Computers and Prejudice

April 6th, 2010
ism, nlp
Consider the following fake snippet of news reporting:

Sarah Jones reporting for News Station X. The CEO of Big Powerful Company announced today that BPC was on track to meet its earnings targets for the quarter. She went on to attribute the continuing success of BPC to the quality of its employees.

Imagine I were to tell you that I interpreted the "she" as being a reference back to "sarah jones" instead of "the ceo of big powerful company". What if I say that this is because I think of "ceo" as a male-gendered word, so I interpret the "she" as being more likely to refer to "sarah jones". You'd say I was being sexist, no? The assumption that a ceo is male is a sexist assumption. I'd agree with you.

Now imagine I am working on a computer program that is supposed to learn how to connect pronouns (and other words) back to their antecedents. (I am, in fact, working on such a program) Imagine it learns, from lots and lots of data, that ceos are much more likely to be male than female. So it makes the mistake I described above, identifying "she" as the reporter instead of the ceo. Is the program being sexist? Were the programmers sexist in desigining the program with the capacity to learn the gender of words? Is there sexism here, aside from that which results in there being more male ceos than female?

Comment via: facebook, substack

Recent posts on blogs I like:

The benchmarkpocalypse

There's been a lot of talk about the vulnpocalypse, to which I don't have much to add because I'm not a security person, but I haven't seen much discussion on the closely related (and to be fair, less serious, issue), the benchmarkpocalyps…

via Posts on August 17, 2026

You can't join the secret inner circle of cool effective altruists

I, and a lot of other people I know, seem to believe something like this:

via Thing of Things August 11, 2026

Microfictions 2

A few more microfictions, still no LLMs.

via Evan Fields August 7, 2026

more     (via openring)