Hitoshi Kokumai

9ヶ月前 · 1 分の読書時間 · visibility 0 ·

chat 著者への問い合わせ

thumb_up 関連性 message コメント

Arguments over ‘Clearview.ai’- Scaring or Laughable?

Many people have long been arguing over the scare of Clearview.ai. 


A simple question has occurred to me; I wonder if these arguments are based on the knowledge of the empirical data of False Acceptance (false match/false positive) rate and the corresponding False Rejection (false non-match/false negative) rate of Clearview..ai face matching algorithm.

Has someone been able to confirm the true empirical data?

Unless the true False Acceptance rate is just 0% when the corresponding False Rejection is close to 0%, you could well be identified as some other people without your and their knowledge, and some other people could well be identified as you without their and your knowledge.

Should the figure be far above 0%, ‘some other people’ needs to be altered to ‘many other people

Are you more scared?

Or, do you burst into laughter?

Arguments over ‘Clearview.ai’- Scaring or Laughable?Distracted “A

‘AI-Enhanced Biometrics’ sounds as if it were very powerful, dreadful and, therefore, most threatening to citizens. 

However, the effect of ‘AI-enhanced biometrics’ is more psychological than substantial, say, more like a ghostly paper tiger coming with frightening sound effects, 

 If someone tells you “Our AI knows you were there”, your reply could be “Have you considered the empirical probability of False Acceptance of me and False Rejection of your true target?” 

 How come?

 Reducing errors caused by imperfect algorithm of pattern-matching is one thing. Reducing the errors coming from the inherently probabilistic nature of the objects of the pattern-matching is another. Human’s body features are unpredictably variable; predicting something is not possible where predictability does not exist. AI does not change this fact.

thumb_up 関連性 message コメント

その他の記事 Hitoshi Kokumai