3 ms·
What's the rate of false positives? Would like to see a peer-reviewed article of the most advanced recognition models. Simply put - no matter the accuracy, you'
by CDBackslash 4y ago
What's the rate of false positives? Would like to see a peer-reviewed article of the most advanced recognition models. Simply put - no matter the accuracy, you'll only have some type 2 errors, and hence a journalist will be able to write an anecdotal story about the injustice that a improperly identified person faced. All processes (criminal justice or otherwise) have error rates - yet not all anecdotal stories are covered. There is typically a straw-man argument that facial recognition is not accurate enough to be used - but I've never seen an argument why the models cannot be improved to the point where they are extremely accurate.
The result should not be to ban facial recognition, it should be to have an expeditious process to recognize and compensate people who have been identified in error.
- 1970-01-01 4y agoThe article is a terrible on critical details. No mention of hit confidence levels, software, hardware, or review process.
- humanistbot 4y ago> All processes (criminal justice or otherwise) have error rates - yet not all anecdotal stories are covered Normally this is a correct take on AI criticism; how many car crashes do the news report versus a not-so-self-driving car? But with policing, at least in the US, the past 10 years has been a continuous cycle of reporting and discussion about traditional analog false positive errors in policing, which often correlate with race. Over the past 10 years, I don't think there is another sector of decision-making that has been subjected to as much scrutiny as analog policing in the US.