Monday, March 21, 2005

"No child of mine will ever cower before an imaginary God. It is beneath the dignity of human beings and it is beneath the dignity of our descendants. If the lightning is beautiful, then let us see the beauty in electricity without need for thunder deities; for if we cannot learn to take joy in the merely real, our lives will be empty indeed."

~Eliezer Yudkowsky


One great fear of mine is the tyrrany of sounding good. Eliezer is a prime example of a person who has tuned and tuned in the search for rationality. Unfortunately, I suspect, there is a class of 'improvements' one can make which correspond more to obscuring flaws than making true statements. It is a problem I have noted myself on more than one occasion. Here Eliezer is responding to a theist who complained that we need to instill our mind children with some kind of religion that will force them to respect their elders, lest we be obsoleted. He continues:

"But I'm not going to try to hardcode that, not in a child nor in an AI. As an atheist, I have a simple, matter-of-fact confidence that religionists once had and relinquished long ago. I don't think I need to load the dice for my answer to win. All I need is to set in motion the dynamics that seek truth, i.e., some computable approximation of Solomonoff induction. If there were the tiniest shred of truth to religion, that would be enough to uncover it. If you have even a droplet of honest belief left, not just empty excuses for a faith you lost long ago, you will not ask me to load an AI's dice in favor of your pet theory. Let the truth win out."

All true statements. All very appealing (at least to this rationalist and this truthseeker). But, the subtle shift in conversation here is quite nearly unnoticed. We've transitioned to instilling beliefs in a mind, to better them and ourselves, to talking about the structure of the mind, to fixing it so there is only one answer. Perhaps because the theist is muddled in his thinking this blanket approach is valid. It's true that Eliezer's objections do entirely refute John C Wright's theistic aspirations. But his argument does not directly address his points.

A general question: What is intellectual honesty? Eliezer has a real commitment to truth. However, and I fear this is a general point, being committed to truth is not sufficient. Eliezer in this example, and others in many examples (I choose Eliezer because I believe he's not making any other mistakes here) has changed to context, the discussion has been shifted to allow him his total commitment to certainty. By changing the context slightly he's found a place where he can shoot down this theistic argument with perfect aplomb and sound like a hero. But is he? He's making arguments that are true, and insightful(even poetic, perhaps) but they aren't in the original exact vein of discussion. Isn't that somewhat misleading? Or am I making something of a molehill? Perhaps Eliezer has simply reframed the question in general terms, much as I'm generalizing his statements for logical effect.

Let's continue in that vein, and move reducto ad absurdum. Suppose a fully rational, truthful being, that only chooses to engage in discussion when certain, and always seeks to twist contexts to those he's more comfortable in, to the limits of his self respect and intellectual honesty. Luckily, mythology is replete with examples of this type. The Zen Master, the Oracle, Yoda, all inscrutable characters who are right, and insightful, and powerful creatures, but maddening, because they only rouse themselves to croak factually accurate and unassailable arguments, and refuse to engage in fringe discussion.

There are two factors here. One is the very real problem of authority acceptance. Many self-aware Masters rage at their disciples on both sides, chiding them for accepting the Master's word without question, and also being annoyed when they don't recognize and internalize the truth the Master offers them. So the good master retreats into relative silence to avoid corrupting and doing a disservice to all those who listen to him. Speaking when certain, and able to tell how his words will affect. This admirable strategy is always blended with persona maintenence, a despicable practice of hiding, changing, and sculpting information to maintain certain relationships and reputations. Shame on the Master who can't bear to have students see him wrong.

The second factor is subtler, and the one I have been trying to explore above. The Master categorizes within his subject. He divides the realm of his understanding by function or taxonomy, he asks questions and answers with statements which exist along those lines. The Fool asks sweeping, conjoined questions, stabs at understanding that smears across the subject. The Master, presented with these questions, maps the question to his understanding, finding pieces of it within some division of his knowledge, addresses this part (perhaps rightly) believing himself to have dealt with the entirety. After all, a single contradiction is all you need to invalidate an entire argument.

I don't know whether this represents an important distinction when you're just learning to be a rationalist. I haven't yet reached the point where it even constitutes a significant portion of my mistakes. But it is A mistake.

I know there is a rank above that of Inscrutable Master. I don't have all the details yet, but he's humble and detailed and truthful. She answers questions in the spirit they are asked, but as correctly as she knows how. He presents his uncertainty, his incomplete scraps of knowledge, and his current thinking, because it too, is information. And she never takes an easy win, when there are more interesting and informative portions of an argument. Noisy Errors are to be preferred.

No comments: