Dana Fried
AI is biased because reality is biased.
ML gives you a picture of the world as it is - or rather the subset of the world in its training data, the selection and labeling of which also introduces bias.
All ML can ever do, therefore, is reproduce existing biases; it does this automatically and by default.
ML can never be made unbiased because you would have to account for every structural inequality specifically. At that point you might as well just have a person do it.
Dana Fried
That isn't to say that ML is bad, per se. But it's irresponsible to use an ML model in any capacity where its inevitable bias could negatively affect people - especially and specifically people who are not its owners.
Vrimj
I replace ML or AI or LMM with "trained weasels" when I am doing legal risk assessments. It makes me ask better questions about who is training the weasels and how and do we really trust weasels to do that.
jz.tusk
I wonder if there is a simple way to explain how ML training works to non-techy folk, so that they realize it's mostly just tweaking equations until they reproduce the corpus they are fed, and they're going to do their best to reproduce that corpus, biases and defects included?
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.