> If you've heard 5-10 plausible meanings for "intelligence", and every one of them is heritable, then you can be pretty sure that "intelligence" is heritable even if you're not sure which particular meaning someone is using.

What if you’ve heard definitions for intelligence that aren’t measured by IQ test scores and then someone uses IQ test scores to argue for the heritability of intelligence?

> And the article says: "on average, psychological traits are extremely heritable, at around 50 per cent." It's hard to argue for a definition of "intelligence" that wouldn't be a psychological trait.

Its equally hard to argue for an average of things which are not faithfully represented by numbers.

I'd be skeptical of a definition of intelligence that had no correlation with IQ scores. If you managed to put forth and defend such a definition, then that could turn into a valid objection; but that would then run into the "psychological trait heritability" prior.

> Its equally hard to argue for an average of things which are not faithfully represented by numbers.

How do you represent, say, extroversion with numbers? It's messy and any particular measure is probably not a good one, but what people do is come up with a list of questions like "Statement: I am excited when I go to a social event full of strangers; rate your agreement from 0 (Strongly Disagree) to 4 (Strongly Agree)", and add up the numbers to create a score. It's kind of arbitrary and has a lot of "noise" in it, but on the other hand there is enough signal that it is measuring something real. If you need to turn it into a binary classification, you choose a cutoff number, which will again be semi-arbitrary and make errors, but again can be good enough to draw useful conclusions. You can then apply that measure to, say, biological siblings who were both adopted, or (best) to identical twins separated at birth vs fraternal twins separated at birth, and see if genetic similarity correlates with similarity of your measure. From that kind of data, you can get a vague idea of heritability.

If the thing you're discussing isn't measured by existing tests, fine. If the thing you're discussing has a logical definition and is important, but you think there can be no tests to produce even a vague measure, then I would guess you just haven't tried very hard.

> I'd be skeptical of a definition of intelligence that had no correlation with IQ scores. If you managed to put forth and defend such a definition, then that could turn into a valid objection; but that would then run into the "psychological trait heritability" prior.

Its hard to measure a correlation between a set of numbers and a set of NaN’s. However the prior here is that 50% of psychological traits are not heritable, so our prior is that we have approximately equal chances of a trait being heritable or not. What is likely is that some aspects of intelligence are heritable and some are not, and if the statistics that indicate heritability are measuring anything, they are measuring that part of intelligence that is measured by tests AND is heritable. Which is obviously poor evidence for those being the only relevant aspects of intelligence.

> It's kind of arbitrary and has a lot of "noise" in it, but on the other hand there is enough signal that it is measuring something real.

The “something real” is “how people respond to tests”, the assertion that “how people respond to tests” is positively correlated with “extraversion” is a claim that needs evidence. Mere assertion is not sufficient.

> If you need to turn it into a binary classification, you choose a cutoff number, which will again be semi-arbitrary and make errors, but again can be good enough to draw useful conclusions.

The results of a research program are more often useful to the person who is credited with conducting the research than they are useful to the person who is concerned with making accurate predictions of the real world phenomena allegedly represented by tests.

> You can then apply that measure to, say, biological siblings who were both adopted, or (best) to identical twins separated at birth vs fraternal twins separated at birth, and see if genetic similarity correlates with similarity of your measure. From that kind of data, you can get a vague idea of heritability.

Yes, you can get a good idea of the heritability of how a person responds to tests. However the program you described does nothing to validate the responses on tests to any real-world phenomena.

> If the thing you're discussing isn't measured by existing tests, fine. If the thing you're discussing has a logical definition and is important, but you think there can be no tests to produce even a vague measure, then I would guess you just haven't tried very hard.

Perhaps one has tried hard enough to realize that the problem is less tractable than you suppose. One could also observe that “intelligence” is socially constructed and does not in fact have a logical definition that is based solely on objective factors.