If there is a mistake deep in the belief of someone
Are they not ideal Bayesians? Also, do they update based off other people's priors? It could be intresting to make them all ultra-finitists.
Mimemis land is confusing from the outside. I'm not sure how they could avoid stumbling upon "correct" forms of manipulating beliefs, if they persist for long enough and there are large enough stochastic shocks to the communities beliefs. If they also copid succesful people in the past, I feel like this would be even more likely. Unless they happen to be the equivalent of chinese rooms: just an archive of if else clauses.
Anyway, thank you for introducing this delightful style of thought experiments.
Happy to hear you enjoyed it!
On First Principles Land:
Even if they are ideal Bayesians, they could come to a mistake with unfortunate evidence. I'm not sure how we should handle updating on the information of others, that complicates things significantly. I was mostly imagining this as each person independently acts as a semi-ideal Bayesian agent and knows everything from the fundamental truths and evidence themselves. I would be interested in variations with various kinds of knowledge sharing.
On Mimesis Land:
Yea, this land is confusing to me too. I guess belief manipulation would essentially act as an evolutionary process. Some clusters would learn some techniques for belief selection, and the successful clusters would pass on these belief-selection techniques. That said, this would take a while, and a most people could be oblivious to this.
To make "Mimesis Land" work at all, I'd posit that there is some amount of experimentation of techniques and beliefs happening at all times.
It feels like this gives Mimesis Land too much of an edge. They get to experiment with their practices AND pass on the best ones, while First Principles land just has to act on their best guess given the evidence they have at the time.
I think to make it a more fair comparison, neither can have access to "Experimentation Land", the only change in Mimesis land comes from accidents and misunderstandings when passing down beliefs and practices, whereas the only change in First Principles land comes from happenstance direct observations
Interesting point. This leaves open the question of where Mimesis land gets its ideas in the first place.
Epistemic Status: wildly speculative
I've been thinking about the ideas of mimesis and cultural learning by René Girard (known in part for being recommended by Peter Thiel, explanation here) and Joseph Henrich (Secrets of our Success). These seem like both profound, though vastly unexplored ideas. René Girard had a few bold ideas on mimesis, but seemed to focus more on scapegoats and religious details. Joseph Henrich catalogued lots of specific evidence, but didn't really investigate the broader implications.[1]
There are obvious possibilities of these ideas to our epistemics. If it is true that humans copy their ideas from each other instead of deriving them based on reason, often without knowing it, this would have vast implications on our beliefs.
To understand these ideas better, I constructed two hypothetical and extremized societies. The first is "first principles land", where everyone constructs their beliefs individually using first principles. These people act basically as ideal Bayesian agents and approximate gears-levels understandings of everything. This is infinitely difficult to do in practice, but simple to reason about in abstract. The second is "Mimesis Land", where people copy their beliefs from those they consider successful, typically without realizing it (this bit taken from Girard). To make "Mimesis Land" work at all, I'd posit that there is some amount of experimentation of techniques and beliefs happening at all times.
I then compared these two hypothetically societies based on how I expected each to do on 17 various attributes.
Needless to say, Mimesis Land was more interesting to me than First Principles Land.
One tricky thing is that neither land was trying to model real humans; but humans that followed my simple view of a version of a set of extreme theories that might be somewhat coherent. There was a fair bit of subjectivity in this. I'm sure others would come up with significantly different versions of First Principles Land and Mimesis Land. As such I'm not particularly satisfied with this methodology, but at the same time I'm not sure what procedures would be strictly better.
I am bullish on trying to more clearly formulate what first principles thinking and mimesis fully look like in practice, and the costs and benefits of both. In both cases there seem to be fairly patchy terminology of several clusters discussing them without all too much structure. Many thanks to LessWrong for some of my original thoughts on both so far, but there's still a lot of work left.
If I were to continue on a similar path, I would consider adding several other worlds. First Principles Land may get a less extreme version. The Mimesis Land here assumes that people copy the most successful people, and do so mostly unknowingly. It would be interesting to try variants where they copy others (for instance, the most credible people), and do so intentionally (so they are fully aware that their beliefs are inconsistent and often likely to be wrong.) There could be "Chesterton's Fence Land", where traditional beliefs are intentionally chosen. There could be "Forecaster Land", where everyone delegates all thoughts to a prediction market.
I'm really curious what readers may think of this methodology, and this piece in general. Comments highly appreciated.
Disclaimers of things I don't believe:
[1] Social constructivism and has a much more robust literature. To be honest I'm quite rusty on this and need to catch up some time. Interestingly, I don't remember reading about connections between this and Girard or Henrich, though they seem to share a lot of beliefs.
Thanks to Brangus for discussion on the ideas that helped lead to this final post.