Comment by mjburgess | Hacker News Reader

Comment by mjburgess | Hacker News Reader

Parent Full thread

You should take seriously that a vast amount of research of this kind is pseudoscience, and that there isn't much science that can be done from it.

All "heritability" research is just an analysis of correlations between genes and expressed behaviour. On these grounds a scottish accent is highly heritable, since the accent correlates extremely highly with some genes.

We do not have the means to create new human beings with different genetic profiles and observe them accross many environments, as they develop, to determine if "somehow" a gene causes anything so macroscopic like "intelligence".

So, quite literally: we cannot do science. We cannot create experiments to test hypotheses.

Genes have no plausible mechanism to "cause" intelligence. If they play a significant role, it is a very large number of genes building tissues which are completely opaque (brains, etc.) whose interactions with the environment are completely opaque.

Heritability research, and "psych-genetic research" of this kind is, necessarily, pseudoscience. It's adhearents are the poorest statistians and have the least understanding of the scientific method you could hope for.

We have no scientific theory of intelligence which provides causal variables that we can experiment with; we have no theory of biological mechanism, likewise; we have no theory of genetic mechanism --- all we have are some aging loons with stats software content to put iq-test-scores (ie., people competing patterns of shapes...) and Genes into it.

Whatever can be found by these loons isn't worth a scientist finding. "Data" isnt innocent, it is collected and used for a reason; and pseudoscience isnt innocent.

Correlating genes with ESP ability is likewise nonesense, and that has as much scientific merit.

I think this post reveals a lot of ignorance in how scientists study heritability.

To give a simple example, one of the best tools is comparing identical twins to non-identical ones. If the correlation between the fraternal twins is substantially less than the correlation between the identical twins then we can know a genetic factor us in play.

Your argument is also too general in that it labels all observational science as pseudoscience. Is astronomy pseudoscience? You don't need to be able to manipulate the environment to do science, just have a hypothesis and a way I'd getting new observations that would probe/disprove the hypothesis.

I'm very well aware of how "heritability" is studied. It is defined to be a correlation, so twin studies do not estimate heritability. They attempt to estimate how much heritability is explained by gene expression being the cause of the relevant behaviour.

Twin studies assume a model of environmental variation which cannot be evidenced, because the relevant target of study (a human being in its environment) cannot be experimented on to evidence anything. Twin studies are not RCTs (which are themselves the minimum an experiment needs to be to count as science).

Indeed, the assumed model of the interaction between genes and the environment is *plainly false* (namely, that "equality" between environments exists and that it has linear non-feedback, non-cyclical effects).

Twin studies on star signs would likewise produce as compelling evidence that star signs cause personality. (Can you see why that would be?)

Science requires such experiments take place. No amount of post-hoc analysis of correlation can produce a causal explanation. One must control causes to do so.

Hence, much of this abuse of statistics is unreporducible pseudoscience; it is well known to be outside these fields by anyone with any credible understanding of statistics.

Every chemist will defend physics; and every chemist, biology. And every biologist neuroscience.

Can you cite me anythihg other than the abject horror of real scientists outside these fields on reading their reports?

I'm sorry to break the news: this is peudoscience in the tradiiton of pseudoscience which was invented by psychology's repurposing of statistics for post-hoc analysis of correlation on human data (ie., the most complex systems known to exist) which fails even basic tests of scientific credibility.

Complex systems resist study by the use of the scientific method, and require the best minds in modelling and statistics to even produce "rule of thumb" models (eg., consider the climate).

What we have in heritability, psychometric, etc. research is loons studying the climate (ie., analogously, humans) by putting thermometer readings into GUI stats software and publishing the results.

They're fools. Neo-homeopaths, whose lab coat is "statistics"/.

That post cements that you do not understand the fundamentals of the scientific method or how we do scientific analysis. E.g. that we assume "that "equality" between environments exists"

You're very confused about how a lot of things work on fairly fundamental levels which would take hours to disentangle, so all I can do is recommend learning more on philosophy of science and read some more papers on heritability. Maybe this one picked not entirely at random: https://www.researchgate.net/profile/James-Holland-2/publica...

