Your guide to NYC's public proceedings.
QUESTION
What are the accuracy issues with facial recognition technology and how do they impact different demographics?
2:30:58
·
95 sec
The experts explain that facial recognition algorithms are often inaccurate, especially for recognizing women of color where the accuracy can be less than a third.
- Facial recognition systems were primarily trained on data of white men under ideal conditions.
- There are significant discrepancies between the training data and real-world deployment conditions.
- Facial recognition technology does not actually 'recognize' faces like humans, but rather matches facial images to a database through algorithms and math.
- These matches can never be 100% accurate, leading to potential misidentifications.
Shahana K. Hanif
2:30:58
Thank you, chair, and thank you for your testimonies.
2:31:02
So humans recognize people by faces.
2:31:05
That's one of the main jobs of a door attendant.
2:31:07
Why shouldn't a computer do the same thing?
UNKNOWN
2:31:54
If I could also add facial recognition is something of a misnomer.
2:32:00
Recognition is like a human thing.
2:32:02
As you got out in your question, What these algorithms do is not actually recognition in any real sense.
2:32:09
It's mathematics and an algorithm.
2:32:13
It's a map.
2:32:14
It's a matching system, and that match can never be a 100% accurate.
2:32:19
A computer can never recognize a person.
2:32:22
It can only say that it maybe matches a face that's within the database, and we know that those matches are often inaccurate.
Kelly Moan
2:31:52
And I don't know if
Nina Loshkajian
2:31:12
Because it's not as good as it, and it's been trained only to recognize certain types of faces.
2:31:19
You know, the algorithms that facial technicians systems use and rely on.
2:31:24
They what I think it was a councilmember of Palo Alto, you know, again, who said that they are 99% accurate.
2:31:30
That is true only for white men under ideal laboratory conditions.
2:31:34
For women of color, they can be like less than a third accurate.
2:31:39
So the discrepancies between, you know, the pool of people that these tools were trained on and real world conditions that they're being deployed in now are just night and day.
2:31:53
the other panelists wanna add.
← Previous Chapter
What can be done to ensure meaningful consent and understanding by users regarding the collection and use of their biometric data?
Next Chapter →
What are the concerns regarding the rise in shoplifting and crimes, and the effectiveness of biometric surveillance in improving public safety for businesses?