Philosophy: Doubt & Faith

Descartes stood at the crossroads of collapsing scholastic theology and Galileo's trial — he needed a foundation for new science immune to every authority. His method: push doubt to its extreme so that it breaks itself. Senses, mathematics, even the existence of the external world are suspended by the demon hypothesis. What remains undoubtable: the very "I" that is doubting must exist. The point is not the slogan but the methodological inversion: doubt is the forge of certainty, and only what survives the harshest doubt earns the right to be foundation.

This is the structure of falsifiability (Popper): a theory is only provisionally accepted after surviving the strongest refutation. In AI engineering, red-teaming, adversarial examples, and formal verification are Cartesian doubt engineered — actively construct the most hostile input to expose a system's real boundary. The "evil demon" has a concrete modern incarnation: out-of-distribution adversarial perturbations.

Pascal was a founder of probability and decision theory (with Fermat, originating expected-value calculation). He faced an embarrassment squarely: reason cannot prove whether God exists. But "not betting" is itself a bet — you must live in some posture. Since betting is unavoidable, use expected value: the believing side has infinite gain, the disbelieving side has infinite loss, and any finite worldly cost shrinks to zero against infinity. This is the first time in philosophy that decision theory tackled a metaphysical question; the real point is not religious persuasion but the principle: when evidence is insufficient, act by expected utility, not by waiting "until the evidence is complete."

Directly maps to Bayesian decision theory and risk management: under prior uncertainty, bet via the payoff matrix. It is also the core of Taleb's antifragility — what matters is not probability prediction but payoff asymmetry: bets with bounded downside and unbounded upside should be taken with eyes closed. Bostrom on AI existential risk, the precautionary principle in climate policy, and option theory in entrepreneurship are all isomorphic extensions of Pascal's wager.

Hume pushed doubt where Descartes dared not go: even "I think therefore I am" cannot be saved. First cut — at causation: we never observe "cause compelling effect," only "B always follows A"; causation is the brain's habit-projection from constant conjunction. Second cut — at self: introspection finds only a fluid bundle of perceptions, never a "self that bears them." Third cut — at induction: "the sun will rise tomorrow" has no logical guarantee. But he did not slide into nihilism. His stance is mitigated skepticism — acknowledge the limits of reason and yet keep acting via habit, sentiment, and shared life.

"Constant conjunction" is virtually the definition of statistical learning two centuries early: models learn only from correlation and cannot see causation — exactly the gap Judea Pearl's do-calculus aims to fill. The "bundle theory" resonates strikingly with neuroscience's predictive processing (Friston, Anil Seth) and Buddhist five aggregates without self: what we call "self" is a narrative dynamically synthesized by the brain, with no independent substance. LLMs are textbook Humean systems — they extract conjunctions from massive text and never truly "understand" cause.