Data Digest: AI Bias, Gullibility, and Positive Uses
A new algorithm shows biased training data, why humans can fool machine learning, and AI for social good.
- By Upside Staff
- June 30, 2020
An algorithm that tries to extrapolate faces from pixelated images defaults to white features, indicating flaws in the training data and possibly deeper bias issues.
Read more at The Verge
New research suggests that today's machine learning algorithms cannot learn to consistently handle input intended to deceive them.
Read more at SciTech Daily
Here are 10 examples of how AI can be applied for social good.
Read more at Forbes