How do you make decisions?

[DataPacRat]

Here are a pair of philosophical puzzles; I'm interested in knowing what answers you would pick for them, why, and if you see any conflict between your answers.


The first problem:

Imagine that a superintelligence from another galaxy, whom we shall call the Predictor, comes to Earth and at once sets about playing a strange and incomprehensible game. In this game, the superintelligent Predictor selects a human being, then offers this human being two boxes. The first box, Box A, is transparent and contains a thousand dollars. The second box, Box B, is opaque and contains either a million dollars or nothing. You may take only box B, or you may take boxes A and B. But there's a twist: If the superintelligent Predictor thinks that you'll take both boxes, the Predictor has left box B empty; and you will receive only a thousand dollars. If the Predictor thinks that you'll take only box B, then It has placed a million dollars in box B. Before you make your choice, the Predictor has already moved on to Its next game; there is no possible way for the contents of box B to change after you make your decision. If you like, imagine that box B has no back, so that your friend can look inside box B, though she can't signal you in any way. Either your friend sees that box B already contains a million dollars, or she sees that it already contains nothing. Imagine that you have watched the Predictor play a thousand such games, against people like you, some of whom two-boxed and some of whom one-boxed, and on each and every occasion the Predictor has predicted accurately. Do you take both boxes, or only box B?

The second problem:

You know that you will shortly be administered one of two sodas in a double-blind clinical test. After drinking your assigned soda, you will enter a room in which you find a chocolate ice cream and a vanilla ice cream. The first soda produces a strong but entirely subconscious desire for chocolate ice cream, and the second soda produces a strong subconscious desire for vanilla ice cream. By "subconscious" I mean that you have no introspective access to the change, any more than you can answer questions about individual neurons firing in your cerebral cortex. You can only infer your changed tastes by observing which kind of ice cream you pick.

It so happens that all participants in the study who test the Chocolate Soda are rewarded with a million dollars after the study is over, while participants in the study who test the Vanilla Soda receive nothing. But subjects who actually eat vanilla ice cream receive an additional thousand dollars, while subjects who actually eat chocolate ice cream receive no additional payment. You can choose one and only one ice cream to eat. A pseudo-random algorithm assigns sodas to experimental subjects, who are evenly divided (50/50) between Chocolate and Vanilla Sodas. You are told that 90% of previous research subjects who chose chocolate ice cream did in fact drink the Chocolate Soda, while 90% of previous research subjects who chose vanilla ice cream did in fact drink the Vanilla Soda.


Which ice cream would you eat?

 

[3sigma]

Which ice cream would you eat?

I don’t see how I could answer that question with any certainty. The second problem states that my ice cream choice is controlled mainly by which soda I am assigned and I have no choice in or knowledge of that assignment.

 

[Eudaimonist]

These sound merely like game theory problems, not the sort of interesting decisions people are likely to face day to day, which are more about means-end reasoning, prioritizing one's values, deciding what values to include or exclude, deciding which virtues apply to the situation, budgeting one's time, measuring progress to long term goals, and selecting the long term goals themselves. What is being measured in game theory seems to be a narrow range of cleverness, not practical wisdom.

My serious answer to ask myself if I should even play the game to begin with. Where did the money come from? Was it obtained ethically? Should I be homo economicus or homo sapiens?

But let's say that I decide to play.

The first problem:

Given that Predictor always predicts accurately (or seems to), then the choices/outcomes are limited to these:

1) Choose box B -- Predictor predicts this, and the payoff is $1,000,000
2) Choose boxes A and B -- Predictor predicts this, and the payoff is $1000

I choose (1) and collect my $1,000,000. I thank Predictor for being so generous.

The second problem:

1) Eat vanilla ice cream -- Payoff is $1,000 (plus random amount out of my control)
2) Eat chocolate ice cream -- Payoff is $0 (plus random amount out of my control)

I choose (1) and walk off with my $1,000, plus the amount I got automatically with the random soda selection. I take the cute research assistant out for an ice cream -- the kind I actually wanted.

This assumes that I can choose against my subconscious preference, and that my ability to choose hasn't been destroyed by the soda, in which case I would have no choice whatsoever and would have to accept any results that occur.

It also assumes that it wasn't actually worth it to me to lose $1,000 to taste chocolate ice cream if I were to drink the chocolate-craving soda. After all, in that case I've just won a million dollars, and a thousand dollar ice cream craving is just the sort of luxury I could afford.

