,
6 tweets,
2 min read
Read on Twitter
I founded a research centre specialising in machine learning algorithms. I’m neither a Democrat nor a Republican. She’s completely right & I’m delighted a politician is actually talking about this as it’s going to become a huge issue.
(And this video - which was circulating last year - is a very visual illustration of how bias can be built in & automated in apparently ‘neutral’ machines).
How can maths be racist? Asks Ryan Saavedra. Automated bias is widely understood as a problem in machine learning, because algorithms make choices based on data - and what if the data input are biased? (See Cathy O’Neil, Sayifa Noble great work on this). 1/4
Case in point: Imagine a police force uses an algorithm to determine where to send officers, based on historic arrest data. For varied and complicated reasons, poor minority neighbourhoods have higher crime levels. Ergo, more officers are sent there. 2/4
Simply by virtue of physically being there, they’ll arrest more minority groups - which the algorithm interprets as proof the model is good. It starts to suggests more minority neighbourhoods are targeted. Result? Apparently ‘neutral’ algorithm = discriminatory policy. 3/4
Also see my colleague @akrasodomski’s great essay in @HuffPostUKTech about this - and the perils of machine reliance. m.huffingtonpost.co.uk/amp/entry/alex…
