PSYCH: Re: Emotion Selection

Freeman Craig Presson (dhr@iname.com)
Sun, 4 Jul 1999 12:22:56 -0500

On 3 Jul 99, Anders wrote:
[...]
> What is needed is to balance the emotions, which provide us with
> motivation and creates a mode of thinking, with rational evaluation of
> situations. Just like we need to learn how to learn, think rationally and
> act, we need to learn how to manage our emotions. They work great in
> certain respects, but not all.

This and the suggestion of refining emotions (so that, by analogy, you get a specific warning message instead of "General Protection Fault" :-) sound really useful to me.

Meditation, again, is useful for the former. I haven't been doing it much lately, but when it was a regular habit, I had the experience several times of sitting down to meditate when I was emotionally stirred up. Eventually, I would get to a place where I had a calm, still perspective, but could still see my stirred-up "stuff." It would be as if, instead of having the stuff inside my body-mind, I was holding it on my lap. Once you've done this, by whatever method, it becomes possible to do it at will, at least until something comes along to kick you into an even deeper morass of emotions.

Actually, there are two general flavors of meditation: stillness and insight. I am talking above about the effect of stillness, and insight "should" be helpful in the refining process. but I suspect it really only gets us to the point of correctly identifying the source of whatever emotion is "up" at the time, rather than necessarily developing more acute ones at a low level.

I'd love to know about working shortcuts here, since for most people, enough meditation practice to be useful for this amounts to years.

Some formulas I have encountered and partially tested are:

Anders again, different post:
> (Hmm, what are the proper english words for these two states? In Swedish it
> is likely "avund" and "missunsamhet" - trust the Swedes to have subtle
> nuances for envy :-)
I think the main english word for "missunsamhet" is "covetousness," from "to covet" as in "Thou shalt not covet thy neighbor's wife's ass."

Greg Burch wrote:

> Almost all emotions serve vital functions in mental life. They serve as
> motors and gauges. I think transhumanists should aim to refine our
> emotions, but not to do away with them. (I also think that the idea of a
> truly sentient AI without emotions is like imagining an automobile without
> a motor or instruments.)

I have a simple example germane to the parenthetical above. "To be able to play chess is a sign of a fine mind. To be able to play it well is a sign of a misspent youth." There were no human-strength chess computers during my misspent youth, but as they became available and I played against them, I noticed that I would lose more games than I "should" to them (when I was doing a lot of this, I was at USCF 2000 strength, +/- 50 points, and the highest ratings claimed by commercial computers were around 2100).

Introspection provided a compelling answer: I wasn't treating the process of playing chess with a computer as a real game. The reason? *The computer didn't give a damn whether it won or lost.* I see a chess game as a contest between two sentients; me sitting at my PC moving pieces on a virtual board connected to, say, gnuchess, _looks_ like a game of chess, but doesn't have any contest aspect. I can use exactly the same interface to connect to a chess server and play a human; in that case, there is a contest, we both "care" about the outcome (or agree to for the moment) and indeed, I play up to my (declining) abilities in that case.

It seems like even Kasparov had trouble taking Deep Blue seriously at first, but he learned quickly. He, of course, was playing publicly and had the computer's developers very much in the game, and besides, he has a bigger ego than I do :-)

My example addresses roughly half of Greg's comment, actually. I would say that the lack of expressions of emotion would be one way an AI could fail a Turing test, just as it makes gnuchess fail my chess-opponent Turing test (gnuchess doesn't talk back at all, so it's clearly not trying to pass the test); and OTOH, giving the AI some basic drives and a sense of self may well be necessary for it to be able to formulate its own goals and act on them (nothing new here, really).