#60 - Annie Duke, decision strategist: Poker as a model system for life—how to improve decision making, use frameworks for learning, and apply 'backcasting' to boost your odds for future success
Former World Series of Poker champion and author, Annie Duke, discusses how poker models real-world decision-making under uncertainty, blending luck and skill. She explores the decision matrix, the 'resulting' heuristic, and frameworks for learning and improving probabilistic thinking, including the concept of "backcasting."
Deep Dive Analysis
19 Topic Outline
Annie Duke's Background and Entry into Poker
Chess vs. Poker: A Model for Real-World Decision Making
The Challenge of Probabilistic Thinking
Variable Reinforcement and Poker's Psychological Draw
Defining Skill and Luck in Games
Understanding Texas Hold'em Poker Basics
The Art of Reading Opponents and Game Dynamics
The Value of 'Quitting Fast' in Poker and Life
Evolution of Poker: Limit vs. No-Limit Play
The Impact of Analytics on Modern Poker
The Decision Matrix and the 'Resulting' Heuristic
Consequences of Avoiding Non-Status Quo Decisions
Poker as a Model System for Life Decisions
Leadership and Status Quo Decisions
Learning from Unexpected Outcomes (Good and Bad)
A Framework for Levels of Thought in Learning
The Capacity for Self-Deception and Soft Landings
Backcasting and Pre-Mortems for Better Planning
Navigating Decision Complexity and Speed
6 Key Concepts
Resulting
Resulting is a cognitive shortcut where people judge the quality of a decision based solely on its outcome, rather than the process used to make it. This often leads to misattributing good outcomes to good decisions and bad outcomes to bad decisions, even when luck played a significant role.
Variable Reinforcement Schedule
This psychological concept describes how rewards given on an unpredictable, average basis (rather than a fixed one) lead to highly persistent behavior. Individuals continue an action because they believe they are 'due' for a reward, as seen in slot machines or poker.
Skill vs. Luck
Skill refers to actions within one's control that affect outcomes, while luck refers to stochastic elements outside of one's control. A key test for skill is whether one can intentionally lose while following the rules of the game. The influence of luck becomes more apparent as the skill gap between competitors narrows.
Transparency Problem
This describes the tendency to blame individuals more for bad outcomes resulting from unexpected or non-status quo decisions, but less for bad outcomes from expected or consensus decisions. Conversely, good outcomes from unexpected decisions are highly praised, while those from expected decisions are merely acknowledged.
Backcasting
A planning technique that involves imagining a desired future outcome (e.g., achieving a goal) and then looking backward from that future point to identify the steps, decisions, and potential interventions (including luck or hedges against bad luck) that led to that success.
Pre-mortem
A planning technique where one imagines a future failure (e.g., a project failing) and then works backward to identify all the possible reasons why that failure might have occurred, including factors outside of one's control. This helps in proactively identifying risks and developing contingency plans.
9 Questions Answered
Poker is a better metaphor because, unlike chess, it involves incomplete information and a strong influence of luck, which are both prevalent in real-world decision-making scenarios.
A good test for skill is whether you can intentionally lose while following the rules of the game. If you can, it's a game of skill; if not, it's purely luck.
This persistence is due to variable reinforcement schedules, where rewards are unpredictable but occur often enough to keep players engaged, leading them to believe they are 'due' for a win.
Modern professional poker players use analytics and simulations to understand optimal strategies, calculate probabilities for various situations, and identify bluffing opportunities with greater precision, making the game much more data-driven.
Resulting is the cognitive bias of judging a decision's quality solely by its outcome. It's problematic because it ignores the process and the role of luck, leading to unfair blame for good decisions with bad outcomes and missed learning opportunities from bad decisions with good outcomes.
Leaders tend to make status quo or consensus decisions to avoid being blamed as an 'idiot' if an unexpected, non-consensus decision leads to a bad outcome, even if the unexpected decision had a higher expected value.
The levels progress from examining losses through luck (Level 1), to examining losses through both luck and skill (Level 2), to examining all four quadrants of decision/outcome (Level 3), and finally to questioning if even good decisions with good outcomes could have been optimized further (Level 4).
These techniques involve imagining a future success or failure and working backward to identify contributing factors, including the role of luck. This process helps uncover potential risks, develop hedges, and plan responses in advance, leading to more robust strategic plans.
