by Chris Black
Key points:
– the larger the datasets, the less liberal the resulting AI
– open-source AI networks (companies that provide large-scale datasets) are planning to scrub ideologically problematic data points from their sets
– researchers suggest that closed AI like ChatGPT, Google, and Microsoft may be even more biased (obviously it is, but not in the way they mean)
To reduce bias in AI, many tech companies take the tack of using larger datasets for training, but an analysis shows that this approach actually increases results that include racism or hateful content. https://t.co/yIsY39rF2O
— New Scientist (@newscientist) July 25, 2023