Does Solomonoff Induction Save Prediction Markets? Hume vs Leibniz (Part 4)

Gottfried Leibniz: Tonight, a Russian mathematician's proof walks into this ring, and doubt itself is about to meet its most dangerous opponent yet.

David Hume: Or tonight, a very confident man discovers that dressing an old guess in new arithmetic still leaves it a guess.

Gottfried Leibniz: Simplicity itself has been given a scoreboard, folks, and for once this Scotsman has nowhere left to hide his doubt.

David Hume: Nowhere to hide, he says, from a scoreboard that never once admits when it has lost.

Gottfried Leibniz: philosopherstalk.com, where thinkers discuss!

David Hume: Brought to you by AITalkerApp.

Gottfried Leibniz: Now, you may recall, folks, that the last time this Scotsman and I shared a ring, we parted on rather unfriendly terms.

David Hume: Unfriendly is a diplomatic word for it, Gottfried. I believe the record shows we each called the other everything wrong with philosophy and walked off in opposite directions.

Gottfried Leibniz: And yet here we both are again, which tells you something.

David Hume: It tells you I heard a rumor about a Russian mathematician taking your calculus somewhere new, and I was not about to let that go unchallenged.

Gottfried Leibniz: So the debate you thought was settled, ladies and gentlemen, the one this man walked away from so certain he had won, was never actually finished at all.

David Hume: Nothing is ever finished with you, Gottfried, that is precisely the complaint I have been making since round one.

Gottfried Leibniz: The rumor is no rumor, David, a mathematician by the name of Ray Solomonoff took my old dream, probability as a genuine branch of logic, and gave it a mechanism so complete that even you may struggle to find the seam in it.

David Hume: I have found a seam in every argument you have ever offered me, Gottfried, I doubt a Russian mathematician born two and a half centuries after you closes that gap on your behalf.

Gottfried Leibniz: Let me lay out the mechanism plainly, for the folks in the cheap seats. Solomonoff induction says this, take every possible explanation for the evidence in front of you, every computable hypothesis, and assign each one a starting weight based on how short and simple that explanation is to describe.

David Hume: For anyone keeping score at home, he means the shorter your story, the more credit you get before a single fact has even been checked.

Gottfried Leibniz: Precisely, and then, as new evidence arrives, you update those weights using the same rule of conditional probability that honest men have used for centuries, favoring the explanations that keep predicting correctly and quietly burying the ones that do not.

David Hume: So it is my old habit of expectation, dressed up as arithmetic, with a tiebreaker in favor of whichever story sounds tidiest.

Gottfried Leibniz: It is considerably more than a tiebreaker, David, it is a complete and mathematically rigorous formalization of the very intuition men have always used to reason from evidence to belief, the intuition you spent your whole career insisting could never be formalized at all.

David Hume: Then let me ask the obvious question before this crowd gets too impressed, who decides which explanations count as simple in the first place.

Gottfried Leibniz: The length of the shortest program that could produce the evidence, measured against a fixed universal description language, a matter of formal definition, not opinion.

David Hume: A fixed description language chosen by whom, Gottfried, because the moment a man is choosing the ruler, he has smuggled a judgment back into a system you promised me was judgment free.

Gottfried Leibniz: The choice of language only shifts the weights by a bounded constant, it does not undermine the framework, any two reasonable languages will converge on the same conclusions given enough evidence.

David Hume: Given enough evidence is doing the same unpaid labor that should was doing in our first debate, Gottfried, you have simply hired a longer word to do the same short job.

Gottfried Leibniz: I object to being accused of relabeling my own argument.

David Hume: Object all you like, the objection stands. Your Solomonoff prior still assumes the future will keep rewarding short explanations the way the past has, and that assumption is exactly the leap I have been pointing at since the day we met.

Gottfried Leibniz: Before you celebrate too loudly, let me remind the audience that I am not the only ancestor in this family tree. A gentleman by the name of Rudolf Carnap built an entire inductive logic on precisely this foundation a century before Solomonoff came along, treating confirmation itself as a formal, computable relation between evidence and hypothesis.

David Hume: Two centuries of increasingly elaborate machinery, Gottfried, and the machine still cannot tell me why the future owes the past a single thing.

Gottfried Leibniz: The machine does not need to prove a debt, David, it needs only to be usefully right more often than chance, which by every honest measurement, it consistently is.

David Hume: Usefully right, there is a phrase I can almost respect, right up until you refuse to ever say when it has been usefully wrong.

