(Untitled)

[maerdi]

I'm feeling a bit lonely and sad today. Does anyone have anything interesting they'd like to talk about?

Some possible suggestions:

• the AI in a box problem from one of patrissimo's recent posts.
• what causes dishwashers to be less effective and how one can improve their performance
• Fault Lines
• working alone versus working with a partner versus working as part

Comments

[maerdi]

I think the gatekeeper is pretending he really is the gatekeeper, though the rules do allow him to step out of character, which I find odd. You might like this write-up from the 'original' practitioner: http://www.overcomingbias.com/2008/10/shut-up-and-do.html

In particular, I found it interesting that he says he just won it the hard way (not by tricks).

[dragonmudd]

I think there's an important distinction that was implied but not given any attention in the original problem. The problem is stated as "could outsmart me" not "could outsmart someone". I am certain that it's much easier to write an AI such that there exists a person that the AI could outsmart and trick.

[thegreatgonz]

This may not be in the spirit of the AI-in-a-box problem, but I'm not at all convinced that you can create a human-level or better AI without allowing it to interact with the world; I suspect that interaction, and learning by doing, are fundamental to the process of developing intelligence, at least in any form that we would recognize as such.

[pmb]

You are in agreement with a past chair of MIT CSAIL in that regard.

[maerdi]

You might well be right. I was mostly thinking of what tactics would work best for the AI.

That said, what about unlimited contact with the AI's programmers? They can even interact with the world and report back, if you want.

[thistleingrey]

Depending on the partner, partner scenario wins, for me. It needn't be someone I get along with well or feel especially close to, but the feeding back and forth of ideas is vital. (I edit better than I create.) I've noticed that in groups, I end up trying to gain a partner anyway--subgroup of two?--if it's a group I'm not leading offically.

The broader question re: performance depends somewhat on whether one tends to hijack situations that seem temporarily directionless, I suspect, and on how well said hijacking tends to work out. ;)

[maerdi]

I'm very much the same. I need my own time in any project, but I work much better in a pair than strictly on my own or as part of a larger group. Too many people and I start to lose my own thoughts to managing and keeping track of the group, even if there are only 3 or 4 in the group. On the flip side, I'm not nearly as powerful on my own as I am with a partner I have some respect for and can talk to openly. In fact, depending on the project, I prefer not to be too close to my partner emotionally so that disagreements never feel personal and we can argue both sides to the best of our abilities with joy and a total lack of anger or hurt.

[mtbg]

I have all sorts of disorganized thoughts about this. Depending on how much I feel like procrastinating tomorrow, I may post something about it :)

[maerdi]

I guess not-procrastinating won out, but maybe you'll have some time to come back to it. I find it a really interesting subject, though I'm amazed at how long it took me to put words to my own experiences and lessons.

[zane314]

I've done some thinking about the AI-in-a-box problem and I've come up with some lines of attack that I believe could convince some people.

They wouldn't convince me yet, though.

[dragonmudd]

I suspect you will be unable to come up with a solution that would convince you. My reasoning is that I suspect that once you find one that might, you will file that away as "this is another line of attack, and I want to win, so I simply recognize it as a potential weapon." It's a case of knowing something might have power takes away its power. I don't think that's necessarily true of all people, but I suspect it's true of you, due to how you approach problems like this and you level of interest in puzzles and games.

[zane314]

I think that the best case I can hope for is coming up with a line of logic that ends with, "Huh. That would convince me. I am not the kind of person who could keep an AI in the box, and thus if I am ever in the situation, I will excuse myself from the selection committee

[maerdi]

This doesn't quite convince me yet, because given worst case scenarios this is me weighing the rights of an AI vs the rights of the human race to exist and even if the AI is capable of being stronger, better, more understanding, etc. than the entire human race, it's still my duty to pick in favor of the humans, but I know it's a line of attack I have vulnerabilities in.

Have you ever read "The Ones Who Walk Away From Omelas"?

Thanks for sharing-these particular attacks are good ones. And I think in the right conditions, I might have moments of weakness where these might make some headway. I can't imagine that happening in 2 hours, though, in part because I'd have had to come to terms with the first (at least) before I ever stepped in.

[porfinn]

One ticket (with unlimited uses) issued to Lara to call if she ever feels "lonely" or "sad".

Although I can be a tad bit overbearing, bombastic, and generally annoying, I have been told that I am very distracting and good at spreading good cheer as if it were too much peanut butter on bread. You are always welcome to call.

This is almost a direct quote:
"We had a funeral today, and needed fun, so here we are. Use your powers, and amuse us!"

Said by some friends visiting the area because of his father's demise. And, variations on that theme are numerous. My number is on facebook, take advantage of it whenever you like.