In both cases, I ask myself: what difference in outcome does my choice produce? Which possibilities are out of my control?


eudaimonia,

Mark

 

[ChrisLeishy]

question 1

A super intelligent life form gives a human an incomprehensible game. Everything after that word is irrelevant as you are asking something to comprehend something incomprehensible. The human or the life form cannot continue with out comprehension of what they are doing as intelligence requires comprehension. Actions without intelligent comprehension is just random movement.

How-logic.

question 2

Both contain free ice cream eat up, anything else is a bonus.
reason- Experience. Never knock back a free lunch. Get the ice cream before he changes his mind. That is guaranteed.


Just for fun...

Old proverb-

You never really know about a man until you walk a mile in his shoes.

from this you gain 2 very important things.

1 Empathy and understanding
2 A new pair of shoes with low mileage.

 

[Wiccan_Child]

"I would take just one box in the first problem and eat chocolate ice cream in the second problem"

I can only presume the superintelligence knows my logic, and thus knows I will pick Box B. I would pick Box B because it contains $1,000,000 if the being predicts correctly, and it predicts correctly if it follows my logic. Being superintelligent, it is capable of deducing the same things I am.

As for the chocolate ice cream, I thought that only those who ate their corresponding ice-cream get the bonus - so vanilla-drinkers who ate vanilla get $1000, but chocolate-drinkers who ate vanilla would get nothing.

In hindsight, I agree with Eudaimonist: take the $1000. The two payouts are unrelated events. The sheer fact that the first event (drinking soda) gives us an urge to eat a certain type of ice-cream won't overcome the urge to get $1000, and the $1000 can be attained without jeopardising the $1,000,000 we may or may not be getting. Thus, eat vanilla, and hope for chocolate soda.

I also agree that these are exercises in game theory, not philosophy.

 

[SithDoughnut]

What Eudaimonist said; there's no point re-phrasing it in another post.

 

[Chesterton]

How do you make decisions?

 

[DataPacRat]

I don’t see how I could answer that question with any certainty. The second problem states that my ice cream choice is controlled mainly by which soda I am assigned and I have no choice in or knowledge of that assignment.

Not exactly; the second problem says that there's an /observation/ that most people who drink the chocolate soda end up /choosing/ chocolate ice cream, and vice versa... not that there's no choice being made.

These sound merely like game theory problems, not the sort of interesting decisions people are likely to face day to day,

I also agree that these are exercises in game theory, not philosophy.

If you like, then of them as thought experiments, trying to gather evidence to come up with a theoretic model for /how/ people make decisions of various sorts.

In both cases, I ask myself: what difference in outcome does my choice produce? Which possibilities are out of my control?

And that is a very interesting question to ask yourself.

The first problem is called "Newcomb's Problem" or "Newcomb's Paradox", and has led to the creation of two major philosophical camps, "causal decision theory" and "evidential decision theory". A quickie summary of each:

The evidential theorists would take only box B in Newcomb's Problem, and their stance is easy to understand. Everyone who has previously taken both boxes has received a mere thousand dollars, and everyone who has previously taken only box B has received a million dollars. This is a simple dilemma and anyone who comes up with an elaborate reason why it is "rational" to take both boxes is just outwitting themselves. The "rational" chooser is the one with a million dollars.

The causal theorists analyze Newcomb's Problem as follows: Because the Predictor has already made Its prediction and moved on to Its next game, it is impossible for your choice to affect the contents of box B in any way. Suppose you knew for a fact that box B contained a million dollars; you would then prefer the situation where you receive both boxes ($1,001,000) to the situation where you receive only box B ($1,000,000). Suppose you knew for a fact that box B were empty; you would then prefer to receive both boxes ($1,000) to only box B ($0). Given that your choice is physically incapable of affecting the content of box B, the rational choice must be to take both boxes - following the dominance principle, which is that if we prefer A to B given X, and also prefer A to B given ~X (not-X), and our choice cannot causally affect X, then we should prefer A to B. How then to explain the uncomfortable fact that evidential decision theorists end up holding all the money and taking Caribbean vacations, while causal decision theorists grit their teeth and go on struggling for tenure? According to causal decision theorists, the Predictor has chosen to reward people for being irrational; Newcomb's Problem is no different from a scenario in which a superintelligence decides to arbitrarily reward people who believe that the sky is green. Suppose you could make yourself believe the sky was green; would you do so in exchange for a million dollars? In essence, the Predictor offers you a large monetary bribe to relinquish your rationality.

