An AI cyber-cop crackdown on chess conversations for racism tells us what we should have known all along: that algorithms created by fallible humans cannot be infallible
Michael Jackson was wrong, it turns out. It does matter if you're "Black or White". Algorithms that constitute the top oxymoron of our times, Artificial Intelligence (AI), have affirmed as much in their latest avataar as cyber-cops. Last year, a popular YouTube chess channel called Agadmator got suspended for flouting the site's guidelines. The reason for this, it has just emerged, was the 'racism' detected by AI cyber-cops in the use of words like "black" and "white" in conjunction with terms of aggression like "attack". Cyber-policing is done sneakily, and offenders are not notified of their violations, so it took an Indian avid chess player and scientist, Ashique KhudaBakhsh, 38, six weeks of experimentation to identify what led to the crackdown.
Chess players, of course, are known for an ability to escape traps, but that is usually if they have a human mind at the other end. On the internet, short of playing with chess pieces acceptable to rule-enforcers (pink and grey, anyone?), they may find the going hard.
Defenders of AI will offer the usual defence: that AI is imperfect right now, but is evolving so fast--a language tool called GPT-3 has started writing what many geeks consider poetry--that its 'intelligence' will start showing up soon. This optimism is drawn from another prized contradiction in terms called Machine Learning. All this is part of a technological revolution to minimize the role of human judgement, which is seen as fallible at the best of times. Human folly, we are all acquainted with. What few care to explain is how algorithms designed, fed and taught by fallible beings can be infallible.
Courtesy - Livemint.
0 comments:
Post a Comment