Understanding a broader decision-making framework helps identify which decisions require deep analysis (the 'curves') and which can be made quickly (the 'straightaways'). For big decisions, embracing the probabilistic nature helps move past analysis paralysis by focusing on relative optimality rather than absolute certainty.
26 Actionable Insights
1. Think Probabilistically
Recognize that decision outcomes are not deterministic; many things could occur with varying likelihoods. Understanding this probabilistic nature is crucial for better decision-making.
2. Avoid “Resulting” Bias
Do not judge the quality of a decision solely by its outcome, especially in situations with high uncertainty. A good decision can lead to a bad outcome, and vice-versa.
3. Relentlessly Explore Counterfactuals
After any outcome (good or bad), actively consider alternative scenarios that could have occurred, including those that didn’t happen but were possible. This deep dive into “what if” scenarios is crucial for significant progress and understanding decision quality.
4. Practice Backcasting
To plan effectively, envision having already achieved your goal (e.g., “it’s a year and a day, and I achieved X”). Then, look backward and ask, “How did I actually achieve this?” to identify the steps and factors, including luck, that led to success.
5. Conduct a Premortem
Before starting a project or pursuing a goal, imagine it’s a year later and you’ve failed. Then, work backward to identify all the potential reasons for that failure, including luck, allowing you to proactively mitigate risks.
6. Plan for Luck’s Intervention
After identifying potential lucky or unlucky events (via backcasting/premortem), proactively ask: Can I increase good luck? Can I decrease bad luck? Can I hedge against bad luck? And most importantly, what will be my pre-planned response if bad luck occurs, to ensure calm, effective decision-making.
7. Advance Through Outcome Analysis Levels
Progress from attributing losses to luck and wins to skill (Level 1) to critically analyzing all four quadrants (good/bad decision, good/bad outcome) equally (Level 3), and ultimately questioning even successful outcomes for even better paths (Level 4).
8. Evaluate Decisions Systematically
To assess decision quality, analyze all possible outcomes, their likelihoods, and your preferences for each. Compare this expected value against other potential decisions to choose the path with the highest chance of a preferred outcome.
9. Catch Decision Errors Faster
Aim to identify and correct errors in thinking (e.g., overconfidence, deterministic thinking) more quickly, as even small improvements in this area significantly enhance decision-making.
10. Embrace “Quit Fast” Strategy
When exploring new activities, try many low-risk options, then quickly abandon those that don’t yield positive expected value (happiness, health, etc.) to free up time for things you truly love and want to commit to.
11. Analyze All Unexpected Outcomes
Just as negative outcomes trigger analysis, unexpectedly good outcomes should also be thoroughly examined. This helps identify overlooked risks, refine models, and learn from successful strategies that may not have been fully understood.
12. Forecast & Analyze Performance Deviations
Before an activity, forecast your expected performance. Afterward, analyze deviations: if you perform significantly below or above your forecast, investigate the underlying reasons to learn and improve.
13. Overcome Identity-Protecting Biases
To reach higher levels of decision analysis, be willing to challenge your core beliefs about your competence, even if it means re-evaluating a successful outcome and potentially “turning a win into a loss” for deeper learning.
14. Actively Seek Feedback
Cultivate a hunger for feedback on your decisions and outcomes. This proactive approach helps you understand potential futures, anticipate reactions, and learn from your experiences to continuously improve.
15. Guard Against Status Quo Bias
Be aware that fear of blame or a desire to avoid being seen as an “idiot” can lead to making status quo decisions rather than optimal ones, even when data suggests otherwise.
16. Foster Innovation by Reducing Fear
To encourage innovation and thinking outside the box, leaders and individuals should reduce the fear of being labeled an “idiot” for unexpected decisions that lead to bad outcomes, as this fear suppresses progress.
17. Prioritize Risk-Adjusted Return
When evaluating investments or decisions, focus not just on potential returns but also on the risk involved (risk-adjusted return on capital) to make more robust and sustainable choices.
18. Ask “Should We Have Done Worse?”
In post-mortems for negative events, expand the inquiry beyond “Could we have done better?” to also ask “Should we have done worse?” This helps identify situations where the decision process was actually worse than the outcome suggested, offering deeper learning.
19. Evaluate Against Perfect Information
After an outcome, consider what you would have done with perfect information. This reveals if you underplayed or overplayed a situation, helping to refine your strategy regardless of the actual result.
20. Use Poker as Decision Model
Poker, with its incomplete information and strong influence of luck, is a better model for real-world decision-making than chess, which has perfect information and no luck. This helps understand the nuances of decisions in uncertain environments.
21. Increase “Skin in the Game”
When personal accountability (e.g., using your own money, reputation) is high, it creates a stronger incentive to engage in higher-level, more objective decision-making and analysis.
22. Cultivate Rapid Improvement Conditions
To accelerate decision-making skill, seek environments with high “skin in the game,” a significant blend of luck and skill, and short feedback loops, as these conditions force rapid learning or failure.
23. Employ Systematic Thinking for Long Horizons
In decisions with extended feedback loops, systematic and probabilistic thinking becomes even more critical, as longer time horizons offer greater opportunity for self-deception and bias to go unchecked.
24. Treat Long-Term Decisions as Immediate
For decisions with long feedback loops, analyze them with the same rigor as if you would receive immediate feedback, like in chess. This helps counteract the tendency for self-deception over time.
25. Avoid “Soft Landing” Complacency
When decisions lead to positive but suboptimal outcomes, resist the urge to stop analyzing. Dig deeper to identify the truly optimal path and capture the full potential value that was missed.
26. Optimize Decision Speed
Understanding a robust decision-making framework allows you to decide quickly on routine matters and also to navigate complex, uncertain situations more efficiently, avoiding paralysis by recognizing when a “good enough” path is sufficient.
8 Key Quotes
If you could meet anybody famous, who would it be? I mean, at the top of my list would have to be him [Bill Belichick]. Like, and again, it wouldn't be in the setting of him giving glib answers in a news conference, cause that wouldn't be that much fun. Although it's funny as hell to watch. If I could spend a day on his boat with him, really understanding the discipline that he brings to what he does, it would just, it would just amaze me.
Peter Attia
We do not spend enough time encouraging people to quit.
Annie Duke
I'm due. I'm due. And what that means is that it doesn't matter what you do to that machine, you could just turn it off. What they think is no, it must be the next press, it must be the next press. And this is exactly what happens to the rats in the Skinner box.
Annie Duke
The further you get from shore, the deeper the water.
Peter Attia
I know less about this subject today than I did 10 years ago. And they said, how is that possible? And I said, well, to be completely accurate, I mean on a relative basis. Obviously on an absolute basis, I know more than I did 10 years ago about subject Y, but my knowledge of how much broader it is today has dwarfed my absolute knowledge. So on a relative basis, I've gone backwards.
Peter Attia
If I had to say, like, if someone asked me, what's the single biggest factor between a player who doesn't make a ton of progress or make slower progress compared to a player who really, really makes a lot of progress and becomes elite in poker, I would say that's it.
Annie Duke
Uncertainty gives you the leeway to allow bias to come in.
Annie Duke
The longer the time horizon, the more capacity you have, the more leeway the world is giving you to fool yourself.
Annie Duke
2 Protocols
Framework for Learning and Decision Analysis (Four Levels of Thought)
Annie Duke- Level 1: Examine losses primarily through the lens of luck and assume wins are due to skill without deep analysis.
- Level 2: Begin to examine losses through both luck and skill, taking responsibility for one's own actions, but still lean on concordance for wins.
- Level 3: Actively look at all four quadrants of decision outcomes (good decision/good outcome, bad decision/bad outcome, good decision/bad outcome, bad decision/good outcome) and try to examine them equally.
- Level 4: Go beyond concordance on the win side; even if a good decision led to a good outcome, explore if there was an even better 'primary line of play' or a more optimal solution, willing to turn a perceived win into a 'loss' for long-term improvement.
Strategic Planning using Backcasting and Pre-Mortems
Annie Duke- Define a clear goal: State what you want to achieve (e.g., 'I'd like to be able to do X in a year').
- Backcast success: Imagine it's the day after you achieved your goal and look back, asking 'How did I actually achieve this?'
- Perform a pre-mortem: Imagine it's the day after you failed to achieve your goal and look back, asking 'Why did I fail?'
- Identify luck and interventions: Through backcasting and pre-mortems, identify how luck (good or bad) might intervene in the process.
- Increase good luck/decrease bad luck: For positive interventions, explore ways to increase their chances. For negative interventions, explore ways to decrease their chances.
- Develop hedges: If bad luck cannot be fully mitigated, identify and set up hedges against it.
- Plan responses: For unavoidable bad luck, plan in advance what actions to take in response, allowing for calmer, more effective decision-making rather than being reactive.