3 ms·
How to make a racist AI without really trying
- newman8r 9y agoI think it raises larger issues, because as far as NLP goes - yeah - making a classifier isn't hard (as the article states). But what happens when an algorithm that analyze data objectively, presents a result that would be deemed 'racist' if it were released by a human? yeah, this is useful in a corporate or marketing setting, but as we start to integrate more NLP tech into interpreting ML/statistical results, I don't think I'd want to inject bias and risk missing out on difficult yet important truths. Tay was just a marketing gimmick though - the biggest thing I took away form it was that a major corporation can blame AI for a faux pas and nobody will hate them for it. In 5 years, the "someone who hacked my twitter" excuse will be replaced by "oops, our AI made a boo boo"