Friday, March 3, 2017

Adams' Razor

In one of Scott Adams' more viral blog posts, he described how one could determine who was hallucinating when some people observe some phenomenon and others do not. His method is similar to an application of Occam's razor. It states: whichever observation requires an unlikely addition to reality is the hallucination. He provides an example.
If a friend said he could see a pink elephant in the room, standing right in front of you, but you don’t see it, which one of you is hallucinating?
Answer: The one who sees the pink elephant is hallucinating.
And he's correct, but is there any great insight here? It seems to be a simple exercise of conditional probability. Given the odds of an elephant in the room are vanishingly small, as are the odds of there being a pink elephant, the odds of a pink elephant in the room are vanishingly low. So Adams's principle might be stated as:
Whichever of the conflicting observations is least likely is the hallucination.
And that just opens us up to the normal can-of-worms that probabilistic analysis can bring. "Least likely" according to what? Let's look at some of the complexities of applying Adams' Razor in the real world.


The case that Adams is working towards is the divergent opinion in the US during the election as to whether Trump was a racist. While I agree with his conclusion - that those who saw rampant racism were hallucinating - I question his logic in getting there. What was being judged was a pattern of behavior, as opposed to a single concrete observation like an elephant in the room. Even proponents of the racism hypothesis could not point to any clear examples of racism; it was always some subtle pattern of racism that he was communicating.

There is no disputing that some are better at discerning patterns than others. It is the basis of the standard IQ test. Training, intuition, and observation skills also help. What locals see as normal volcanic events the vulcanologist sees as signs of imminent eruption. The shrewd investor sees an emerging bear market that others miss. The ability to see crime patterns others miss is the driving force behind such fictional characters as Sherlock Holmes, Columbo, and Monk. Yet Adam's asserts that if anyone doesn't share the observation of the phenomenon, then it almost certainly is false.

The fact that expertise does play a role in determining whether an observed pattern is authentic is all the wriggle room the Trump haters need to claim superiority. In their view they have a more sophisticated comprehension of racism. They are Sherlock Holmes, and the Trump electorate is the clueless constable. But which is more likely: that college-educated city-dwelling liberals possess powers of discrimination discernment to levels country bumpkins just can't comprehend, or that they are narcissists?

The question becomes: is half of America too inept to see racism that exists, or is the other half so deluded that they see racism that doesn't exist? It's not apparent that we can call one case more probable than the other. Perhaps I am in error by focusing too heavily on the observers rather than the observation. Which is more probable, that Trump is a racist, or that he is not? If the question is if Trump is an open racist, then the answer is obvious. As I mentioned in Religion of the Secular West, racism is the cardinal sin of our modern liberal society. That anyone would be openly racist while running for the presidency would be like vying for the papacy while denouncing Christianity. (Although to be fair the current pope is pretty close to that mark.)

The notion that he was openly racist fails Adams' Razor. Some people do claim that he's openly racist, but anytime I've pushed one of them on the issue they quickly fall back to the assertion that he is a covert racist. So what is more probable: that Trump is a covert racist or that he isn't? For starters we'd have to have some idea of how many people are actually racist and know they are but keep it hidden. How do you measure that? Who's going to confess to secretive racism in a survey? And what of the people who are racist and don't realize it, thinking they're not? White-bashing liberals would fall into this category. All this supposes we have some sort of agreed-upon definition for racism to begin with. Good luck with that. Many on the left purport that only whites can be racist. It uncertain how one might find common ground with blatant logical absurdity.

Determining whether someone is covertly racist is impossible to model because we have no way of knowing how many people are racist to begin with. We don't really know how surprising the addition to reality is. The response one might give is that it doesn't matter how improbable the addition is, just that it is there. But that ignores the implicit addition of the other case. In our example, the media incessantly portrayed Trump as having racial bias. If the addition of Trump racism is false, it implies the addition of massive media bias. Which is more unlikely, that a man is biased, or an entire industry?

The only thing here that is "almost certain not to be real" is our ability to quantify any of this. It all comes down to human intuition. And we are back to where we started. Some people will make an honest conclusion after weighing all the possible evidence, and others will start with the conclusion and concoct supporting evidence and discard all conflicting evidence. Adams' approach assumes the Razor wielder is of the first variety, but those people are hardly the problem.


Perspective can skew the relevant probabilities of the observer.

Adams' probability rule doesn't help here since a giant six lying on the ground is no more likely than a giant nine. Let's look at a more complicated example. In the runup to the election those observing the mainstream media were barraged with propaganda and slander against one candidate. This had an effect on the perception of reality. Voters were less likely to admit they were Trump supporters for fear of social stigma, they were less likely to hold the opinion that Trump would win, and the media was more likely to err on the side of skewing the results to favor his opponent. The resulting disparity of their perspective with reality can be seen here.

How would we begin to apply the Adams formula if we were still in the delusion timeframe? Which is more likely, that there was mass co-ordinated propaganda from America's most trusted news sources, or that Trump was a hopeless candidate? From an Occam's razor viewpoint the latter would be the preferred choice. The model of propaganda and collusion is inherently much more complex than the model of Trump ineptitude. It requires a surprising addition to reality: mass collusion between the media outlets and the Democratic party. I don't know how to organize the situation in a way that Adams' Razor gives the proper outcome.


A person's own biases can skew what they determine to be the addition to reality. For instance Adams asserts that Trump being racist would be an addition to the world. But it is really an addition to his generic view of the world. It reflects his own belief: that the default state of a human is not to be racist. He probably is not racist and projects his own qualities onto the world. However, many in this country believe that there is widespread institutional racism, that all economic inequality stems from oppression, and that whites are inherently privileged and racist. For those people, a white billionaire like Trump being racist would not at all be a surprising addition to their generic take on reality. Tolerance would be the "unlikely addition to their reality."


Advances of human understanding alter the probabilities afforded to different observations. In prehistoric times, a group observation of angry deities hurling thunderbolts during a storm would be considered entirely probable, even inevitable. Today's understanding of weather science would compel us to assign very low probability to such an observation, if any at all. Such understanding is not evenly distributed. There are many who remain superstitious enough to find the observation probable.

Similarly if a group of astronomers observed slight discrepancies in star positions based on the time of year, the assigned probability of the observation being considered accurate would depend on whether the Theory of Special Relativity had yet been discovered.

Thus we should see that Adams' Razor doesn't give us any promise of absolute truth. Probabilities are computed in reference to the observer.


Adam's asserts that if there are two conflicting observations, the one with the unlikely addition to reality is the hallucination. The term hallucination doesn't add much. It could be restated as given two conflicting observations, the one with the unlikely addition to reality is incorrect. We know the term unlikely throws probabilistic monkey wrenches into our attempts at objectivity, and we've shown here that the term addition is similarly prone to be subjective. Ultimately the Adams Razor boils down to something like this.
The more probable observation is almost certainly the correct one.
Not only does the principle not gain us anything, it does away with the benefits of probability to begin with. Adams' Razor was intended as sage wisdom from a self-styled expert of hypnosis and hallucination, but as an epistemological tool it really provides nothing of benefit. It is merely a crude rule of thumb for those with nothing else to go on, not a method for achieving objective truth.

No comments:

Post a Comment