Preventing Harmful AI Bias with Fairness Through Awareness

forbes.com

2 points·fairytalemtg··2 comments

The absolute train wreck coming to the technology industry through unbiased math is going to be both hilarious and sad to see in the coming years.

The harsh reality is that equality-of-outcome between different groups is impossible (short of a Harrison Bergeron type of scenario) and an unbiased algorithm uses pattern recognition to see that.

If you use ML to do anything like decide loans, educational selection, jobs, anything to do with crime, etc, you're going to have one group or another do better than one group or the other.

ML will either have to produce politically incorrect results, will have to be intentionally weighted to favor certain groups, or will have to be banned.

See this for example. Even Amazon couldn't get the desired outcome.

https://www.reuters.com/article/us-amazon-com-jobs-automatio...

So many startups with such low sample size pitch big on AI/ML/NLP and every things else under the sun. AMZN and other bigs not able to win it, says loud that it is complex, and very algorithm can always be introduced to weight a bias.

AI telling me it is time to eat/drink/workout is very different from AI evaluating my social media feeds and recommending topic/time to post, to get me more reactions. Whether what I share has value or not.

AI/NLP to transform mundane tasks into automation is always a goal to achieve :)

Reply on news.ycombinator.com