Rachel Shu•4 months ago
Literate predicting (in progress, V1)Overview You want to write posts that are epistemically verifiable, but you also want to discuss causal chains even if they are hard to verify, and keep track of your causal reasoning. How do you accomplish this? Another advantage: having lots of little markets sprinkled through your paper is a way to engage readers better. In the spirit of epistemic spot checks, a paper should be written as a series of verifiable claims. Each claim has an associated yes/no market. There is a top-level market which aggregates other markets in the paper. Effectively this market stands for whether or not the thesis of the post is true. Tooling Inspired by Knuth's conception of literate programming, where the bulk of the text is the documentation, not the code. So what does the tooling look like? WYSIWYG editor for posts, where markets would be easily creatable within the post editor. Sort of like a Jupyter notebook. Ability to automate conditional markets: Simple relations between yes/no markets establishable with propositional logic. I guess you could use multiple choice markets here with ternary operators. Higher-order logics and bayesian reasoning also have application in this domain, obviously. You could make very complex structures of reasoning with real-time updating between relationally linked markets, do fancy stuff with numerics, quantifiers. Would be nice to have a flowchart view and you can just drag and drop relations between markets. Example Suppose I claim in my post that Joe Biden will be re-elected in 2024 because Donald Trump will not be unbanned from Twitter before the election. Within the post I should establish markets for the following: Donald Trump is unbanned from Twitter before the election. Joe Biden is re-elected in 2024. [Joe Biden is re-elected in 2024] and not [Donald Trump is unbanned from Twitter before the election]. The third market is entirely automated, and should just be the inverse of Donald Trump is unbanned from Twitter before the election multiplied by Joe Biden is re-elected in 2024. Markets on causal reasoning Note that there is no causal reasoning involved in the previous example. Causation is hard to prove! But, you could have a fourth market that never closes and never resolves titled Joe Biden won the election because Donald Trump was not unbanned from Twitter. On this market people just trade on their beliefs; you can't get an is from an ought. Consensus wins! What Is Truth, anyways? Since it will never resolve, traders get their money back in the long term via Manifold's loan system. Loans would need to be adjusted for this to work: rather than asymptotically getting back your initial investment, you need to asymptotically get back something relative to the current value of your investment. (If you shorted, you get something relative to 1-MKT, obviously.) So moving to a dividends-based approach. Idk maybe this would need to be slightly inflationary to work. Still thinking it through. Perhaps inflation on dividends both positive and negative are paid out from some initial subsidy, and that subsidy is well-publicized; it's a pay-to-publish model where you pay to incentivize scrutiny.
No comments yet