BiasInAlgorithms

ThoughtStorms Wiki

See also RacistAI, RacistComputers, AIFailures

Go beyond TheSocialDilemma to people who have been investigating Algorirhmic abuse and inequality for a while :

Via https://threadreaderapp.com/thread/1308542908291661824.html

https://blog.getpocket.com/2020/06/the-bias-embedded-in-algorithms/

The Problem

https://www.theguardian.com/technology/2021/aug/10/twitters-image-cropping-algorithm-prefers-younger-slimmer-faces-with-lighter-skin-analysis

ThreadView

Doing something about it

Removing bias in Large LanguageModels https://www.alignmentforum.org/posts/fFF3G4W8FbXigS4gr/cognitive-biases-in-large-language-models

Quora Answer : Is AI just as biased as the humans creating it?

Mar 11, 2020

It's as biased as the data it's trained on.

That's subtly different from "as biased as the humans creating it".

Yes, the humans might be feeding the AI training data that fits their own biases.

Or they might have a method of gathering data (eg. scraping the internet) which gets data that isn't strictly speaking their biases, but which has its own biases.

When there was a scandal a few years ago that Google's image recognition was racist it was clear that the racism didn't necessarily come from Google's own engineers. And probably didn't reflect their prejudices.

Nevertheless, Google's engineers allowed this data-set through perhaps without thinking that it would contain the bias it had. Or more likely, without realizing that black people were under-represented in the set and so more likely to be subject to false negatives.

So AIs are getting trained up full of human biases. And yes, this is a real concern. But it's not necessarily as simplistic as the exact biases of the engineers getting transferred across.

Backlinks (3 items)