things, we can imagine two qualitatively identical billiard balls that have a further fact about them—their persistent identity—that makes them distinct.

But it seems to be a basic lesson of physics that “numerical identity” just does not exist. Where “qualitative identity” exists, you can set up quantum evolutions that refute the illusion of individuality—Feynman diagrams that sum over different permutations of the identicals.

We should always have been suspicious of “numerical identity”, since it was not experimentally detectable; but physics swoops in and drop-kicks the whole argument out the window.

Parfit p. 241:

“Reductionists admit that there is a difference between numerical identity and exact similarity. In some cases, there would be a real difference between some person’s being me, and his being someone else who is merely exactly like me.”

This reductionist admits no such thing.

Parfit even describes a wise-seeming reductionist refusal to answer questions as to when one person becomes another, when you are “replacing” the atoms inside them. P. 235:

(The reductionist says:) “The resulting person will be psychologically continuous with me as I am now. This is all there is to know. I do not know whether the resulting person will be me, or will be someone else who is merely exactly like me. But this is not, here, a real question, which must have an answer. It does not describe two different possibilities, one of which must be true. It is here an empty question. There is not a real difference here between the resulting person’s being me, and his being someone else. This is why, even though I do not know whether I am about to die, I know everything.”

Almost but not quite reductionist enough! When you master quantum mechanics, you see that, in the thought experiment where your atoms are being “replaced” in various quantities by “different” atoms, nothing whatsoever is actually happening—the thought experiment itself is physically empty.

So this reductionist, at least, triumphantly says—not, “It is an empty question; I know everything that there is to know, even though I don’t know if I will live or die”—but simply, “I will live; nothing happened.”

This whole episode is one of the main reasons why I hope that when I really understand matters such as these, and they have ceased to be mysteries unto me, that I will be able to give definite answers to questions that seem like they ought to have definite answers.

And it is a reason why I am suspicious, of philosophies that too early—before the dispelling of mystery—say, “There is no answer to the question.” Sometimes there is no answer, but then the absence of the answer comes with a shock of understanding, a click like thunder, that makes the question vanish in a puff of smoke. As opposed to a dull empty sort of feeling, as of being told to shut up and stop asking questions.

And another lesson: Though the thought experiment of having atoms “replaced” seems easy to imagine in the abstract, anyone knowing a fully detailed physical visualization would have immediately seen that the thought experiment was physical nonsense. Let zombie theorists take note!

Additional physics can shift our view of identity even further:

In Timeless Physics, we looked at a speculative, but even more beautiful view of quantum mechanics: We don’t need to suppose the amplitude distribution over the configuration space is changing, since the universe never repeats itself. We never see any particular joint configuration (of the whole universe) change amplitude from one time to another; from one time to another, the universe will have expanded. There is just a timeless amplitude distribution (aka wavefunction) over a configuration space that includes compressed configurations of the universe (early times) and expanded configurations of the universe (later times).

Then we will need to discover people and their identities embodied within a timeless set of relations between configurations that never repeat themselves, and never change from one time to another.

As we saw in Timeless Beauty, timeless physics is beautiful because it would make everything that exists either perfectly global—like the uniform, exceptionless laws of physics that apply everywhere and everywhen—or perfectly local—like points in the configuration space that only affect or are affected by their immediate local neighborhood. Everything that exists fundamentally, would be qualitatively unique: there would never be two fundamental entities that have the same properties but are not the same entity.

(Note: The you on Earth, and the you on Mars, are not ontologically basic. You are factors of a joint amplitude distribution that is ontologically basic. Suppose the integer 18 exists: the factorization of 18 will include two factors of 3, not one factor of 3. This does not mean that inside the Platonic integer 18 there are two little 3s hanging around with persistent identities, living in different houses.)

We also saw in Timeless Causality that the end of time is not necessarily the end of cause and effect; causality can be defined (and detected statistically!) without mentioning “time”. This is important because it preserves arguments about personal identity that rely on causal continuity rather than “physical continuity”.

Previously I drew this diagram of you in a timeless, branching universe:

Manybranches4

To understand many-worlds: The gold head only remembers the green heads, creating the illusion of a unique line through time, and the intuitive question, “Where does the line go next?” But it goes to both possible futures, and both possible futures will look back and see a single line through time. In many-worlds, there is no fact of the matter as to which future you personally will end up in. There is no copy; there are two originals.

To understand timeless physics: The heads are not popping in and out of existence as some Global Now sweeps forward. They are all just there, each thinking that now is a different time.

