Demands for Particular Proof: Appendices


26


Eliezer_Yudkowsky

Appendices toYou're Entitled to Arguments, But Not (That Particular) Proof

(The main article was getting long, so I decided to move the appendices to a separate article which wouldn't be promoted, thus minimizing the size of the article landing in a promoted-article-only-reader's feed.)

A.  The absence of unobtainable proof is not even weak evidence of absence.

The wise will already know that absence of evidence actually is evidence of absence; and they may ask, "Since a time-lapse video record of apes evolving into humans would, in fact, be strong evidence in favor of the theory of evolution, is it not mandated by the laws of probability theory that the absence of this videotape constitute some degree of evidence against the theory of evolution?"

(Before you reject that proposition out of hand for containing the substring "evidence against the theory of evolution", bear in mind that grownups understand that evidence accumulates.  You don't get to pick out just one piece of evidence and ignore all the rest; true hypotheses can easily generate a minority of weak pieces of evidence against themselves; conceding one point of evidence does not mean conceding the debate; and people who try to act as if it does are nitwits.  Also there are probably no creationists reading this blog.)

The laws of probability theory do mandate that if P(H|E) > P(H), then P(H|~E) < P(H).  So - even if absence of proof is by no means proof of absence, and even if we reject the philosophy that absence of a particular proof means you get to discard all the other arguments about evidence and priors - must we not at least concede that absence of proof is necessarily evidence of absence, even though it may be very weak evidence?

Actually, in cases like creationism, not even that much follows.  Suppose we had a time camera - a device that lets us look into the distant past and see historical events with our own eyes.  Let the proposition "we have a time camera" be labeled Camera.  Then we would either be able to videotape the evolution of humans from apes; let the presence of this video record be labeled Video, and its absence ~Video.  Let Evolution stand for the hypothesis that evolution is true.  And let True and False stand for the epistemic states "Pretty much likely to be true" and "Pretty much likely to be false", respectively.

Then, given that evolution is true and that we have a time camera, we should expect to see Video:

P(Video|Evolution,Camera) = True   and   P(Video|~Evolution,Camera) = False

So if we had a time camera, if we could look into the past, then "no one has seen apes evolving into humans" would be strong evidence against the theory of natural selection:

P(Evolution|~Video,Camera) = False

But if we don't have a time camera, then regardless of whether evolution is true or false, we can't expect to have "seen apes evolving into humans":

P(Video|Evolution,~Camera) = False   and   P(Video|~Evolution,~Camera) = False

From which it follows that once you know ~Camera, observing ~Video tells you nothing further about Evolution:

P(Evolution|~Video,~Camera) = P(Evolution|~Camera)

If you didn't know whether or not we had time cameras, and I told you only that no one had ever seen apes evolving into humans, then you would have to evaluate P(Evolution|~Video) which includes some contributions from both P(Evolution|~Video,Camera) and P(Evolution|~Video,~Camera).  You don't know whether the video is missing because evolution is false, or the video is missing because we don't have a time camera.  And so, as the laws of probability require, P(Evolution|~Video) < P(Evolution) just as P(Evolution|Video) > P(Evolution).  But once you know you don't have a time camera, you can start by evaluating P(Evolution|~Camera) - and it's hard to see why lack of time cameras should, in and of itself, be evidence against evolution.  The process of natural selection hardly requires a time camera.  So P(Evolution|~Camera) = P(Evolution), and then ~Video isn't evidence one way or another.  The observation ~Camera screens off any evidence from observing ~Video.

And this is only what should be expected: once you know you don't have a time camera, and once you've updated your views on evolution in light of the fact that time cameras don't exist to begin with (which doesn't seem to have much of an impact on the matter of evolution), it makes no further difference when you learn that no one has ever witnessed apes evolving into humans.


 

B.  Subtext on cryonics:  Demanding that cryonicists produce a successful revival before you'll credit the possibility of cryonics, is logically rude; specifically, it is a demand for particular proof.

A successful cryonics revival performed with modern-day technology is not a piece of evidence you could possibly expect modern cryonicists to provide, even given that the proposition of interest is true.  The whole point of cryonics is as an ambulance ride to the future; to take advantage of the asymmetry between the technology needed to successfully preserve a patient (cryoprotectants, liquid nitrogen storage) and the technology needed to revive a patient (probably molecular nanotechnology).

In particular, the screening-off condition (playing the role of ~Camera in the example above) is the observation that we presently lack molecular nanotechnology.  Given that you don't currently have molecular nanotechnology, you can't reasonably expect to revive a cryonics patient today even given that they could in fact be revived using future molecular nanotechnology.

You are entitled to arguments, though not that particular proof, and cryonicists have done their best to provide you with whatever evidence can be obtained.  For example:

A study on rat hippocampal slices showed that it is possible for vitrified slices cooled to a solid state at -130ºC to have viability upon re-warming comparable to that of control slices that had not been vitrified or cryopreserved. Ultrastructure of the CA1 region (the region of the brain most vulnerable to ischemic damage) of the re-warmed slices is seen to be quite well preserved compared to the ultrastructure of control CA1 tissue (24). Cryonics organizations perfuse brains with vitrification solution until saturation is achieved...
A rabbit kidney has been vitrified, cooled to -135ºC, re-warmed and transplanted into a rabbit. The formerly vitrified transplant functioned well enough as the sole kidney to keep the rabbit alive indefinitely (25)... The vitrification mixture used in preserving the rabbit kidney is known as M22. M22 is used by the cryonics organization Alcor for vitrifying cryonics subjects. Perfusion of rabbits with M22 has been shown to preserve brain ultrastructure without ice formation (26).

