## What is the difference between a stereotype bias and a heuristic?

A heuristic is accurate.

(Reposted because bad at blogging again. Need practice.)

Rough sketch of the method of prediction via combination of models, data, and heuristics

In statistical reasoning, heuristics (instincts, intuitions) are used to modify estimates from base rate cases. So if I’m trying to estimate the likelihood that podrag’s startup will succeed, I first begin with a base rate: how often do startups succeed? Let’s say 1/10. But I use a heuristic device: my intuition tells me that tech startups are more likely to succeed than the base rate. How much more? The answer is what is known as a “true” value vs. a “measured” value. Maybe I guess that they are twice as likely to succeed, then I say 2/10. This is not a “true” value but rather an “educated guess” using one or more heuristics and a base rate.

(Note that I am making use of empirically based rates, which themselves rest on heuristics, see: https://aeolipera.wordpress.com/2015/09/11/the-plural-of-anecdote-is-data/. I am trusting that the survey conductors, who were close to each anecdote, made reasonable choices about which anecdotes to include or throw out. Here’s a corollary: always distrust studies conducted by feminists.)

Maybe a study comes out that says the measured value for tech startups is 3/20. My guess has overshot the measured value. But the true value could also be closer to 1/10 or 2/10, we don’t know. We can reason from data, but we can’t know a “true” probablility- it is a category error.

Pay attention to this point, because it is similar to the idea of general intelligence. “General” intelligence is, by necessity, a true value. We can define it as something, say “the general ability to solve problems”, and we can take measurements of something which we think is strongly related to that, like the ability to solve the problems on a particular IQ test. But it is impossible to measure the ability to solve any possible problem.

From the more specific base rate I can use a similar heuristic to make a more specific guess: do I think podrag is more or less likely to succeed than the average tech startup? Maybe I think he is or is not, based on reasons I can give pro and con- maybe he has less than average entrepreneurial experience, or more than average understanding of economics, or less than average education, etc.

It is possible to empirically study the tendencies of these heuristics. One such finding is that people are very good at determining whether a familiar person is more or less smart than themselves, and this becomes more robust as IQ increases. For instance, we can say with great certainty that von Neumann was a man of extraordinary intelligence. Why? Because men whose intelligence reached the 160 ceiling for testability consistently considered him to be markedly superior. An example:

I have sometimes wondered whether a brain like von Neumann’s does not indicate a species superior to that of man.
Hans Bethe

Such anecdotes exist in abundance, and so we can reasonably conclude that von Neumann was significantly more intelligent than men of Bethe’s caliber.