Doubt What Your Work Depends On

You might think a good skeptic doubts everything. You can't, and you shouldn't try. The real question is what to doubt.

A short course film for INFO 7375 Computational Skepticism for AI, built on Descartes: the Meditations (1641), where he tears every doubtable belief down once in his life to rebuild from the foundations, and the Discourse on the Method (1637), where he keeps a provisional code to live by while the doubting runs. The course borrows his move as one question you can ask of any claim: what would have to be true for this to be wrong? Then it adds a rule for where to aim that question: risk. How likely is the claim to be wrong, and what would it cost if it were?

Three terms: radical doubt, asking what would have to be true for a claim to be wrong; risk, how likely a claim is to be wrong and what it costs if it is; a provisional code, rules acted on while still doubting.

In the Meditations (1641), Descartes set out, once in his life, to tear down every belief he could find any reason to doubt and rebuild from the foundations — safe to do, he says, because he was after knowledge, not action. This course borrows that move as one question for any claim: what would have to be true for this to be wrong? The answer isn't a mood — it's a list of things that can actually be checked. But that question can't run on everything; every check costs time, and doubting every claim means never acting at all.

Descartes saw this too. In the Discourse on the Method (1637), he says that while pulling down a house to rebuild it, another house is needed to live in — so while he doubted, he kept a provisional code, a few rules to act by. One rule: a traveler lost in a forest shouldn't wander side to side or stand still — pick a direction and walk straight, and you'll likely end up somewhere better than the middle of the forest. When truth can't be told and action is required anyway, act on what's most probable.

So aim the doubt: ask how likely a claim is to be wrong, and what it would cost if it were — together, that's risk, treated as a way of thinking rather than a measured model. A chair that's held for years gets no doubt — sit down. Most claims in a working day live in that corner: act on them, fix them if they break. A few claims sit elsewhere: a triage score, a medication dose, the number a job depends on — if one of those is wrong, someone pays for it. Those get the doubt. The rule is proportion: the more it costs to be wrong, the more doubt a claim earns.

Your Turn: Write down five claims your work relied on this week, and paste this into Claude: "Here are five claims my work relied on this week: [paste them]. For each one, help me say what it would cost if it were wrong, and rank them by that cost. Then, for the top one or two only, ask: what would have to be true for this to be wrong? List conditions I can check." Then check two things yourself: are the top two really the ones that could hurt someone, and could each condition actually be tested this week?

Sources: René Descartes, Meditationes de prima philosophia (1641), Latin text via Project Gutenberg 23306; Meditations on First Philosophy, tr. John Veitch, Wikisource. René Descartes, Discourse on the Method (1637), tr. John Veitch, Project Gutenberg 59. René Descartes, Principles of Philosophy (1644), Part I, tr. John Veitch, Project Gutenberg 4391. Nik Bear Brown, INFO 7375 Computational Skepticism for AI, Chapter 1, The Skeptic's Toolkit — risk-weighted doubt is the course's rule, not Descartes'. Every factual claim checked against these primary sources before rendering.

0:00 The question: what to doubt 0:13 Three terms: radical doubt, risk, provisional code 0:25 Descartes, Meditations (1641): doubt once in a life 0:40 The move: what would have to be true for this to be wrong? 0:57 Discourse on the Method (1637): a house to live in while you rebuild 1:10 The traveler lost in the forest: act on what is most probable 1:24 Aim the doubt: how likely, and what it costs 1:44 The claims that get the doubt 1:54 The rule is proportion 2:06 Your Turn

Course: INFO 7375 Computational Skepticism for AI · Northeastern University · Nik Bear Brown. Part of the Claude & Agentic AI series. Made with the brutalist.art toolkit: Kokoro for the voice, Manim for the drawings. @NikBearBrown

SEO Keyword Tags

computational skepticism, Descartes Meditations, Discourse on the Method, radical doubt, risk-weighted doubt, INFO 7375, Northeastern AI course, Cartesian doubt explained, what to doubt, epistemology for AI, critical thinking AI claims, philosophy of skepticism, provisional code Descartes, evaluating claims risk, AI skepticism course

What this video covers

• 0:00 — The question: what to doubt 0:13 Three terms: radical doubt, risk, provisional code 0:25 Descartes, Meditations (1641): doubt once in a life 0:40 The move: what would have to be true for this to be wrong? 0:57 Discourse on the Method (1637): a house to live in while you rebuild 1:10 The traveler lost in the forest: act on what is most probable 1:24 Aim the doubt: how likely, and what it costs 1:44 The claims that get the doubt 1:54 The rule is proportion 2:06 Your Turn

Transcript

A lightly formatted transcript of the video, with timestamps so you can jump to any section on YouTube.

The question: what to doubt 0:13 Three terms: radical doubt, risk, provisional code 0:25 Descartes, Meditations (1641): doubt once in a life 0:40 The move: what would have to be true for this to be wrong? 0:57 Discourse on the Method (1637): a house to live in while you rebuild 1:10 The traveler lost in the forest: act on what is most probable 1:24 Aim the doubt: how likely, and what it costs 1:44 The claims that get the doubt 1:54 The rule is proportion 2:06 Your Turn

[0:00] Salam, this is Liam in for Bear with a short film for info 7375 computational skepticism for AI. You might think a good skeptic doubts everything. You can't and you shouldn't try. The real question is what to doubt. Three terms: radical doubt, asking what would have to be true for a claim to be wrong, risk, how likely a claim is to be wrong and what it costs if it is, and a provisional code, rules you act by while you're still doubting. In his meditations, published in 1641, Descartes set out once in his life to tear down every belief he could find any reason to doubt and rebuild from the foundations. He says it was safe to go that far because he was after knowledge, not action. This course borrows his move as one question you can ask of

Any claim. What would have to be true for this to be wrong? The answer isn't a mood, it's a list of things you can go and check. But you can't run that question on everything. Every check costs time. Doubt every claim you meet and you never get to act and the work just waits. Descartes saw this in his Discourse on the Method from 1637. He says that while you pull down your house to rebuild it, you need another house to live in. So while he doubted, he kept a provisional code, a few rules to act by. One rule, a traveler lost in a forest shouldn't wander from side to side or stand still. Pick a direction and walk straight and you'll likely end up somewhere better than the middle of the forest. When you can't tell what's

True and you have to act, act on what's most probable. So aim the doubt, ask two things about a claim. How likely is it to be wrong and what would it cost if it were? Together, that's risk. Treat this as a way of thinking, not a measured model. The chair you're sitting on has held you for years and if it failed, you'd land on the floor. Don't doubt it, sit down. Most claims in a working day live in this corner. Act on them and fix them if they break. A few claims land over here. A triage score on a patient, a medication dose, the number your job depends on. If one of those is wrong, someone pays for it. Those get the doubt. The rule is proportion. The more it costs to be wrong, the more doubt

A claim gets. Cheap to be wrong, act and watch. Costly to be wrong, stop and ask what would have to be true for it to be wrong before the world answers for you. Your turn. Write down five claims your work relied on this week and paste this into Claude. Here are five claims my work relied on this week. Paste them. For each one, help me say what it would cost if it were wrong and rank them by that cost. Then for the top one or two only, ask what would have to be true for this to be wrong. List conditions I can check, then check two things yourself. Are the top two really the ones that could hurt someone? And could you test each condition this week? Doubt what your work depends on at Nick Bear Brown.