But the /fascinating/ thing... is that evidential theorists would take box B, and go for chocolate ice cream; while the causal theorists would take both boxes and go for vanilla ice cream.

But, even with just a few votes so far, we're seeing some people aren't obeying the predictions of /either/ form of decision theory - according to either causal or evidential decision theory, they are making inconsistent choices. What do you think the implications of this might be?

 

[Wiccan_Child]

If you like, then of them as thought experiments, trying to gather evidence to come up with a theoretic model for /how/ people make decisions of various sorts.

The logical, rational decision is dictated by game theory. It's mathematics, nothing more. Unless the point is to see if someone makes decisions mathematically to maximise potential winnings, or by some other, less rational method.

And that is a very interesting question to ask yourself.

The first problem is called "Newcomb's Problem" or "Newcomb's Paradox", and has led to the creation of two major philosophical camps, "causal decision theory" and "evidential decision theory". A quickie summary of each:

But the /fascinating/ thing... is that evidential theorists would take box B, and go for chocolate ice cream; while the causal theorists would take both boxes and go for vanilla ice cream.

But, even with just a few votes so far, we're seeing some people aren't obeying the predictions of /either/ form of decision theory - according to either causal or evidential decision theory, they are making inconsistent choices. What do you think the implications of this might be?

Why is this inconsistent? We've given our reasons for our decisions, where is the inconsistency there? It's ultimately the choice that maximises personal gain, or is the 'winning' scenario. In the first case, this is to take Box B, in the second case this is to eat vanilla ice-cream. I honestly don't see the inconsistency. Perhaps you've incorrectly predicted what causal and evidential theorists would take? Or perhaps we aren't either causal or evidential theorists?

Take the 'proof' that evidential theorists won't take both boxes: "anyone who comes up with an elaborate reason why it is "rational" to take both boxes is just outwitting themselves". That's hardly an airtight piece of reasoning. I'm inclined to favour the causal theorists simply because the summary seems so biased against them.

 

[DataPacRat]

Or perhaps we aren't either causal or evidential theorists?

Congratulations; you've just proven you're smarter than most university philosophy departments.

A third form of decision theory has been described, called 'timeless decision theory', which leads to the choices you describe as 'rational'... and for much the same reasoning as you describe, come to think of it... but nearly everyone who's studied Newcomb's paradox has ended up as either an evidential or causal theorist. And make choices in such problems consistent with their chosen camp's theory... while rationalists/timelessians manage to outperform them both (outside of pathological problems involving superbeings who hate rationalists and deliberately punish them).

 

[3sigma]

Not exactly; the second problem says that there's an /observation/ that most people who drink the chocolate soda end up /choosing/ chocolate ice cream, and vice versa... not that there's no choice being made.

The second problem states: The first soda produces a strong but entirely subconscious desire for chocolate ice cream, and the second soda produces a strong subconscious desire for vanilla ice cream. Within the context of the problem, that isn’t stated as an observation: it’s stated as a fact.

However, if we disregard that then I’m with Eudaimonist. I would choose those things under my control to maximise my gain (assuming no unstated counter conditions) so I would choose Box B alone and the vanilla ice cream.

The difference between the two problems is that in the first, whether you receive the million dollars is predictable whereas in the second, it is random.

 

[Eudaimonist]

But, even with just a few votes so far, we're seeing some people aren't obeying the predictions of /either/ form of decision theory - according to either causal or evidential decision theory, they are making inconsistent choices. What do you think the implications of this might be?

Are you saying that I made inconsistent choices? I thought I was being perfectly consistent to my own pattern of reasoning.

Incidentally, why would the evidentials go for the chocolate ice cream?


eudaimonia,

Mark

 

[Wiccan_Child]

Congratulations; you've just proven you're smarter than most university philosophy departments.

I'm a physicist dear, of course I am

(jokes )

A third form of decision theory has been described, called 'timeless decision theory', which leads to the choices you describe as 'rational'... and for much the same reasoning as you describe, come to think of it... but nearly everyone who's studied Newcomb's paradox has ended up as either an evidential or causal theorist. And make choices in such problems consistent with their chosen camp's theory... while rationalists/timelessians manage to outperform them both (outside of pathological problems involving superbeings who hate rationalists and deliberately punish them).