Sure, and likewise: https://link.springer.com/article/10.1023/A:1012896802713 .

I am presently researching a PhD in the philosophy of science and causal analysis; I have two degrees in physics, and I have spend the last nearly 10 years with major lead researchers and data scientists across many industries.

My opinions here are a summary of a talk I recently delivered at a conference on issues applying "ordinary statistical methods" to complex systems, esp. those with profound hysteresis, such as people. But likewise, my opinions arent rare amongst people with similar experience to me: psychmetrics, psychology, heritability research, and so on -- are widely regarded amongst my peers as pseudoscience.

It's unlikely I've made a basic mistake.

Very refreshing to see someone actually providing a link for further reading. This raises the level of discourse on HN.

His argument as I understand it, is that the MZ twins have a different environment in ways such as being treated as more similar by their parents. This means that high similarity in one respect (such as appearance or temperament) can contaminate the natural experiment by making people assume they are more similar in another respect (such as intelligence or learning style) and treating them the same where that second trait is relevant, thus making the second trait more highly correlated than they would be by genes alone (e.g. you assume they're equally intelligent and thus give them the same education).

This has some merit but is quite far from a knock-down argument.

My previous talks with a professor studying genetics in Oxford has given me the impression that heritability is settled science and theres a huge volume of papers in the area, so I'm not going to change my mind based on an anonymous comment and a paper by a non-academic with an axe to grind without first going back and consulting with her.

Given your statements here I am quite confused as to how you can have a degree in physics while dismissing science where you can't make interventions... did you study astronomy at all in your degree?

I'm also confused as to how you state that genes do not cause intelligence, do you think a mouse is less intelligent than a human due to its lack of soul or something?

[1] https://geneillusions.files.wordpress.com/2021/06/twin_2002_...

I dont have that high an opinion of astronomy, but my issue is what counts as an explanation (a model with causal variables), and what counts as an experiment (setting the value, etc.) -- astronomy can often pass these tests on the underlying models.. and are usually good at reporting the confidence intervals as a appropriate.

My issue with something like a human being is that the state space is going to be non-linear, in the sense that suppose we have an explanatory model of the human in their environment, state = h(x_env, x_body, t, ...)

This makes controlling, really in any sense, those variables is going to be nearly impossible. So close values in `x` space, eg., two environments that look nearly-identical, are going to have radically different values in state space.

Psychologists, using linear "mereological intuitions" based on classical physics, will deny this. A physicist however, would not make this mistake. In cases of hysteresis and complexity, we're well-aware tiny variations in an environment can produce radically different paths thru' state space.

(And I think people are more complex than magnets!).

The fundamental issue with the field of studying human behaviour with scientific methods, is that (imv), we dont have scientific methods. We shouldn't expect science to provide us a method which works in every env.

I have, perhaps, high standards on what counts as science; it's esteem imv comes from disciplines that can meet those standards. Everyone else is rent-seeking.

----

Edit:

> genes do not cause intelligence

If there's an infinite number of possible people P, our genes make a subset of those impossible because they're not "biologically possible".

There is a narrow sense of "cause" which therefore says that genes play a causal role.

I was taking a more robust sense, in which (one plausible view), is that the set of effects we call "intelligence" are gounded in the body, but a cause of the environment.

Splitting these issues out is very subtle however.

One simple way of making this clear: most possible bodies arent intelligent; they dont speak; they dont think; etc. And all those are biologically possible.

Just as genes dont cause English speaking, they dont "obviously" cause intelligence. Most biologically possible people are basically unable to think, never having been developed in a social environment.

----

Very cool! Reminds me of the Thompson compiler hack.

Not only they tell us genetics doesn't exist, but we can't say it's shitty science because they're the ones telling us what's shitty science and what's not! Brilliant!

Would be nice if you could supply a none-paywalled link to download the research paper. Thanks!

Reply on news.ycombinator.com