Diversity as an end in itself
On optimising for a rich and vibrant ecosystem of culture and thought
2021-12-12 — 2021-12-12
Wherein It Is Considered Whether Cultivating Human Oddity Is to Be Pursued as an End in Itself, and the Probing of a Configuration Space of Possible Intelligences Is Proposed as a Constituent Aim.
Is cultivating diversity and difference an end in itself? The end? We might argue that it is good for innovation to foster weirdos. Or that diverse teams are more effective. But those are all instrumental goods. What if being a freak is a good in itself? Or do we want to optimise for something more like optionality?
I have a short essay in my mind here about the intrinsic worth of human oddity, as opposed to human uniformity. Is there something beneficial about probing the configuration space of possible humans? Of possible intelligences? Of how maybe it is more efficient to have a monoculture but that way lies despair?
For me, personally, it is important to be around diverse and different people. This is not to say that I always enjoy being around people different to me. It is more that I like better who I am if I am forced to deal with slightly more different sorts of people than I would lazily include in my bubble.
TBD.
- Does the moral worth of chickens uneaten scale linearly in number of chickens?
1 Quality-diversity algorithms
Sometimes, we make diversity a training goal for an algorithm. Maybe that idea is useful to thing about?
In a textbook evolutionary algorithm, under a fixed goal, the artificial selection process typically converges to some optimum. Which is to say, after enough generations, not only might we hope it has found a good solution, but we might also expect that the whole virtual population is very similar, all mildly mutated variants of the reigning champion. Evolutionary optimization algorithms are still optimization, after all, and diversity is the first casualty of optimization. The field maintains a sub-literature of patches — niche-construction, fitness sharing, novelty bonuses — to slow the collapse.
Quality-diversity algorithms promote diversity using an explicit diversity goal. Rather than returning the single best solution, they return an archive, partitioning the space of behaviours into niches and keeping the best solution found in each. MAP-Elites (Cully et al. 2015) does that, imposing a grid of pseudo-niches on behaviour descriptors; novelty search (Lehman and Stanley 2011) drops the quality term altogether and rewards only behaviours the archive has not seen before. This is nominally inspired by biology: diversity survives natural selection, when it does, because there is no single prize to converge on — a landscape of distinct niches is a landscape of distinct local optima, and the best beetle and the best whale are not competing in the same slot. MAP-Elites hard-codes that, with grid cells.
There is an overt informal motivation in terms of option value. In the Cully et al. paper, a hexapod that breaks a leg need not relearn locomotion from scratch; it can consult its pre-computed archive of qualitatively different gaits until it finds one that still works with 5 legs. The archive is insurance against an environment shift that the training objective never anticipated. Formally, though, this cashes out as a static diversity measure over currently-realized behaviours, evaluated at a snapshot — diversity-as-end in the sense of this page, not optionality, since the algorithm is not forward-looking.
I’m sure I’ve seen some DPP-algorithms that resembled this without being explicitly evolutionary. Also, does not a Stein VGD algorithm have a similar flavour?