Toward Legible Adaption

Our dominant world monoculture seems to be drifting into maladaption, and if we don’t fix this it will have much less influence over the future. When natural selection controls future change, one must package features one values with adaptive packages if one wants a hope of their lasting into that future. And I want some of our unusual monoculture features, like open abstract inquiry, to last.

One big cause of our modern cultural drift into maladaption is: a weakening of selection pressures. Compared to centuries ago, cultures (as units) today just don’t die much due to wars, famines, or pandemics. But one “selection” pressure remains strong: people want to join and copy cultures that seem to be “winning”, to gain the respect that comes from being and associating with “winners”.

So maybe if we could get more of us to see (natural selection type) adaption as “winning”, our behaviors might get more adaptive faster. Yes, one big obstacle is that after WWII many (somewhat falsely) blamed “social Darwinism” for Nazism, creating a taboo against trying to consciously win at adaption. But even if we can overcome that, a perhaps bigger obstacle is that we find it hard to see clearly who is winning in this sense, or which policies would help win, and so find it hard to encourage our leaders to help us so win. (Nazi strategies worked out quite badly, for example.) Thus even if we could raise the status of adaption, people might instead pursue other ends while giving lip service to adaption.

I have a solution idea: use speculative markets to create clear well-informed estimates of the future adaptive success of groups who could plausible coordinate to adapt better. That is, create “adaption futures.” With visible, focal, respected market estimates of group adaption, groups could reward leaders who improved their adaptation, or use conditional adaption estimates to pick policies that would so improve. And if we could get down to N=1 “groups”, we might even advise personal decisions re adaption.

To make this work, the main thing we need is good-enough ex-post measures of group adaptive success. With that, we could create assets which pay monotonically in such measures, and subsidize trading in such assets. After which the market prices of such assets can give our desired clear estimates of group adaption. And prices in conditional-called-off adaption markets could give us estimates of which policies promote group adaption.

Now if all we wanted was a measure of DNA adaption success, we’d only need to count the number of descendants of each group perhaps a century or two later. For example, we might measure the DNA of later folks and compare those to DNA measures of current folk. Or maybe just use records of who parents whom. We might also want to weigh final descendant counts by estimates of the future adaptive success of those later descendants, such as from their health, wealth, or status. These could give groups clear signals of how well a group did in the adaption competition.

However, once we realize that most natural selection of humans today is done via cultural evolution, not via DNA evolution, we will want a way to also count cultural descendants, not just DNA descendants. These are counts of how many people later inherited how much of the culture of a group. How can we measure that? Not only accurately, but also canonically, to avoid suspicions that those who define our measures, or who implement their measurement, might do so in ways that favor some groups or policies over others?

I suggest that we take detailed surveys of current cultural behaviors and markers, described in ways that we hope could also apply well to people in the future. Maybe hundreds or even thousands of diverse cultural features per person. And also surveys of who is how much in what groups today. Do such surveys now for a big random sample of people around the world, and also commit to later (say in a century or two) doing such a survey again, of people then. If we see the possible cultures of a person as a big vector space of points x, these surveys could allow us to estimate two normalized distributions over x, n(x) for now, and f(x) for the future. And for each group g, a normalized distribution g(x) of how that group is distributed over x today.

One simple approach would be to define cultural adaptive success per x as a(x) = f(x)/n(x), i.e., how many future people there are at each “culture” x, per each current person with that culture. Then the average success of group g could be the group average of a(x) over x, i.e., a_g = Int_x g(x) a(x) dx. Group market assets m_g might pay according to some monotonic transform of a_g, such as m_g = m(a_g) = ln(a_g). A more advanced sort of combinatorial market might even produce consensus functions that estimate a(x) for all x, not just for particular groups g. That actually seems technically feasible to me.

This approach assumes that the main way that cultural behaviors x change over time is that the people now at x create and influence more future people to be at x later. But what if there are also ways that behaviors x tend to change over time due to internal processes, or due to ways that behaviors depend on changed shared external factors, such as world wealth, peace, health, or tech? If so, the above approach would credit people at x that happen to be toward where the world moves with unusual cultural influence, when in fact such folks needn’t of had much influence at all.

To deal with this, I suggest that we also collect whatever data we can on such systematic cultural change processes, and estimate a change function c(x) that says how points x today tend to change into points y = c(x) later, independent of cultural influence from others, and then estimate cultural adaptive success via a(x) = |Det Dc(x)| f(c(x))/n(x). That first term with a matrix determinant of derivatives corrects for x volume changes due to a complex change function c(x). But keep the analysis to find c(x) simple, so we can stay canonical; when in doubt, attribute change to f/n, not c.

Okay, that’s the basic idea. Now let’s consider some pesky details. We need to make this adaption measure a(x) canonical, so as to avoid suspicions of biases re groups or policies. This is a key problem and I don’t claim to have solved it. But here is a plausible start. First, include as many cultural behavior variables as people are willing to pay to measure. Transform such variables to be more normally distributed, then use mean-zero unit-variance transforms of those. If stat and market methods prefer low dimensionality, do a factor analysis of all these culture variables and focus on the largest factors. Then use standard normality-based stat methods to build best fit models of g(x), n(x), f(x) from the collected data.

As n(x), f(x) are normalized, the above method only estimate relative cultural success. But they can easily be combined with any other estimates, market or otherwise, re the overall future success of humanity. If AI descendants later matter re future success of groups today, then cultures of such future AIs can be included in the later surveys.

As discussed above re estimates of DNA adaptive success, we might improve cultural success estimates via estimates of adaptive success of the population after the time of the future survey, using parameters like individual wealth or status. We might even commit to making a new set of markets then, and use their prices for such estimates.

There might be big disputes re the relative weighting of key culture factors, or of culture vs DNA vs org influence. In this case it seems okay if we just estimate up to a dozen different dimensions of adaption separately. People can then use different functions to combine these into their preferred adaption estimates. There could still be strong incentives to increase these dozen measures of “winning”.

Markets might find it easier to use assets with bounded asset values, obtained via using payoff functions m(a) that are also bounded.

To pay for all this, group representatives might pay survey owners for the right to create and trade assets based on group adaption estimates a_g from the surveys. Such representatives might also pay for market maker subsidies to make those markets more informative, and also for conditional markets to advise particular group decisions. Other funding might come from philanthropy.

You might worry about the long time scales. But stock prices can reveal expectations about commerce issues decades into the future even if individual traders don't hold each stock for more than a year. They just have to expect to sell to other traders with similar expectations.

A strong criticism of this proposal is that firms with a market price do in fact have an estimate of their long term adaptiveness, but this doesn’t stop cultural decay from consistently killing them, even when CEOs have great incentives, powers, and knowledge re preventing this. So without enough macro culture variety, markets estimating the adaptive success of each one might still not be enough to prevent maladaptive macro culture drift.

But let’s turn this critique around: working to prevent firm cultural decay is quite economically valuable in its own right, and succeeding more at that should raise our hopes for also succeeding with analogous efforts for macro cultures.