User:Tmtoulouse/Neurocrankism

From RationalWiki
Jump to navigation Jump to search

Neurocrankism[1] is the study of cranks using behavioral and cognitive psychology, computational modeling, and neuroimaging. Cranks inhabit every imaginable field of human discovery, from the extremely technical or obscure to the mundane and ubiquitous. At first glance it might be difficult to see how a unified field of study could be applied to the diversity that is crankdom. However, an amazing property of the crank makes it relatively easy. Whether the crank is arguing for an algebraic proof of Fermat's last theorem or that their was no plane that hit on 9/11 and all the supposed passengers are on an island somewhere--the same style of argument, the same approach to evidence, the same views of the status quo, the same martyrdom complex, heck even the same font choices unite every single piece of their writing. It is often very easy to take to the writings of a crank in two very divergent fields and merely switch the topic between them with a cut-and-paste and still wind up with a nearly identical raving. This commonality is extremely suggestive that while the ultimate expression of the crank's favorite topic is as diverse as human knowledge, the underlying psychology that leads to break from reality could be identical.

Hypothetical underpinnings[edit]

Constructing world models[edit]

World models are internal cognitive models for how people believe the world works. These models are constructed through a combination of innate and instinctual base beliefs and incoming perceptual data. The task of an organism is to be able to make predictions based on incoming sensory data and adjusts actions choices to best maximize fitness oriented goals. Humans have the most advanced world modeling psychology yet discovered in animals. However, it is still based on innate beliefs and perceptual information.

These interact in interesting ways. For example, there is a fascinating sensory illusion called the cutaneous rabbit. A needle is used to lightly tap on the lower forearm of a subject, then two taps are quickly made in another location further up the forearm. The subject actually feels the second tap as if it occurred between the first and last. This stems from an instinctual belief about the maximum velocity at which something should be moving on ones arm, combined with the perceptual data available.

There is a strong drive to create world models for just about everything that shows a dynamic response in the world. People are evolved pattern recognition machines. But the cognitive processing behind the pattern recognition has had strong internal biases built in. The cost of miss-attributing intention and agency to noise is relatively low, wheres the cost of failing to attribute noise to an actual agency can be very high (is the rustling in the jungle a tiger or the wind? better to assume the tiger and run away). This innate bias leads to the construction of world models with a general over representation of intelligent and intentional agency being the source of perceived patterns.

Updating the models[edit]

The conceptualization of crank theories differ little in there structural creation from any of the more widely accepted modeled concepts that humans have developed. The cognitive and neurological mechanisms behind the construction of world models likely does not differ significantly between cranks and more normal people. People regularly construct models for how the world functions, many of these constructs are highly imaginable and often false. However, in normally functioning individuals models can be updated or abandoned based on new incoming data. To understand the mechanisms underlying crank psychology the answers are not likely to be found in trying to differentiate how they construct their world models, but rather on how these models are updated or abandoned based on incoming data. What separates cranks from merely wrong ideas is that no amount of incoming information that contradicts their model will cause that model to be abandoned or substantially changed.

Interesting areas to explore include: initial confidence in the accuracy of world models; perception, conceptualization and processing of evidence; filtration and integration of data into existing world models; adjustments of world models; perception of alternative world models.

Social elements[edit]

A final interesting piece are the social elements for how cranks interact with those around them. The "voice in the wilderness" mentality that they alone have the true answers to deep questions, and the vehement degree to which cranks seem to go to voice their theories, combined with the classic martyrdom and out right aggression towards consensus all seem to be apart of what makes a crank a crank. This maybe a trap though. There is a strong self-selection bias taking place with the more vocal cranks. There are probably many cranks who never attempt to push their theories on anyone else, and certainly not to the degree that their more vocal counterparts take part in. While analyzing the social elements of cranks could be interesting, it may not be a part of what it means to be a crank. Instead it maybe analyzing what it means to be vocal, extroverted and have a strong opinion. Are vocal proponents of non-crank ideas versus crank ideas substantially different in their method and approaches? Are vocal cranks, and solitary cranks more similar than vocal non-cranks and vocal cranks?

Co-variation[edit]

One of the most ubiquitous examples of standard model building going awry is in the co-variation task. The basics of the co-variation task is to try and model whether one phenomenon causes another when all you have is correlational observations. Essentially, when I see or do x then y happens, so x must cause y. In a formal sense this is always a fallacy (correlation does not equal causation) but in "everyday reasoning" it is often a good approximation. Many crank ideas are an over generalization of a co-variation of phenomenon. In medicine attributing a causal relationship between "getting over" an illness with "whatever I did to myself" is the primary mechanism that quack medicine propagates itself. In the realm of the paranormal attributing things like dust balls, temperature changes, or cold reading "hits" as being caused by some kind of supernatural force is another classic example.

Modeling co-variation learning[edit]

First lets break down how it might be possible to reason inside the co-variation task normally. The task is to determine if phenomenon x causes y. Phenomenon can be sensory data, or actions, essentially anything dynamic in the environment. There are then four categories of observational data that can be used to model this task. These are broken down in the co-variation matrix below:

Pulling this out of the abstract a bit lets break down the four categories A, B, C and D. We want to see if pushing the big red button is what turns on the television. Data in the A category is the number of times that pushing the red button leads to the television turning on, data in the B category is the number of times pushing the big red button does not lead to the television turning on, data in the C category is the number of times the television has turned on when the big red button was never pushed, data in the D category is any time the big red button was not pushed and the television did not turn on. All four of these categories should share equal weight in their "value" towards assessing the initial question of whether the big red button turns on the television.

Something interesting happens though when we start to take into account the base rate of each phenomenon. If the example is changed instead to trying to determine if smoking causes cancer we have some a priori data for the occurrence of cancer and smoking. The base rate of lung cancer is about 1/1600 people and the base rate of smoking is about 1/5. If we assume that smoking does not cause lung cancer we can break down the chances any given randomly selected person falling into one of the four categories:

How then can this information be used to update our assessment of the original hypothesis that smoking does not cause cancer. The hypothesis that smoking does not cause cancer can be represented as having a particular probability of being true given all of the data we have to this point, this can be written as . To do this we use Bayes' equation. Lets take one person, or one incoming piece of data, and see how the probability of our hypothesis changes based on which category it is in . Assume an initial probability for of 50 percent, and that we get one piece of data from category A change in probability to this data point is:

We can do something similar for each data set and we wind up with the following values:

What these values show give us is essentially the ranking for how valuable data is in each category for updating our model. When base rates are taken into account the informational value of data between the four categories is no longer equal, and the disparity between the categories increases as the number of total observations available before a decision has to be made decreases. Finding multiple people that both smoke and have cancer provides the most significant amount of information in making the claim the one causes the other.

Many experiments in psychology have found that people overvalue data in category A, this has largely been declared a reasoning bias and held as an example of illogical or irrational thinking in people. However, if events are assumed to be relatively rare than A does carry more informational value than any other category. People therefore appear to have an innate bias towards valuing co-occurrence, that maybe adaptive and reasonable in many "real world" conditions.

Footnotes[edit]

  1. The combining of the prefix "neuro" with a cute suffix is the really in thing to do these days. Most of these terms are a a label in search of research. The basic recipe goes: Find some interesting work in behavioral and cognitive psychology that has taken place in the last 30-40 years, group it all up into a category, apply a "neuro" label, write a meta-review on the the topic, call it the next big wave in neuroscience, and get a free publication in Nature out of the deal. In this spirit, I submit my own version. Nature here I come!