In Timeless Causality I drew this diagram:

Causeright

This was part of an illustration of how we could statistically distinguish left-flowing causality from right-flowing causality—an argument that cause and effect could be defined relationally, even the absence of a changing global time. And I said that, because we could keep cause and effect as the glue that binds configurations together, we could go on trying to identify experiences with computations embodied in flows of amplitude, rather than having to identify experiences with individual configurations.

But both diagrams have a common flaw: they show discrete nodes, connected by discrete arrows. In reality, physics is continuous.

So if you want to know “Where is the computation? Where is the experience?” my best guess would be to point to something like a directional braid:

Braid_2

This is not a braid of moving particles. This is a braid of interactions within close neighborhoods of timeless configuration space.

Braidslice

Every point intersected by the red line is unique as a mathematical entity; the points are not moving from one time to another. However, the amplitude at different points is related by physical laws; and there is a direction of causality to the relations.

You could say that the amplitude is flowing, in a river that never changes, but has a direction.

Embodied in this timeless flow are computations; within the computations, experiences. The experiences’ computations’ configurations might even overlap each other:

Braidtime_2

In the causal relations covered by the rectangle 1, there would be one moment of Now; in the causal relations covered by the rectangle 2, another moment of Now. There is a causal direction between them: 1 is the cause of 2, not the other way around. The rectangles overlap—though I really am not sure if I should be drawing them with overlap or not—because the computations are embodied in some of the same configurations. Or if not, there is still causal continuity because the end state of one computation is the start state of another.

But on an ontologically fundamental level, nothing with a persistent identity moves through time.

Even the braid itself is not ontologically fundamental; a human brain is a factor of a larger wavefunction that happens to factorize.

Then what is preserved from one time to another? On an ontologically basic level, absolutely nothing.

But you will recall that I earlier talked about any perturbation which does not disturb your internal narrative, almost certainly not being able to disturb whatever is the true cause of your saying “I think therefore I am”—this is why you can’t leave a person physically unaltered, and subtract their consciousness. When you look at a person on the level of organization of neurons firing, anything which does not disturb, or only infinitesimally disturbs, the pattern of neurons firing—such as flipping a switch from across the room—ought not to disturb your consciousness, or your personal identity.

If you were to describe the brain on the level of neurons and synapses, then this description of the factor of the wavefunction that is your brain, would have a very great deal in common, across different cross-sections of the braid. The pattern of synapses would be “almost the same”—that is, the description would come out almost the same—even though, on an ontologically basic level, nothing that exists fundamentally is held in common between them. The internal narrative goes on, and you can see it within the vastly higher-level view of the firing patterns in the connection of synapses. The computational pattern computes, “I think therefore I am”. The narrative says, today and tomorrow, “I am Eliezer Yudkowsky, I am a rationalist, and I have something to protect.” Even though, in the river that never flows, not a single drop of water is shared between one time and another.

If there’s any basis whatsoever to this notion of “continuity of consciousness”—I haven’t quite given up on it yet, because I don’t have anything better to cling to—then I would guess that this is how it works.

Oh… and I promised you a real-world application, didn’t I?

Well, here it is:

Many throughout time, tempted by the promise of immortality, have consumed strange and often fatal elixirs; they have tried to bargain with devils that failed to appear; and done many other silly things.

But like all superpowers, long-range life extension can only be acquired by seeing, with a shock, that some way of getting it is perfectly normal.

If you can see the moments of now braided into time, the causal dependencies of future states on past states, the high-level pattern of synapses and the internal narrative as a computation within it—if you can viscerally dispel the classical hallucination of a little billiard ball that is you, and see your nows strung out in the river that never flows—then you can see that signing up for cryonics, being vitrified in liquid nitrogen when you die, and having your brain nanotechnologically reconstructed fifty years later, is actually less of a change than going to sleep, dreaming, and forgetting your dreams when you wake up.

You should be able to see that, now, if you’ve followed through this whole series. You should be able to get it on a gut level—that being vitrified in liquid nitrogen for fifty years (around 3e52 Planck intervals) is not very different from waiting an average of 2e26 Planck intervals between neurons firing, on the generous assumption that there are a hundred trillion synapses firing a thousand times per second. You should be able to see that there is nothing preserved from one night’s sleep to the morning’s waking, which cryonic suspension does not preserve also. Assuming the vitrification technology is good enough for a sufficiently powerful Bayesian superintelligence to look at your frozen brain, and figure out “who you were” to the same resolution that your morning’s waking self resembles the person who went to sleep that night.

Do you know what it takes to securely erase a computer’s hard drive? Writing it over with all zeroes isn’t enough. Writing it over with all zeroes, then all ones, then a random pattern, isn’t enough. Someone with the right tools can still examine the final state of a section of magnetic memory, and distinguish the state, “This was a 1 written over by a 1, then a 0, then a 1” from “This was a 0 written over by a 1, then a 0, then a 1″. The best way to securely erase a computer’s hard drive is to destroy it with thermite.

