What's the difference between causal and evidential decision theory?
Rosehip Mole asked, and Nib — askNib's tutor — drew the answer live at a whiteboard. This is the spoken transcript; enable JavaScript to watch it drawn.
explain: Causal decision theory (CDT) is a school of thought within decision theory which states that, when a rational agent is confronted with a set of possible actions, one should select the action which causes the best outcome in expectation. CDT contrasts with evidential decision theory (EDT), which recommends the action which would be indicative of the best outcome if one received the "news" that it had been taken.[1]: 7
Decision theory asks: how should a rational agent pick actions? Two schools disagree about what 'best outcome' even means.
CDT says: pick the action that physically causes the best result. Trace the arrows of cause and effect forward from your choice.
EDT says something different: pick the action whose news would be the best evidence about how things already are — even without causing it.
Newcomb's problem shows the split. A predictor filled two boxes based on a forecast of your choice, made before you decide.
CDT says: take both boxes — your choice now can't reach back and change what's already inside. EDT says: take only B — that choice is the best news you'll ever get about being a million-dollar winner.
CDT chases what your action causes; EDT chases what your action reveals — and Newcomb's boxes are where that gap becomes real money.
I didnt get it could you explain with a better example
Let's try something everyday instead of boxes. Notice: people who wake up early tend to be more successful. Should you force yourself to wake up early?
Here's the twist: it's not waking early that causes success. A hidden trait — ambition — causes both. Ambitious people happen to wake early AND happen to succeed.
CDT looks at real causation: forcing yourself to wake early, with no ambition, causes nothing extra — so CDT says don't bother just for that reason.
EDT looks at correlation as evidence: if you find yourself waking up early, that's good news — it's the kind of thing successful, ambitious people do — so EDT treats it as a favorable sign.
So the two theories split exactly on whether an action's link to a good outcome must be causal, or just a reliable clue — that's the whole disagreement.
CDT asks 'does my action make this happen?' while EDT asks 'does my action tell me this is already happening' — same facts, two very different questions.
and how is this related to the box
Same logic, just dressed as boxes. The predictor is like ambition — a hidden cause that already set things in motion before you act.
The predictor already scanned your mind and filled Box B days ago — just like ambition already shaped whether you're an early riser, long before today.
CDT reasons the same way as before: taking both boxes can't causally change what's already sitting inside Box B, so grab both — free extra money.
EDT reasons like the waking-early case: if you're the type of person who one-boxes, that's strong evidence the predictor foresaw it and loaded Box B with the million.
So the predictor plays exactly the role ambition played — a hidden cause behind both your action and the outcome — and that's why the two examples teach the same lesson.
The waking-early story and Newcomb's boxes share one skeleton — a hidden cause behind your action and the payoff — CDT and EDT just answer it in opposite directions.
I feel like I am missing the premise of the box thing
Let's slow way down and lay out the setup itself, step by step, before any decision theory touches it.
There's a Predictor — imagine a superintelligent being who has studied you closely and is almost always right about what you'll choose.
Two boxes sit in front of you. Box A is clear — you can always see one thousand dollars inside it, guaranteed.
Box B is opaque. It either has one million dollars, or it's empty — and here's the key: it was filled BEFORE you make your choice.
You have exactly two options: take only Box B, or take both boxes. The catch — the predictor already decided what to put in B, based on predicting which option you'd pick.
The rule: if the predictor foresaw you'd take only B, it put a million inside. If it foresaw you'd take both, it left B empty — so you'd only get the thousand.
So the puzzle is: knowing that rule, and that the box is already sealed one way or the other, which do you pick — that's where CDT and EDT gave different answers.
That's the full premise: a near-perfect predictor sealed Box B before you act, and your only real choice is one box or both — everything CDT and EDT argue about starts right there.
What is the smoking lesion problem?
This is the classic case built to attack EDT. Imagine a genetic lesion that causes both a craving to smoke AND lung cancer — smoking itself is harmless.
Notice: there's no arrow from smoking to cancer. The lesion is the hidden common cause of both — just like ambition caused both waking early and success.
EDT looks only at correlation: smokers get cancer more often, so smoking is 'bad news' — EDT wrongly tells you not to smoke, even though smoking causes nothing.
CDT looks at actual causation: smoking doesn't cause cancer here, only the lesion does — so CDT correctly says smoke freely if you enjoy it, it changes nothing.
So this problem flips the earlier score: in Newcomb's box EDT looked smarter, but here CDT clearly wins — which is exactly why philosophers still argue over which theory is right overall.
The smoking lesion case is EDT's Achilles heel — it shows that chasing good news instead of real causes can lead you to avoid something perfectly harmless.
More lessons on askNib · Ask your own question