Causality, wisdom of the crowd and scary thoughts

Prerequisite: there is no such thing as a biased coin.

Theorem (Central limit theorem). Let \(X_1, \dots, X_n\) be independent and identically distributed random variables with mean \(0\), then

Where \(std(X)\) is just a standard deviation of \(X\).

I hope you vaguely remember how it works: take any experiment, repeat it many times independently and Gaussian distribution, \(\mathcal{N}\), comes out. Magic.

Now let’s try to apply it. The following is known as Wisdom of the crowd, although it is also known as Emperor’s height fallacy:

Supposing that each person in China surely knows the height of the Emperor, \(h\), to an accuracy of at least \(\pm 1\) meter; if there are \(N = 1 000 000 000\) inhabitants, then it seems that we could determine his height to an accuracy at least as good as \(1/\sqrt{1 000 000 000}\) m = \(3 × 10^5\) m = \(0.03\) mm, merely by asking each person’s opinion and averaging the results.

E.T. Jaynes

So what is going on with our misapplication of CLT?

Let us first figure out what independent assumption means. The formal definition is the following:

So, knowing the value of the first experiment doesn’t change the probability of the second experiment.

However, this is where the way you perceive probability starts to matter.

I. Orthodox statistics

If your brain was twisted by modern teachings of probability theory you start thinking that there’s something random in this universe which drives the outcomes of the experiments.

Take flip of a coin. If you say that coin has an intrinsic probability \(p\) of coming up heads, you say that the flips of a coin are driven by some random process and each experiment is a realisation of such randomness, where random process is something that can not predicted by anyone in this world.

If you think I am straw manning, I am not. And it’s no surprise this idea was not favoured by physicists.

Question: are guesses of two people (who didn’t not communicate to each other) independent?

If the first guess, \(X_1\), was simply a realisation of some random process, be it positions of the atoms in the universe or anything else, it can not predispose another person to guess anything related to \(X_1\). If it could, then what you are effectively saying is that there’s a physical causal effect between individual guesses. So mutterings of the first respondent changes the minds of the respondents afterwards. Nonsense as there is no such thing as telepathy. (No communication between respondents is assumed.)

Therefore, in orthodox statistics, individual guesses are independent.

This conclusion is absurd: if the respondents keep overestimating the Emperor’s height, you should start taking this data into account: as it is likely that the next guess will also be over. But not in orthodox stats, where independence is the same as “no physical cause”.

II. Physical causality contrasted with logical causality

Bayesian dependence is also known as logical causality. Examples, where orthodoxy treats events as independent, but Bayesians treat them as dependent, include:

Coin tosses

Orthodoxy: probability is a property of a coin, tossing a coin once won’t change the coin (unless you damage it), so first toss tells us nothing about the outcome of the second toss.

Bayesians: probability is in the head, so the first toss will reveal to me useful information about the propensity of a coin to come up heads and I will adjust my probability with each toss.

Future events

Orthodoxy: probability of a die doesn’t change even if I am told about the future outcomes, because probability is a property of a die, and unless you deform the die, nothing can change.

Bayesians: probability is in the mind, I can condition on future events and this will change the way I predict present events. No axioms of probability theory refer to time at all, so there’s no reason to not do it.

Responses on a survey

Orthodoxy: individual responses don’t affect each other as long as people are not allowed to communicate since telepathy doesn’t exist.

Bayesians: my ability to predict a response of the next person improves with more data. If I detected a systematic error in the previous responses, it is likely due to some underlying cause, such as folklore/media, this error is likely to be present in the future responses.

Therefore, the solution to the Emperor’s paradox follows:

The absurdity of the conclusion tells us rather forcefully that the \(\sqrt{N}\) rule is not always valid, even when the separate data values are causally independent; it is essential that they be logically independent. In this case, we know that the vast majority of the inhabitants of China have never seen the Emperor; yet they have been discussing the Emperor among themselves, and some kind of mental image of him has evolved as folklore. Then, knowledge of the answer given by one does tell us something about the answer likely to be given by another, so they are not logically independent. Indeed, folklore has almost surely generated a systematic error, which survives the averaging; thus the above estimate would tell us something about the folklore, but almost nothing about the Emperor.

E.T. Jaynes

Game theory: Prisoners’ dilemma

Orthodoxy: my decision is physically independent of my opponent’s decision as there is no telepathy, therefore I should just defect regardless of anything else.

Bayesians: my decision is anthropomorphic evidence of how humans behave in this situation, so the way I model the opponent depends on what I decide to do. So there are situations when I shouldn’t defect, since my cooperation makes it more likely that a similarly-thinking opponent would also cooperate.

Game theory: Newcomb’s paradox

Orthodoxy: my decision physically can not affect the prediction, therefore taking both boxes is the optimal thing to do.

Bayesians: my decision provides me with some evidence on the prediction, therefore I should sometimes one-box with a credible predictor.

III. Scary thoughts, finally

If you subscribe to a Bayesian point of view, and chances are you do as people don’t bother reading what they disagree with in the first place, you now have an all-encompassing, timeless (logical) causality governing your every day life.

Every time you procrastinate doing X, chances are, you will procrastinate doing X again in the future. Arguing with this is not taking evidence into the account. So you only thought your distraction was going to take a minute… Boom now months of your future life are being wasted. That’s if you don’t break out of the bad habits. However, every time you give in you are making it less likely that you ever will.

If you are a smoker and you pledge that this will be your last cigarette, chances are it won’t be. And it won’t be your last bet, if you are a gambler.

You remember Roko basilisk?

If you don’t, there’s a short summary. Somebody “clever” posted this on LW: if you believe that strong AI will be developed within your lifespan, that AI might decide to torture all the people who were aware of such possibility but didn’t contribute enough to bring its development forward because a delay in AI development costs millions of lives (strong Friendly AI is likely to discover immortality). So now all of you, readers, should go and dedicate your every living moment facilitating AI development.

Yawn. But then Eliezer Yudkowski freaked out and deleted this whole affair and banned such discussions on LW all together.

When I first heard about such reaction I scoffed at it. It appeared bizarre that Eliezer treated such idea with fear. However, after understanding timeless causality on a deeper level, it makes sense. If you managed to convince a person to believe in some kind of version of Roko’s basilisk than he’ll live in fear and misery slaving away everyday. Arguably, some religions already do just that: promise some kind of after life. But we are talking about really “smart” individuals here slaving away because of the timeless causality.

So if you already subscribe to timeless causality philosophy, you might want to be careful about what new pieces of knowledge you want to acquire…