I really don’t think that carefully vitrifying a brain to prevent ice crystal formation and then freezing it in liquid nitrogen is going to be a secure erase procedure, if you can examine atomic-level differences in the synapses.

Someone hears about cryonics and thinks for 10 seconds and says, “But if you’re frozen and then revived, are you really the same person?”

And if they happened to know all about quantum physics and could apply the abstract knowledge to real life, and they had followed the whole debate about zombies and resolved it against epiphenomenalism in general, then they would be able to visualize the braids in the river that never flows, and say, “Yes.”

But this knowledge is not common.

So they die.

There are numerous other reasons that people seize on, when they search for a rationalization for a negative initial flinch against cryonics. And numerous other knowledges that would be required to answer those objections. “But wouldn’t it be boring to live such a long time?” (Can be answered if you know hedonic psychology, and have developed a theory of fun, and can visualize accessible fun spaces that increase in volume with increasing intelligence.) “Why would future civilizations bother to revive me?” (Requires understanding either economic growth diminishing the cost, or knowledge of history and how societies have become kinder over time, or knowing about Friendly AI.) “Isn’t it wrong to live so long?” (Requires knowing about the “sour grapes” bias. See also transhumanism as simplified humanism and the meaning that immortality gives to life.) Then there’s the meta-knowledge of how to question all these deeply wise cached thoughts that pop into your head about the futility of life; and the ability to do things that might make people look at you weird, and so on...

Some of these are series of posts I haven’t done yet. But if you anticipate updating your probabilities when you read those future posts, then you should update them now. Or, if you prefer, trust me:

If you would rather live happily ever after, than die, and you are willing to spend between $300 and $2000 per year(*) to express this preference, then sign up for cryonics.

If you’ve been cryocrastinating, putting off signing up for cryonics “until later”, don’t think that you’ve “gotten away with it so far”. Many worlds, remember? There are branched versions of you that are dying of cancer, and not signed up for cryonics, and it’s too late for them to get life insurance.

See, knowing about many worlds can help you visualize probabilities as frequencies, because they usually are.

It might encourage you to get around to getting health insurance, too, or wearing a helmet on your motorcycle, or whatever: don’t think you’ve gotten away with it so far.

And if you’re planning to play the lottery, don’t think you might win this time. A vanishingly small fraction of you wins, every time. So either learn to discount small fractions of the future by shutting up and multiplying, or spend all your money on lottery tickets—your call.

It is a very important lesson in rationality, that at any time, the Environment may suddenly ask you almost any question, which requires you to draw on 7 different fields of knowledge. If you missed studying a single one of them, you may suffer arbitrarily large penalties up to and including capital punishment. You can die for an answer you gave in 10 seconds, without realizing that a field of knowledge existed of which you were ignorant.

This is why there is a virtue of scholarship.

150,000 people die every day. Some of those deaths are truly unavoidable, but most are the result of inadequate knowledge of cognitive biases, advanced futurism, and quantum mechanics.(**)

If you disagree with my premises or my conclusion, take a moment to consider nonetheless, that the very existence of an argument about life-or-death stakes, whatever position you take in that argument, constitutes a sufficient lesson on the sudden relevance of scholarship.

(*) The way cryonics works is that you get a life insurance policy, and the policy pays for your cryonic suspension. The Cryonics Institute is the cheapest provider, Alcor is the high-class one. Rudi Hoffman set up my own insurance policy, with CI. I have no affiliate agreements with any of these entities, nor, to my knowledge, do they have affiliate agreements with anyone. They’re trying to look respectable, and so they rely on altruism and word-of-mouth to grow, instead of paid salespeople. So there’s a vastly smaller worldwide market for immortality than lung-cancer-in-a-stick. Welcome to your Earth; it’s going to stay this way until you fix it.

(**) Most deaths? Yes: If cryonics were widely seen in the same terms as any other medical procedure, economies of scale would considerably diminish the cost; it would be applied routinely in hospitals; and foreign aid would enable it to be applied even in poor countries. So children in Africa are dying because citizens and politicians and philanthropists in the First World don’t have a gut-level understanding of quantum mechanics.

Added: For some of the questions that are being asked, see Alcor’s FAQ for scientists and Ben Best’s Cryonics FAQ (archived snapshot).

Part of The Quantum Physics Sequence

Next post: “Thou Art Physics”

Previous post: “Timeless Causality”

Eliezer Yudkowsky3 Jun 2008 8:16 UTC
61 points
249 comments14 min readLW linkArchive