Re: Suicide by committee (was: How hard a Singularity?)

From: Eliezer S. Yudkowsky (
Date: Thu Jun 27 2002 - 13:38:04 MDT

James Higgins wrote:
> God damn it. Do you speak English? Do you actually READ what I type?
> How many times do I have to say that NO ONE want's to turn the problem
> over to committee! As Ben put it this would be an advisory board.
> What's, your telling us neither yourself nor anyone working on such
> things could even benefit from advice from your peers?

Ahem. Do YOU actually read what you type?

James Higgins wrote:
> Ideally, I think deployment (kick off) of a Singularity project would be
> impossible without the agreement of this group. (the keys would not be
> in the possession of the developers). All 10 people would have to agree
> in order to launch a Singularity attempt. Ideally this same group would
> oversee all potential Singularity projects, so that they could analyze,
> compare and pick the one with the best potential to be launched.


I understand that you later, in addition to the suggestion above, floated a
modified suggestion for an advisory board. Just to be absolutely clear, I
am panicking over the suggestion above and not the suggestion of an advisory
board. SIAI has already considered creating an advisory board.

>> My interest is *not* in convincing people that solutions will work. I
>> want a solution that *does work*. I suppose, as a secondary goal, that
>> I want people to know the truth, but that is not primary; solving the
>> problem is primary. It is not supposed to be persuasive, it is
>> supposed to ACTUALLY WORK. Lose sight of that and game over.
> Yes, but have you actually considered the idea that you could be wrong?
> That YOUR ideas may not work?

Oh, for the love of -

Would I have spent O(N^2) hours working out a goal system architecture that
lets an AI sanely spot and fix architectural errors in its own goal system
if I thought I was infallible?

Now ask me whether I think your "not a committee, just 10 bickering people
with individual veto power over the Singularity" idea is going to work.

> Have you considered what would happen if
> this were the case? As they say, two heads are better than one (which,
> btw, I very much agree with). So having a few different people all
> working on the Friendliness problem would be highly beneficial. If for
> no other reason than it would give all of them alternate ideas to look at
> and think about, which may then improve their own designs.

CFAI is on the Web.

>> I know a *lot* of AI morality concepts that sound appealing but are
>> utterly unworkable. For that reason, above all else, I am scared to
>> death of committees. It seems very predictable what the result will be
>> and it will be absolutely deadly.
> Would you please get off the committee thing. Your sounding like a
> broken record because you just keep repeating the same point, over and
> over. Advisory board, not committee (I should NEVER have called it such
> - my bad). Everyone can always use some good advice.

Okay, if the resolution of this is that you call for the creation of the
advisory board and the earlier committee concept gets sunk, buried, and
burned, then I'd accept that as a temporary "win" for the good guys. Not
that the notion isn't going to shamble forth from its grave and have to be
defeated again every six months.

> Its seems to be inherently difficult to convince you that your not god
> and shouldn't personally be making all the decisions that will
> permanently seal the fate of the human race.

If I thought that I'd be advocating the idea of SIAI having veto power over
all AI projects, on the basis of our greater expertise, dedication,
altruism, and understanding of the Singularity. *NOBODY* should have that
power. Not me, not your committee, not anyone. Once we start fighting over
who gets to give orders instead of working out the best engineering
solution, the main issue has been lost.

No, I'm not God. Why would your ten people with veto power be God? What
exactly has been gained here? Maybe you need to stew over the problem and
let it slowly drive you insane for a few weeks, rather than picking on the
first solution that comes to mind.

>> Here's an idea: Instead of convening the Committee to Fuck Up Friendly
>> AI, let's convene the Committee to Decide Whether the CFUFAI Should
>> Exist in the First Place, with a clear understanding that the members
>> of CDWCSEFP will probably *not* serve on CFUFAI.
> Huh?

Calling a committee to solve a problem presumes that a committee is the best
way to solve the problem. Wouldn't it be more logical to first convene a
committee to decide whether a committee is the best way of making the decision?

Eliezer S. Yudkowsky                
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT