Counterfactual Regret Minimization for Life
Run poker's solver on your calendar: pick the behavior your future self will least regret.
- Difficulty
- Advanced
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 88%
Perkins ports counterfactual regret minimization — the technique behind poker and chess solvers — from the table to life. In poker, CFRM asks which decision maximises the probability of winning across a series of choices. In life, Perkins reframes the objective function: which behavior or set of behaviors leads to the least regret in the future, given your values? The algorithm is deliberately value-agnostic — it takes your values as inputs and returns your behavior, so a single-person's answer and a father's answer can both be correct. Perkins built it because his greatest fear is not running out of money but wasting his life or any particular period of it.
Origin
As a trader and serious tournament poker player, Perkins used counterfactual regret minimization models to make positive-expected-outcome bets under imperfect information. He noticed his deepest fear wasn't financial loss — it was wasting a period of his life — and set out to build a CFRM algorithm that solved for not wasting his life rather than for chips.
Core principles
- 01CFRM is what-if analysis: which decision leads to the best outcome across a series of choices.
- 02In poker you solve for expected value; in life you solve for minimum future regret.
- 03The algorithm is value-agnostic — you plug in your values and it returns your behavior.
- 04Each domain has its own solve target: money at work, meaning in relationships, lifespan in health.
- 05The real scarce asset is life, not money — you can waste money, you cannot un-waste a decade.
How to run it
- 1
Pick the domain and name the solve target
State explicitly what you are optimising in this domain: most money at work, the most meaningful relationship at home, optimal lifespan in health.
Pro tip Different domains have different objective functions — don't run the work solver on your family.
- 2
Plug in your values
The algorithm is value-agnostic. Supply your actual values as inputs — what you want out of your life, not what you think you should want.
Watch out Borrowed values produce borrowed regrets.
- 3
Lay out the branches
Enumerate the candidate behaviors as a what-if tree: play poker with the fellas or walk with your daughter; visit your mother or rage in Vegas.
- 4
Score each branch by future regret
For each branch ask whether your future self will regret this earlier self's action. Choose the path with the least anticipated regret.
Pro tip Add counterweights — Perkins won't eat nasty food for optimal lifespan, because that's a regret too.
- 5
Re-run as your inputs change
Values, health and life stage all shift. Re-run the solve rather than freezing a decision rule made by a self who no longer exists.
In the wild
Perkins uses deliberately stark small decisions as illustrations: play poker with the guys or take a walk with his daughter; call and visit his mother or go raging in Vegas. Each is a branch scored against future regret rather than immediate appeal.
→ The framework resolves ordinary calendar decisions by asking which choice his future self will be angry about, converting vague guilt into a decision rule.
When you play a chess computer, Perkins notes, it is running a CFRM for not being checkmated and for checkmating you — a series of what-if branches evaluated against one objective.
→ This grounds the transfer: the same solver structure works on life once you swap the objective function from winning to not wasting your life.
Common mistakes
Solving for money in every domain
Applying the work objective function to health, family and leisure produces a technically-optimised life that maximises the wrong variable.
Running the algorithm without values
CFRM is value-agnostic by design — with no explicit inputs it silently defaults to whatever the culture or your habits supply.
Is it for you?
Best for
Analytically-minded people who already optimise rigorously at work but run their personal lives on autopilot.
Not ideal for
People who haven't done the work of naming their values — the algorithm returns garbage without inputs.
From the transcript
“in life it's for me what I look at is what is the behavior or set of behaviors that is going to lead to the…”
“my biggest fear is not running out of money uh you can tell by the risk I've taken... the biggest fear I have is wasting…”
“basically make it value agnostic so what are your values what's the algorithm right plug in your values”
“I can waste money I just don't want to waste my life”
From the episode
Bill Perkins: I’m Planning to Die with Zero Dollars in the Bank and You Should Too
Bill Perkins