This is the sort of evidence we can reasonably expect to obtain today, and it took work to provide you with that evidence.  Ignoring it in favor of demanding proof that you couldn't expect to see even if cryonicists were right, is (a) invalid as probability theory, (b) a sign of trying to defend an allowed belief rather than being honestly curious about which possible world we live in, and (c) logically rude.

Formally:

  1. Even given that the proposition put forth by cryonicists is true - that people suspended with modern-day technology will be revivable by future technology - you cannot expect them to revive a cryonics patient using modern-day technology.
  2. Cryonicists have put forth considerable effort, requiring years of work by many people, to provide you with such evidence as can be obtained today.  The lack of that particular proof is not owing to any defect of diligence on the part of cryonicists, or disinterest in their part on doing the research.
  3. The prediction that a properly cryoprotected patient does not suffer information-theoretical death is not a privileged hypothesis pulled out of nowhere; it is the default extrapolation from modern neuroscience.  If we learn that a patient cryoprotected using current technologies has undergone erasure of critical brain information, we have learned something that is not in current neuroscience textbooks - and actually rather surprising, all things considered.  The straight-line extrapolation from the science we do know is that if you can see the neurons nicely preserved under a microscope, the information sure ought to be there.  (The idea that critical brain information is stored dynamically in spiking patterns has already been contraindicated by the evidence; dogs taken to very low (above-freezing) temperatures, sufficient to suppress brain activity, do not seem to suffer any memory loss or personality change.)
  4. Given that the proposition of interest is true, there is something drastically urgent we ought to be doing RIGHT NOW, namely cryopreserving as many as possible of the 150,000 humans per day who undergo mind-state annihilation.  (Economies of scale would very likely drive down costs by an order of magnitude or more; this is an entirely feasible goal economically and technologically, the only question is the political will.)

Given these points: to discard the straight-line extrapolation from modern science and all the hard work that cryonicists have done to provide further distinguishing evidence, in favor of a demand for particular proof that you know cannot possibly be obtained and which you couldn't expect to see even given that the underlying proposition is true, when there are things we ought to be doing NOW given the truth of the proposition and much value will be lost by waiting; all this is indefensible as decision theory in a formal sense, and is, in an informal sense, madness.

Which all goes to say only what Xiaoguang "Mike" Li observed to me some time ago:  That saying you'll only sign up for cryonics when someone demonstrates a successful revival of a cryonics patient is sort of like saying that you won't get on the airplane until after it arrives at the destination.  Only a very small amount of common sense is necessary to see this, and the objection really does demonstrate the degree to which, when most people feel an innate flinch away from an idea, they hardly feel obligated to come up with objections that make the slightest bit of sense.

This beautiful public service announcement, with only a slight change of metaphor, could serve as a PSA for cryonics.  Stop making a big deal out of the decision.  It's not that complicated.


 

C.  Demanding the demonstration of a working nanomachine before you'll credit the possibility of molecular nanotechnology is logically rude, specifically, a demand for particular proof.

Given humanity's current level of technology, you can't reasonably expect a demonstration of molecular nanotechnology right now, even given that the proposition of interest is true: that molecular nanotechnology is physically possible to operate, physically possible to manufacture, and likely to be developed within some number of decades.  Even if we live in that world, you can't necessarily expect to see nanotechnology now.  And yet nonetheless the advocates of nanotechnology have gone to whatever extent possible to provide the arguments by which you could, today, figure out whether or not you live in a world where molecular nanotechnology is possible.  Eric Drexler put forth around six years of hard work to produce Nanosystems, doing as much of the basic physics as one man working more or less alone could be expected to do; and since then Robert Freitas, in particular, has been designing and simulating molecular devices, and even trying to work out simple synthesis paths, which is about as much as one person could do, and the funding hasn't really been provided for more than that.  To ignore all this hard work that has been put into providing you with such observations and arguments as can be reasonably obtained, and throw them out the window because of a demand for particular proof that you think they can't obtain and that they wouldn't be able to obtain even if the proposition at hand is true - this is not just invalid as probability theory, not just defensiveness rather than curiosity, it is logically rude.

Although actually, of course, you can see tiny molecularly-precise machines on parade, including freely rotating gears and general assemblers - just look inside a biological cell, particularly at ATP synthase and the ribosome.  But by that power of cognitive chaos which generates the demand for unobtainable proof in the first place, there can be little doubt that, as soon as this overlooked demonstration is pointed out, the one will immediately find some clause by which to exclude it.  To actually provide the unobtainable proof would hardly be fair, after all.



D.  It is invalid as probability theory, suboptimal as decision theory, and generally insane, to insist that you want to see someone develop a full-blown Artificial General Intelligence before you'll credit that it's worth anyone's time to work on problems of Friendly AI.

Not precisely analogous to the above cases, but it is a demand for particular proof.  Delineating the specifics is left as an exercise to the reader.