Probabilistic Humility
Treat every belief as uncertain and size decisions to the evidence
- Difficulty
- Advanced
- Time to result
- ~ongoing to results
- Steps
- 6
- Confidence
- 97%
Probabilistic Humility begins with the claim that perception and information are incomplete, delayed, and interpreted through imperfect mental models. Therefore, a belief about a company, market, or future event should be expressed as a probability rather than a certainty. The decision-maker identifies missing information, imagines failure paths, and sizes exposure according to both confidence and downside. Convictions remain usable, but they are provisional: new evidence can raise, lower, or reverse them. This creates humility without paralysis because action does not require perfect knowledge; it requires a sufficiently favorable bet. The method also separates decision quality from a single outcome, since good probabilistic decisions can lose and reckless ones can win. Continual updating prevents yesterday's justified position from becoming today's unexamined assumption.
Origin
Extracted from Masters in Business through Dave Nadig's discussion of perception, uncertainty, Annie Duke's thinking in bets, and the fallibility of human experience.
Core principles
- 01Reality is only partially observable
- 02Confidence should reflect evidence quality and missing information
- 03Decisions are bets, not declarations of certainty
- 04Strong conviction still requires continual updating
- 05Acknowledging fallibility improves judgment
How to run it
- 1
Describe the uncertain claim
Convert a categorical belief into a specific proposition whose likelihood can be considered.
Pro tip Include a timeframe and observable outcome.
Watch out Vague claims cannot be updated rigorously.
- 2
Audit the information boundary
List what is known, delayed, inferred, unavailable, or potentially changing outside your view.
Pro tip Ask what even the best-positioned expert could currently be missing.
Watch out Status or expertise does not eliminate blind spots.
- 3
Assign provisional odds
Estimate a probability using available evidence, base rates, and competing explanations.
Pro tip Use ranges when precision would be artificial.
Watch out A number does not make weak evidence objective.
- 4
Design the bet
Choose an action and exposure whose upside, downside, and reversibility fit the estimated odds.
Pro tip Preserve the ability to recover when uncertainty is high.
Watch out High conviction does not justify ignoring ruin risk.
- 5
Seek contradiction
Look deliberately for evidence and scenarios that would make the current belief less likely.
Pro tip Record in advance what evidence would change your mind.
Watch out Consuming only confirming information turns probability into dogma.
- 6
Update continuously
Revise odds and actions as the world, available evidence, and relevant constraints change.
Pro tip Review positions on a schedule and after material events.
Watch out Do not defend an old decision merely because it was reasonable when made.
In the wild
An analyst avoids declaring that Tesla will beat expectations. She assigns a probability, documents gaps in factory and demand data, models downside scenarios, and takes a position small enough to survive being wrong. New production information changes both her probability and exposure.
→ The investment reflects uncertainty while remaining actionable and adaptable.
A traveler estimates that the subway is usually faster but considers service interruptions and the cost of arriving late. For an ordinary trip he takes the subway; for a critical appointment he leaves earlier or chooses a more controllable option.
→ The action reflects probabilities and consequences rather than a false promise of certainty.
Common mistakes
Using uncertainty as an excuse not to act
The framework supports appropriately sized action; it does not require certainty before making a decision.
Expressing false precision
Exact percentages can conceal weak evidence and should not replace honest ranges or qualitative uncertainty.
Confusing outcomes with decision quality
A favorable result does not prove the original odds or process were sound, and an unfavorable result does not prove they were poor.
Is it for you?
Best for
Investors, executives, analysts, and forecasters operating in uncertain or rapidly changing environments.
Not ideal for
Routine situations governed by deterministic procedures where probabilistic framing would add unnecessary complexity.
From the transcript
“we're inherently always wrong about the state of anything that we are asserting truth on.”
“It's actually the fabric of the universe is based on managing probability, not understanding certainty.”
“You have to have the humility of your convictions.”
From the episode
Dave Nadig on Exchange-Traded Funds
Dave Nadig