There is such a thing as diminishing marginal returns. If there were 5 Eliezer Yudkowskys all working on FAI, the sixth Eliezer Yudkowsky ought to go off and direct their efforts to CFAR instead (this is a calculation specifically for copies of Eliezer Yudkowsky, and not precise).
I would say that having one smart non-Eliezer person working on FAI together with you would be preferable to having two or maybe even three copies of EY. The copies are likely to have similar thought patterns, have similar biases and make similar mistakes.
It depends on the topic. If the problem can be solved by lots of hard work and this hard work can be done in parallel, then copies are a good idea. If the problem requires inspiration, perhaps in many incremental steps, then more diverse minds would be preferable.
I think what I am trying to say is that marginal returns diminish faster for copies compared to other people.
In a recent discussion a friend was telling me how he felt he was not as smart as the people he thinks are doing the best research on the most important topics. He said a few jaw-dropping names, which indeed are smarter than him, and mentioned their research agenda, say, A B and C.
From that, a remarkable implication followed, in his cognitive algorithm:
Therefore I should research thing D or thing E.
Which made me pause for a moment. Here is a hypothetical schematic of this conception of the world. Arrows stand for "Ought to research"
Humans by Level of Awesome (HLA) Research Agenda by Level of Importance. (RALI)
HLA RALI
Mrs 1 --------> X-risk #1
2 --------> X-risk #2
3 --------> Longevity
4 --------> Malaria Reduction
5 --------> Enhancement
1344 --------> Increasing Puppies Cuteness
Etc...
It made me think of the problem of creating match making algorithms for websites where people want to pair to do stuff, such as playing tennis, chess or having a romantic relationship.
This reasoning is profoundly mistaken, and I can look back into my mind, and remember dozens of times I have made the exact same mistake. So I thought it would be good to spell out 10 times in different ways for the unconscious bots in my mind that didn't get it yet:
1) Research agenda topics are polygamous, they do not mind if there is someone else researching them, besides the very best people who could be doing such research.
2) The function above should not be one-to-one (biunivocal), but many-to-one.
3) There is no relation of overshadowing based on someone's awesomeness to everyone else who researches the same topic, unless they are researching the same narrow minimal sub-type of the same question coming from the same background.
4) Overdetermination doesn't happen at the "general topic level".
5) Awesome people do not obfuscate what less awesome people do in their area, they catapult it, by creating resources.
6) Being in an area where the most awesome people are is not asking to "lose the game" it is being in an environment that cultivates greatness.
7) The amount of awesomeness in a field does not supervene on the amount of awesomeness in it's best explorer.
8) The Best person in each area would never be able to cause progress alone.
9) To want to be the best in something has absolutely no precedence over doing something that matters.
10) If you believe in monogamous research, you'd be in the akward situation where finding out that no one gives a flying fuck about X-risk should make you ecstatic, and that can't be right. That there are people doing something that matters so well that you currently estimate you can't beat them should be fantastic news!
Well, I hope every last cortical column I have got it now, and the overall surrounding being may be a little less wrong.
Also, this text by Michael Vassar is magnificent, and makes a related set of points.