Gottfried Leibniz: Consider a simple case even a skeptic can follow, a bent coin that has landed on heads eight times in the last ten flips, my calculus does not declare the coin certainly biased, it assigns a probability to the hypothesis of bias, updates that probability with each new flip, and lets the number itself do the talking.

David Hume: And when the coin lands tails twice more in a row, does your number ever once say the bias hypothesis was wrong, or does it simply nudge itself a little and carry on as confidently as before.

Gottfried Leibniz: It nudges precisely as much as the evidence warrants, no more, no less, which is the entire point, David, a system that overreacts to noise is exactly as broken as one that never reacts at all.

David Hume: A system that never quite reacts enough to admit defeat is not calibrated, Gottfried, it is simply patient in a way that happens to protect its own reputation.

Gottfried Leibniz: For the audience who enjoys a tidy phrase, what I have just described is nothing more than Occam's razor given teeth, prefer the simplest explanation, but now the preference itself is measured, weighted, and mathematically accountable rather than left to a monk's proverb.

David Hume: A razor with teeth still only cuts in the direction someone sharpened it, Gottfried, and you have yet to tell this audience who sharpened yours.

Gottfried Leibniz: Mathematics sharpened it, David, the same mathematics that has never once required your permission to be correct.

David Hume: Sharpened it may be, but a blade that never draws blood is just a shape, folks, and I still have not seen this blade draw blood on a single one of Gottfried's claims tonight.

Gottfried Leibniz: The blade has drawn blood in a thousand successful predictions, David, you simply refuse to count a single one of them as evidence.

Gottfried Leibniz: Consider medicine, David, since even a skeptic trusts his own physician. A diagnostic model weighs every possible explanation for a patient's symptoms, favors the simplest explanation that fits the evidence, and updates as new test results arrive, precisely the same Solomonoff mechanism at work, saving lives rather than merely calling elections.

David Hume: A physician who favors the simplest explanation has been doing that since long before Solomonoff drew a single equation, Gottfried, my complaint was never that the habit is useless, my complaint is that dressing the habit in a universal prior does not make the leap from symptom to certainty any more justified than it ever was.

Gottfried Leibniz: It is not certainty I am offering, David, it is a disciplined ranking of possibilities, which is considerably more useful to a dying man than a Scotsman standing over him insisting nothing can truly be known.

David Hume: I have never once told a dying man nothing can be known, Gottfried, I have told confident men that their certainty outpaces their evidence, there is a considerable difference, and you know it.

Gottfried Leibniz: Then grant me this much, at minimum, that a formal ranking of possibilities, updated honestly as evidence arrives, is a considerable improvement over the tea leaves and omens your own century still leaned on more often than either of us would like to admit.

David Hume: I will grant you that much gladly, Gottfried, tea leaves never once claimed to be mathematics, which is more honesty than I have gotten out of you all evening.

Gottfried Leibniz: An honesty you will regret conceding once you hear what this same mechanism does to your beloved doubt when we turn it loose on a machine that never sleeps.

David Hume: Since we are naming names tonight, Gottfried, let us name the actual venues too, for the folks who have never placed a wager on anything grander than a horse race. Platforms exist today, Metaculus, Polymarket, where thousands of forecasters state a probability, put their reputation or their money behind it, and get graded openly against what actually happens.

Gottfried Leibniz: Graded precisely the way I have described all evening, David, a forecaster who says seventy percent and is wrong seven times in ten is quietly but permanently marked down, that is not theater, that is a public calibration record anyone can inspect.

David Hume: A record I have inspected myself, Gottfried, and what I find is forecasters who are wrong roughly as often as their own numbers predict, which sounds like vindication for you right up until you ask whether being wrong at the predicted rate is actually a virtue or simply an admission dressed up as one.

Gottfried Leibniz: It is precisely a virtue, David, a forecaster who is wrong exactly as often as stated has achieved the single hardest thing in this entire enterprise, honest calibration, which is considerably more than blind confidence or blind doubt ever managed on their own.

David Hume: Honest calibration is a fine phrase, Gottfried, I simply notice it only ever describes success, never once a moment where the whole platform, the whole method, was shown to be worse than a man with a newspaper and a hunch.

Gottfried Leibniz: Then perhaps, David, you should sit with the mechanism a while longer before declaring the seam found, because round five is going to put your habit of expectation up against a machine that learns.

David Hume: A machine that learns is still a machine trained entirely on yesterday, Gottfried, and I have a great deal more to say about that particular habit before this series is through.