Very interesting. It sounds to me like this approach exists outside any kind of Bayesian learning framework; i.e. an individual Bayesian agent couldn’t be expected to converge to a correct model here. Is that true?

I would hazard that the RANSAC method sounds a lot like what you’d get out of a larger hybrid model, where e.g. many Bayesian agents with different priors are bred under a genetic algorithm after being ranked by their predictive power. (Like humans surviving to reproduce and pass down models to their children!)