4 ms·
AIs are like every other computer program - perfectly naive. If you train them on bad data, you get bad results. Garbage in, garbage out.
by vorpalhex 7y ago
AIs are like every other computer program - perfectly naive. If you train them on bad data, you get bad results. Garbage in, garbage out.
- core-questions 7y agoWhat happens when we train them on an incredibly wide set of randomly-sampled data, which should show no built-in bias, and uncomfortable conclusions emerge? Do we censor the AI, castigate someone else in the chain for the problem, or just deal with reality as it presents itself?
- megablast 7y agoIn this example, where would you get the tags for the photos from, if not people who are biased?
- core-questions 7y agoWhy assume that the people are "biased"? If the people are labelling things the way that people generally do, the software will return labels that are useful. Any kind of filtering needed to censor this for people's personal comfort should be implemented on top.
- megablast 7y agoOf course, but the issue is also data availability. It is not easy to get large sets of data to be used in training AI. Imagenet was a great attempt to provide some useful data.
- vorpalhex 7y agoMy comment was not intended to be a ding on Imagenet. Data comes from somewhere, and no source is without bias because no human is without bias.
- falsedan 7y agoyes, when I’m preparing my prestigious project with my Ivy League peers I make sure not to sanitize or vet my inputs so as not to bias the results of my programs or publication success. truly it is remarkable that the training set we chose to use is racist but no single person who could have stopped this or be held responsible is
- vorpalhex 7y agoIf your data set was compiled by a single person, well, easy enough. If your data set was compiled by thousands of mechanical turk workers, well, you got a lot of people to blame a little bit. As it goes, "everybody is a little racist sometimes..." and apparently that shows on a big enough data set.
- falsedan 7y agoamazing, how when even more people are involved with selecting and tagging the input, it becomes even less likely any one person was to blame.