Philosophy Philosophy Of Science Simulation

A hidden coin, three rival theories. See why one counterexample kills a universal law while evidence only shifts probabilities for the rest.

Controls

What you see

Readouts

Independent check

...
The science

Popper's falsification. A universal claim ("the coin always lands heads") can never be proved by finite data, but a single tail refutes it. Here its posterior drops to exactly 0 at the first tail.

Bayesian confirmation. For hypotheses with heads-probability p, the likelihood of k heads and m tails is p^k (1-p)^m, and posterior is proportional to prior x likelihood. Updating flip by flip gives the same answer as updating once on all the data, which this page checks.

Evidence accumulates at a rate. If the truth has heads probability q, the expected log likelihood ratio per flip in favour of truth over a rival with probability p is the Kullback-Leibler divergence D = q ln(q/p) + (1-q) ln((1-q)/(1-p)). Bigger gaps between theories are resolved in fewer flips.

Honest simplifications. Only three theories are on the table, and the true value may be none of them; Bayes then picks the least wrong, not the right one. Real science also weighs simplicity, auxiliary assumptions and the choice of priors (Duhem-Quine, Kuhn). This model shows the logic of evidence, not the whole of the practice.