Chapter 9: Causal vs. Evidential Decision Theory Flashcards

1
Q

Beliefs about ______________ ______________ play a significant role in determining what we __________________ think is the ____________ thing to do.

A

causal processes; intuitively; rational

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Example of Casual Processes

A

Newcomb’s Problem

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Newcomb’s Problem

A
  • There is a predictor whose predictions are correct 99% of the time.
  • You are faced with 2 boxes:
    • Box 1 contains $1000 and (You know this, the box is transparent)
    • Box 2 contains either $1M or $0 (the box is not transparent)
  • Your choices are as follows:
    • Alt. 1: Choose BOTH boxes ($1000 and $0 or $1M)
    • Alt. 2: Choose ONLY box 2 ($0 or $1M)
  • If the predictor predicts that you will only choose Box 2, she will place $1M in the box, but otherwise, she will put nothing in there ($0). She knows you know this.
    • REMEMBER, this predictor has been right 99% of the time.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Robert Nozick’s Response

A

Dominance Principle
You should definitely choose BOTH boxes:
- You’ll for sure get $1000 and whatever is in Box 2, be it nothing or $1M.
- The $1M either IS or IS NOT in Box 2. The Predictors placed either $1M or nothing in the box based on her prediction, before you chose a box.
- Her prediction has nothing to do with the fact that $1M is either in the box or not.
- The predictor doesn’t have “magical powers” to adjust circumstances so that $1M appears in Box 2 if you decide to choose only Box 2.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Reply to Nozick’s Approach (See notes for Decision Matrix)

A

Maximizing Expected Utility
Technically, wouldn’t be have a better chance of becoming millionares if we chose ONLY Box 2?
- Think about it, taking into account everything that problem poses, the predictors prediction feasability has consistently been 99%.
- If the decision maker predicted you would choose only Box 2, and hence placed $1M in the box, and you did infact only Box 2, she would be correct.
- It follows that if she predicted you would choose both boxes, and hence placed nothing in Box2, and you selected both boxes, she would be correct.
- This would technically mean we have a higher chance at becoming millionares if we chose only Box 2.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Problem w/ Responses to Newcomb’s Problem

A

We have to different principles of rational decision making telling us to take 2 entirely different courses of action.

The Dominance Principle recommends we take BOTH boxes.
The Principle of Maximizing Expected Utility supposes we should take ONLY box 2.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Source of Misunderstanding 1) and its Response

A

We’re not suppose to know the predictive power of this being. Surely it’s impossible for a single being to be able to make predictions that are 99% correct all the time.

Response: This isn’t helpful. Decision makers find ways of making rationals decisions by IMAGINING that a being liket that could exists. Also, in this proposal, we should accept the predictive power of this being. Let’s we ran an experiment with a select number of people, and because this being is almost certainly right all the time, everyone who 1 Box became millionares, whereas those selected both boxes only receieved thousands of dollars.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Source of Misunderstanding 2) and its Response

A

“The predictor has magical powers.”

Response: No she doesn’t, she’s just a really good predictor.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Causal Decision Theory

A

Decision maker’s beliefs on causal processes should remain fixed in decision making processes, and should choose the best alternative according to their beliefs.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

The ______________ _______________ of the world is going _________________, and can’t be effected by things in the ______________.

A

causal structure; forward; past

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What would Causal Decision Theory tell us to do regarding Newcomb’s Problem, and why?

A

Causal Decision Theory says take both boxes.

  • Based on the predictor’s prediction, she’s either going to place or not place $1M in Box 2. There’s nothing I can do NOW to change the probability that I may recieve $1M (which has already been decided, hence it’s a past event).
  • But, as the rule tells us, I choose the best alternative, and take both boxes because I’ll certainly get $1000.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Review Smoking / Lung Cancer Problem (See notes)

A
  • More to do with Causal Decision Theory.
  • Let’s say you’re a rational decision maker/regular smoker, and a new piece of knowledge surfaces that there is a defect among people that cause Lung Cancer and the drive (impulse) to smoke.
  • Rationally speaking, this shouldn’t change your decision to smoke or not (your causal decision beliefs should remain the SAME).
  • (Ludicrous as it sounds, the problem tells us it’s the defect that causes Lung Cancer, for our purposes, smoking doesn’t actually cause cancer).
  • Any decision / act I make NOW will have no affect on me getting or not getting lung cancer, it’s the defect (state of world).
  • The only thing that smoking does provide is a small benefit of enjoyment that comes with smoking.

