Re: Rationality, Miracles and ETI

Dan Fabulich (
Wed, 3 Nov 1999 00:37:26 -0500 (EST)

'What is your name?' 'Eliezer S. Yudkowsky.' 'IT DOESN'T MATTER WHAT YOUR NAME IS!!!':

> The internalist theory of proof is false. Given time and money, it
> would actually be fairly easy to set up a situation where the most
> rational explanation is the false one, a situation where Occam's Razor
> doesn't work.

OK, gotcha. Now that I better understand what you're saying, I realize that not only is this not a refutation of internalism, Putnam covers a case *exactly* like this in his book. I'll quote, since I don't feel inspired this evening.

"To reject the idea that there is a coherent 'external' perspective, a theory which is simply true 'in itself', apart from all possible observers, is not to *identify* truth with rational acceptability. Truth cannot simply *be* rational acceptability for one fundamental reason; truth is supposed to be a property of a statement that cannot be lost, whereas justification can be lost. Thee statement 'The earth is flat' was, very likely, rationally acceptable 3,000 years ago; but it is not rationally acceptable today. Yet it would be wrong to say that 'the earth is flat' was *true* 3,000 years ago; for that would mean that the earth has changed its shape. ... "

"What this shows, in my opinion is not that the externalist [here, he means metaphysical realist] view is right after all, but that truth is an *idealization* of rational acceptability. We speak as if there were such things as epistemically ideal conditions, and we call a statement 'true' if it would be justified under such conditions. 'Epistemically ideal conditions' of course, are like 'frictionless planes': we cannot really attain epistemically ideal conditions, or even be absolutely certain that we have come sufficiently close to them. But frictionless planes cannot really be attained either, and yet talk of frictionless planes has 'cash value' because we can approximate them to a very high degree fo approximation."

In your example, in which you have a windfall at the same time as you perform an act of charity, it is rationally acceptable to believe that the windfall was a coincidence. However, you're right, it could well be the case that the windfall was NOT a concidence. If we accept internalism, then all this means is that, under ideal epistemic conditions, (for example, you being aware of this fellow who tries to reward people for acts of charity,) it would NOT be rationally acceptable to believe that it was a coincidence.

According to my current favorite model of my experiences, "ideal epistemic conditions" is the situation in which you know everything you can know about the "real world." In this case, the words "real world" refer to a theoretical construct in my mind, which in turn is part of a theory which, I find, correctly explains/predicts my experiences.

Just as it WAS rationally acceptable to believe that the earth was flat back when nobody knew better, so in the imagined scenario would it be reasonable to believe that your finding the book was sheer coincidence. This does not mean that the above claims are true, even under an internalist "idealized justification" view.

Hal pointed out that I may have misapplied the doctrine in the case of the Matrix Hypothesis. My whole point was to show that if you would act as if the Matrix Hypothesis were false, even if some proof of it were presented to you, then the Matrix Hypothesis would be the sort of Hypothesis (and there are very few of these, I think!) which are not rationally acceptable under any circumstances. Some examples of hypotheses like these are: "I do not exist," "the real world does not exist," "my actions have no impact on the real world or on myself," etc. If the Matrix Hypothesis is not acceptable under any circumstances, then it is not acceptable under ideal epistemic conditions, and thus it is false.

(I feel quite certain that the claim: "I am living in a Matrix world, but the creators of the Matrix have never intervened and never will" is false in the manner I've described.)

BTW, I'm almost inclined to say that you can't HELP but install a world-view like this into an AI, since the anti-realist/internalist theory of reference is EASILY explained to an AI. It's so easy, in fact, that you may accidentally wind up coding your AI in an internalist way when you had MEANT to code it to be a metaphysical realist!

Under anti-realism, the internal symbol "ball" simply refers to the internal theoretical construct of a ball. Any theory which incorporates this term must correctly explain/predict the AI's experiences, which are *also* internal. Adopting a realist approach would require you to CODE IN an idea of the real world, and explain to it how the internal symbols refer to the real world out there. Unfortunately, this approach, as I've argued, must necessarily fail, unless you're going to code in a "because I say so!" theory of reference. And, of course, if you accidentally code up the words "real world" to refer to the AI's internal theoretical construct of the "real world," then you've accidentally coded an anti-realist AI. As Putnam put it: "Since the objects [of reference] *and* the signs are both internal to the scheme of description, it is possible to say what matches what. Indeed, it is trivial to say what any word refers to *within* the language the word belongs to, by using the word itself. What does 'rabbit' refer to? Why, to rabbits, of course! What does 'extraterrestrial' refer to? To extraterrestrials (if there are any)."

I presented an example similar to this when I said that the character 7 simply referred to that theoretical invention of ours, the number seven. So with the real world, balls, horses, etc. Each of these words refers to internal theoretical constructs; we judge the theories based on their explanatory/predictive value.


-unless you love someone-
-nothing else makes any sense-

e.e. cummings