Open Thread, November 1 - 7, 2013

witzvo

If it's worth saying, but not worth its own post (even in Discussion), then it goes here.

I'm reminded of many years ago, a coworker coming into my office and asking me a question about the design of a feature that interacts with our tax calculation.

So she and I created this whole whiteboard flowchart working out the design, at the end of which I said "Hrm. So, at a high level, this seems OK. That said, you should definitely talk to Mark about this, because Mark knows a lot more about the tax code than I do, and he might see problems I missed. For example, Mark will probably notice that this bit here will fail when $condition applies, which I... um... completely failed to notice?"

I could certainly describe that as having a "Mark" in my head who is smarter about tax-code-related designs than I am, and there's nothing intrinsically wrong with describing it that way if that makes me more comfortable or provides some other benefit.

But "Mark" in this case would just be pointing to a subset of "Dave", just as "Dave's fantasies about aliens" does.

See also 'rubberducking' and previous discussions of this on LW. My basic theory is that reasoning was developed for adversarial purposes, and by rubberducking you are essentially roleplaying as an 'adversary' which triggers deeper processing (if we ever get brain imaging of system I vs system II thinking, I'd expect that adversarial thinking triggers system II more compared to 'normal' self-centered thinking).

1Armok_GoB12y

Yea in that case presumably the tulpa would help - but not necessarily significantly more than such a non-tulpa model that requires considerably less work and risk. Basically, a tulpa can technically do almost anything you can... but the absence of a tulpa can do them to, and for almost all of them there's some much easier and at least as effective way to do the same thing.