OMG YES THIS.

I don’t know if you’re familiar with Joanna Russ’s book To Write Like a Woman, but there were bits and pieces of it flashing through my brain as I read all this.

From this point of view, Ogas and Gaddam are trying to build a model of sexual cognition with similar properties. The “circuit” will acquire sexualities that are similar in certain respects but different in ways that children learning English and Japanese are different, because they were exposed to English and Japanese. Again, the flaw is that linguistic cognitive modelers (ie, there are definitely people doing for language what Ogas and Gaddam are trying to do for sexuality) are never really nearly as ambitious as they are, because we don’t really even understand what a neural circuit is…

The outcome of this effort is intended to be the validation of a particular machine learning algorithm that reflects their theory about the brain, not conclusions about the communities they are studying. Again, whether this is a valid effort without a better conceptualization of their theory is another question.

Of course, the flaw lies in their notion of functional equivalence which has been teased out at length in those threads (the “transsexuals vs slash” business). But they probably think that they can drown out those issues via the volume of data.

If I were them, I would just have downloaded and used the text of slash fic directly—but I can also see reasons why they wouldn’t want to do that only.

Obviously, both my ideas and neededalj’s are speculations, but everything they said fits pretty closely with this conceptualization. And I don’t think that any modeling venture can really take place with a bad underlying theory, which they have. So as I said before, I agree with critiques of their behaviour, which reflects an ignorance that is the only thing that can produce many aspects of their theory, but some of the “deeper” technical criticisms seem to be off base.

They alluded repeatedly to plans to elicit the same data from a community they believe (and are probably correct) to be very different from the fanfic community.

They can’t elicit the same data from another community, because many of their questions were specific to fanfic. Of what other community could they ask, “At what age did you read your first slash story?” or “Which specific types of fan fiction do you actively seek out to read?”

They will probably construct and train two models based on the two sets of elicited data, and their hypothesis is that the models will “react” in a certain way to the stimuli they plan to subject the models to. They changed questions simply as the need for training data changed—it was not intended for conventional survey purposes, so the “scientificness” of this doesn’t matter to them (they think). Fake answers provide data too, as long as they were “genuine” fake answers that the other population they plan to use for data provides different data. In essence, it really was a “they meant to do that” research troll from the outset.

I part from neededalj in that I don’t view it with as much incredulity as s/he does. Oh, I don’t doubt that the concept is flawed and premature, but the idea of a cognitive modeling project via machine learning over elicited text does not yield a lot of surprise from my corner at least. As I suggested above, one thing I found very interesting was that the incredulity came from social scientists and neuroscientists and like-minded people.

I don’t doubt the concept is deeply flawed, as I said, but it’s neither as crazy nor as stupid as some of their critics are making it to be. This doesn’t excuse their behaviour either.

Oh, and in answer to one of their survey “questions” – my sexual orientation is either horizontal, or west. Depending on whether they’re measuring an absolute or a relative orientation.