Alan_Crowe comments on Trying to Try - Less Wrong

42 Post author: Eliezer_Yudkowsky 01 October 2008 08:58AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (53)

Sort By: Old

You are viewing a single comment's thread.

Comment author: Alan_Crowe 01 October 2008 04:40:53PM 2 points [-]

Wearing my mechanical engineer's hat I say "Don't be heavy handed.". Set your over-force trips low. When the switch is hard to flip or the mechanism is reluctant to operate, fail and signal the default over-force exception.

You can always wiggle it, or lubricate it and try again, provided you haven't forced it and broken it. For me, trying is about running the compiler with the switches set to retain debugging information and running the code in verbose mode. It is about setting up a receiver down-range. Maybe the second rocket will blow up, just like the first did, but at least I will still be recording the telemetry.

I think that Plan A will be stymied by Problem Y, but I try it anyway, before I try to solve Problem Y. My optimistic side is hoping Problem Y might not actually matter, while my pessimistic side thinks Problem X is lurking in the shadows, ready to emerge and kill Plan A whether I solve Problem Y or not.

I *try* in order to gain information.

It is usually important to procede with confidence. When things go wrong they throw off fragments of broken machinery and fragments of information. Suprised, we fail to catch the flying fragments of information, and must try again, forewarned.

Two meanings of the word "try" fight for mind share.

To try: to position oneself in the right spot to catch the flying fragments of information flung out from failure.

To try: The psychological mechanism that lets us fail through faint-heartedness, again and again, but never quite understand why.

Two meanings sharing a word is a common problem with natural language. The particular danger I see for Eliezer is when the second meaning hides the first.

He says he isn't ready to write code. If you don't try to code up a general artificial intelligence you don't succeed, but you don't fail either. So you can't fail earlier and harder than you ever expected and cannot suspect that the singular is far. If you won't try, you'll never know.