Decision Maker:

                                               Defect                                                   No Defect Smoke                  Lung Cancer  /  Enjoyment                     No Lung Cancer / Enjoyment Not Smoke          Lung Cancer   / No Enjoyment               No Lung Cancer  / No Enjoyment
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

To better understand ____________ ______________ theory, we need to ______________ it.

A

causal decision; formalize

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

How to Formalize the statement, “If the decision maker did X, then Y would be the case.”

A

X◻→Y

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What does “p(X◻→Y)” mean? (See Notes for Examples)

A

“probability of decision maker doing X, and getting Y in result is true.”

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Evidential Decision Theory

A

A decision maker shouldn’t make decisions based on “If I did X, then I would get Y as a result,” but instead make decisions based on “If I did X, then I would get Y as a result GIVEN that I did X.”

17
Q

Evidential Decision Theory Formlaized

A

Decision makers shouldn’t make decisions based on p(X◻→Y), but instead p((X◻→Y)|X)

18
Q

Describe Psychopath Situation (See Notes)

A
19
Q

How would Evidential Decision Theory Address Psychopath Situation (See Notes)

A

Paul shouldn’t press the button.
- p(Pressing Button◻→Dead|Pressing Button) is highly likely (0.099), given that everyone who presses the button dies.
- p(Pressing Button◻→Living in a world without psychopaths|Pressing Button) is very unlikely (0.001) considering if Paul presses the button, then he too would be a psychopath, and would hence die – wouldn’t be alive to see a world without psychos.

20
Q

How would Evidential Decision Theory Address Newcomb’s Problem

A

Choose Box 2
- p(taking both boxes◻→$1M|taking both boxes) is pretty low (0.01)
-Whereas p(taking only Box 2◻→ $1M|taking only Box 2) is pretty high (0.99)

21
Q

Problem w/ Evidential Decision Theory

A

Decision maker assigns probability to their own choices / decisions.

22
Q

Why one has to assign probabilities to their own choices/ acts they plan to take? (See Notes for examples)

A

p(X|Y) = p(X⋀Y) / p(Y)

23
Q

What is so inconcistent about a decision maker assigning probabilities to their own choices/decisions? (2)

A

Answer 1: if these probabilities are taken to be objective probabilities, then it wouldn’t make sense that these probabilities are based on any deliberative process (propogated by me) AND a probabalistic process that would govern/guide my decisions anyways.

Answer 2: subjective probability isn’t any better. The Decision maker will express a willingness to accept a bet (considering they’d accept any bet that would keep them from losing). This willingness to accept the bet will be used to measure the proability of an outcome happening. The decision maker will then assign subjective probability to this bet, such that they side with this bet (action they have to take) because they prefer it. So, the agent now has an incentivized bet, and this distorts the agent’s real preferences because they may have otherwise chosen a different bet (but they’ll ALWAYS choose the incentivized one to avoid losing).

24
Q

Death in Damascus

A

Preface: Here’s how death works
- he works by appointment
- predictst that you’ll be in a specific place at a specific time
- makes a book predicting your death / plans weeks in advance.”

Story:
- Man in Damascus is approached by Death. Death was suprised to seem him.
- Death: “I’m coming for you.”
- Man leaves Damascus in fear and heads for Appelo.
- Man finds and stays in hiding place in Appelo.
- Death: “Hey I’m here.”
- Man: “I thought you were expecting me in Damascus?!”
- Death: “Oh, no I predicted you would be here, it’s why I was suprised to see you the other day in Damascus.”

If he the man figured/predicted he’d be in Damascus on the day of his appointment, he would’ve fled to Appelo.
If he predicted he’d be in Appelo on the day of his appointment, he would’ve fled to Damascus.
No matter what man could’ve done, both options, going to Damascus or going to Appelo had about the same utility.

25
Q

The man was right to think his decision to got to ______________ or _____________ was _________________ indepdnent of Death’s prediction about where he would be the day of his appointment.

A

Damascus; Appelo; causally

26
Q

Why couldn’t causal or evidential decision theory be applied to the problem of Death in Damascus?

A

There was not a stable act for the man:

Stable Act: an act is stable if and only if, when it’s performed, there it’s utility is maximal.

The man didn’t have an option that was of the most utility.

27
Q

How is Death in Damascus similar to Newcomb’s Problem?

A

The decision maker’s (me/the man) decision was causally independent of the prediction, yet the prediction (made by being/death) was usually right.