I've used that myself as an argument against God, or at least worship of God: if God is irrational, then any hope of salvation is gone. He says he lets believers in Heaven and lets non-believers burn in Hell... but if he's irrational, who's to say he won't do the reverse at the drop of a hat? Even if God could be shown to exist, if he's an irrational being, he's not worth worship

 

[DataPacRat]

Are you saying that I made inconsistent choices? I thought I was being perfectly consistent to my own pattern of reasoning.

I'm not saying you're being inconsistent; I'm saying that, according to the theories of either of the two main camps of decision theory, the choices you make are inconsistent.

Incidentally, why would the evidentials go for the chocolate ice cream?

According to the author I've been quoting:

An evidential decision agent facing Newcomb's Soda will, at the time of confronting the ice cream, decide to eat chocolate ice cream because expected utility conditional on this decision exceeds expected utility conditional on eating vanilla ice cream.

 

[Eudaimonist]

I'm not saying you're being inconsistent; I'm saying that, according to the theories of either of the two main camps of decision theory, the choices you make are inconsistent.

I see. Noted.

Personally, I think they are focusing too much on the thought processes of the abstraction homo economicus (embodied perhaps in "Sheldon" from Big Bang Theory, who is exceptionally clever and mathematically inclined, but obsessive, unwise, and amoral or nearly so), and not enough on homo sapiens. I find that I simply can't relate to these "clever" solutions as having much to do with the way in which psychologically healthy people reason from day to day.

I might comment on this more in a later post. I'll just mention now that it is tempting to consider these thought experiments as happening in their own self-contained universes (which exist for just as long as the experiment), but any real examination of how people make decisions should take place in the real world throughout one's entire life. IOWs, one should imagine that the experiment is happening in real life today, and that life will continue on afterwards. I think that many more reasoning considerations will be relevant, which may complicate the examples, but necessarily so.

An evidential decision agent facing Newcomb's Soda will, at the time of confronting the ice cream, decide to eat chocolate ice cream because expected utility conditional on this decision exceeds expected utility conditional on eating vanilla ice cream.

This explanation is cryptic to me.


eudaimonia,

Mark

 

[Wiccan_Child]

I see. Noted.

Personally, I think they are focusing too much on the thought processes of the abstraction homo economicus (embodied perhaps in "Sheldon" from Big Bang Theory, who is exceptionally clever and mathematically inclined, but obsessive, unwise, and amoral or nearly so), and not enough on homo sapiens. I find that I simply can't relate to these "clever" solutions as having much to do with the way in which people reason from day to day.

I might comment on this more in a later post.



This explanation is cryptic to me.

I think it's just a fancy-pants way of saying "eating chocolate ice-cream is better".

'Expected utility' is the benefit or boon from eating a given ice-cream - it is 'conditional' on eating that ice-cream, and it is the utility (or boon) we expect to receive. Thus, the 'expected utility' conditional on eating chocolate ice-cream exceeds that expected on eating vanilla ice-cream.

In other words, chocolate yields better results.

 

[Eudaimonist]

In other words, chocolate yields better results.

What results?


eudaimonia,

Mark

 

[Wiccan_Child]

What results?

In general, any and all results. Here specifically, the results are money and free ice-cream.

 

[Eudaimonist]

In general, any and all results. Here specifically, the results are money and free ice-cream.

Did I misread the problem? I though it was vanilla ice cream that produces the extra (choice-influenced) payoff of $1,000, and chocolate ice cream gets you no extra money.

But subjects who actually eat vanilla ice cream receive an additional thousand dollars, while subjects who actually eat chocolate ice cream receive no additional payment. You can choose one and only one ice cream to eat.

So why would anyone eat chocolate ice cream assuming that what people really want is the money?


eudaimonia,

Mark

 

[Wiccan_Child]

Did I misread the problem? I though it was vanilla ice cream that produces the extra (choice-influenced) payoff of $1,000, and chocolate ice cream gets you no extra money.


eudaimonia,

Mark

That's the point, though. Does eating vanilla ice-cream result in you having more money? The original quote, which you said was cryptic and which I was trying to decrypt, states that:

"An evidential decision agent facing Newcomb's Soda will, at the time of confronting the ice cream, decide to eat chocolate ice cream because expected utility conditional on this decision exceeds expected utility conditional on eating vanilla ice cream."

That is, an evidential theorist would eat chocolate, not vanilla. They see people eating chocolate ice-cream and getting $1,000,000, so they too eat chocolate ice-cream, hoping to get $1,000,000.

In my opinion, the problem is poorly/vaguely worded, and the explanation of the evidential theorist's decision doesn't do justice to the theorist - it's basically a strawman.