66
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 20 Feb 2024
66 points (100.0% liked)
World News
22009 readers
66 users here now
Breaking news from around the world.
News that is American but has an international facet may also be posted here.
Guidelines for submissions:
- Where possible, post the original source of information.
- If there is a paywall, you can use alternative sources or provide an archive.today, 12ft.io, etc. link in the body.
- Do not editorialize titles. Preserve the original title when possible; edits for clarity are fine.
- Do not post ragebait or shock stories. These will be removed.
- Do not post tabloid or blogspam stories. These will be removed.
- Social media should be a source of last resort.
These guidelines will be enforced on a know-it-when-I-see-it basis.
For US News, see the US News community.
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 2 years ago
MODERATORS
Building off your last point, with AI models, bias can come in ways you might not expect. For example, I once saw a model that was trained with diversity in mind, but then only ever output Asian people with a high bias towards women. It seems to me like diversity is something that is difficult to train into a model since it'd be really difficult not to overfit it on a specific demographic.
It might be interesting to see if a random input into the model could be used to increase the diversity of the model outputs. This doesn't really help with resume screening tools though (which are probably classifiers), only really generative models.
There isn't really a good way to even define for diversity.
The bad approach is the corporate token diversity, where every picture has to include a white, a black and an asian person, at least 50% have to be women and one of them has to wear a hijab. That might include many groups, but isn't really representative.
You could also use the "blind test" approach many tech solutions are using, where you simply leave out any hints to cultural background, but as has been shown, if the underlying data is biased, AIs will find that (for example by devaluing certain zip codes).
And of course there's the "equal opportunity" approach, where you try to represent the relevant groups in your selection like they are in the underlying population, but that is essentially *-ism by another name.