Completely agreed about "talk to" being metaphorical. Indeed, it would astonish me if over the course of few centuries of the kind of technological development implied by whole-brain-emulation, we didn't develop means of interaction with ourselves and others that made the whole notion of concerning ourselves with identity boundaries in the first place a barely intelligible historical artifact for most minds. But I digress.
That aside, I agree that interaction with "fundamentally different minds" could have a huge effect on our tendency to stagnate, but the notion that other humans have fundamentally different minds in that sense just seems laughably implausible to me. If we want to interact with fundamentally different minds over the long haul, I think our best bet will be to create them.
More generally, it sounds like we just have very different intuitions about how much diversity there is among individual minds today, relative to how much diversity there is within a single mind today. I don't have any particularly compelling additional evidence to offer here, so I think my best move is to accept as additional evidence that your expectations about this differ from mine.
As far as the population problem is concerned, I agree, but this has nothing to do with duplicates-vs-"originals". Distributing resources among N entities reduces the average resources available to one entity, regardless of the nature of the entities. Copying a person is no worse than making a person by any other means from this perspective.
I agree that if we are constantly purging most variation, the variation at any given moment will be small. (Of course, if we're right about the value of variation, it seems to follow that variation will therefore be a rare and valued commodity, which might increase the competitiveness of otherwise-less-competitive individuals.)
More generally, it sounds like we just have very different intuitions about how much diversity there is among individual minds today, relative to how much diversity there is within a single mind today. I don't have any particularly compelling additional evidence to offer here, so I think my best move is to accept as additional evidence that your expectations about this differ from mine.
Well, if Eliezer's FAI theory is correct, then the possible end states of any mind capable of deliberate self-modification should be significantly limited by the values, ...
At some point in the future we may be able to scan someone's brain at very high resolution and "run" them on a computer. [1] When I first heard this as a teenager I thought it was interesting but not hugely important. Running people faster or slower and keeping backups came immediately to mind, and Wikipedia adds space travel, but those three by themselves don't seem like they change that much. Thinking speed doesn't seem to be major limiting factor in coming up with good ideas, we generally only restore from backups in cases of rare failure, and while space travel would dramatically affect the ability of humans to spread [2] it doesn't sound like it changes the conditions of life.
This actually undersells emulation by quite a lot. For example "backups" let you repeatedly run the same copy of a person on different information. You can find identify a person when they're at their intellectual or creative best, and give them an hour to think about a new situation. Add in potentially increased simulation speed and parallelism, and you could run lots of these ones looking into all sorts of candidate approaches to problems.
With emulations you can get around the mental overhead of keeping all your assumptions about a direction of thought in your mind at once. I might not know if X is true, and spend a while thinking about what should happen if it's true and another while about what if it's not, but it's hard for me to get past the problem that I'm still uncertain about X. With an emulation that you can reset to a saved state however, you could have multiple runs where you give some emulations a strong assurance that X is true and some a strong assurance that X is false
You can also run randomized controlled trials where the experimental group and the control group are the same person. This should hugely bring down experimental cost and noise, allowing us to make major and rapid progress in discovering what works in education, motivation, and productivity.
(Backups stop being about error recovery and fundamentally change the way an emulation is useful.)
These ideas aren't new here [3] but I don't see them often in discussions of the impact of emulating people. I also suspect there are many more creative ways of using emulation; what else could you do with it?
[1] I think this is a long way off but don't see any reasons why it wouldn't be possible.
[2] Which has a big effect on estimates of the number of future people.
[3] I think most of these ideas fo back to Carl Schulman's 2010 Whole Brain Emulation and the Evolution of Superorganisms.
I also posted this on my blog