Newcomb's Problem and the tragedy of rationality

I describe my favorite paradox, "Newcomb's Problem," the related "Parfit's Hitchhiker" dilemma, and what they reveal about rationality.

[meatrobot]

"OK murderously greedy samaritan, I'll take the ride and give you your blood money. But it's in a clear box in a tent in a carnival."

[darkranger]

me wanting 1, 000$ and two boxes for my cats : *its free real-estate*

[danielraju]

Imagine the Buddha walking into the tent, the thought reader goes into a recursive infinite loop. The Buddha looks at the Scientist, smiles and says, 'The root of all suffering is desire'.

[jimnesstarlyngdohnonglait]

Think more the Hardest way as how Images are added to web pages using the <img> tag
Vs
Think more the Hardest way as how Definition list helps to define lists with names

[jimnesstarlyngdohnonglait]

Think more the hardest way as *•A Sorry sight:* which is a regrettable and unwelcome aspect or feature.
Vs
Think more the hardest way as *•Acid test:* which is a sure test, given an incontestable result

[SlimThrull]

1:52 I flip a coin. Heads, I take both boxes. Tails, I take the one box. Since the machine wouldn't be able to tell which box I was choosing, I'd have a 50/50 shot of getting a million dollars.

[jcbarendregt]

The second example reminds me of a Seinfeld joke about getting the check for dinner. “Why would I pay for this. I’m not hungry now. I just ate”

[UteChewb]

Parfit's Hitchhiker suggests to me that for rational beings in real life the evolution of trust and reprocity is vital, also the social thing called honour. This avoids the tragedy of rationality in real life by such rational agents committing to an agreement. Sticking to the agreement increases your social standing via reliability, and that improves your future survival.

[jimnesstarlyngdohnonglait]

Think more the Quickest Intuitively way as A is for *Accomplish* again
Vs
Think more the Quickest Intuitively way as B is for *Bold* again

[ChibiRuah]

Honestly the rational act vs rational character makes a lot of sense. One seems to tackle choices the moment maximizing in the moment while one is fixed and fated to make choices ahead of time to maximize expected output overall.

[oriongurtner]

Answer to Parfit’s Hitchhiker: have him drive to Newcomb’s Carnival Tent

He’ll get his 1000 dollars

[greenman]

The obvious answer is to: *Take the Brain Scanning Device* and make untold *Billions* of dollars.

[hdthor]

The Newcomb problem with a predictor which is 90% accurate is equivalent to a problem where there is a 90% probability that an evil time-traveler goes back to change the payout of the unknown box. A rational actor who knows there’s a 90% chance they will be punished for greed by an evil time traveler will choose to play non-greedily.

This is also similar to the class of iterative game problems which is different from the class of single game problems. Humans as social creatures are geared toward acting optimally/rationally in iterative game problems, which may look irrational under the lens of a single game.

A famous example is a game with 2 players and $20. Each player can vote to split or vote to steal. If they vote to split, they each get $10. If one votes to steal, he gets $20 and the other gets $0. If both vote to steal, then both get $0. A rational player will always choose to steal because it either increases his reward (from $10 to $20 if the other player chooses to split), or keeps his reward the same (stays at $0 if the other player chooses to steal). Now, here’s the interesting part: the game gets an additional “punish” action, where any player who is stolen from can pay out of his own pocket $50 to penalize the other player $200.

No rational actor would ever choose to punish in such a way because they would lose $50, and a rational actor only looks selfishly at his own interest, not in whether others are rewarded or punished. However, in iterative game theory, where you will again and again play with that same actor, or with other actors in his clan who might learn from his experiences, it’s important to reward or punish them to train their behavior. So for $50 cost you train the other player (punishing him $200, which will re-weight his neural network) to make him act fairly in future games and build your reputation as someone who won’t tolerate unfair treatment.

This is the element often missing in game theory: adding an option to rewrite the opponent’s neural network for future games. Hustlers are very good at this, they will intentionally lose a game of billiards over a bet of $20, and then ask for a $100 game, hoping you think they stink, when really they’re hustling you and will certainly win. Poker is very similar as well, as you can train your opponents over the course of many hands.

Iterative games are way more complicated than single games. And games with evil time-travelers are just iterative games in disguise. Instead of time-traveling, the predictor could merely observe how you played past games which builds up your reputation for being either greedy or not. And then you play a million games, each time under amnesia, but only one is with real money but you don’t know which. And the predictor in each game has memory of how you played all the past ones.

[jimnesstarlyngdohnonglait]

Think more when I learnt how not to forgive...
Vs
Think more when I learnt how forget

[JimnesstarLyngdohNonglait]

Think more the quickest intuitively way as D is for *Disbursement* again
VS
Think more the quickest intuitively way as D is for *Depreciation* again

[DifferentName]

Thinking about this problem convinced me to vote. For years, causal decision making lead me to skip voting because the chance of my vote making a difference is approximately zero. But this kind of reasoning leads to a world where a large number of intelligent rational people don't vote, which I suspect does make a difference. I would prefer to live in a world where intelligent rational people vote, so now I act accordingly.

[PhiTheHuman]

You have just become one of my favourite teachers on Youtube.

Thankyou, for being you, for simply doing what you do, and for it happening to be exceptional in quality. Good work. <3

[thomaskember]

When I was at university, whenever studying logic came up, I always had a headache. This video illustrates why.

[jimnesstarlyngdohnonglait]

Think more the hardest way when I'm wise enough to know that even in the midst of hardest situation that I may face alone...
Vs
Think more the hardest way to not be fool by unusual myths of dreaming while I sleep at night

[jimnesstarlyngdohnonglait]

Think more as Q is for Quarrelling
Vs
Think more as R is for Roaring