Saturday, February 24, 2018

A NEW FACIAL RECOGNITION COMPANY IS TOUTING A 99 PERCENT ACCURACY RATE. I CALL THAT BULLSHIT

African-Americans recognized as gorillas by one facial recognition system

BY Trey Rusk

Running Code 3
February 21, 1018

The website Nextgov posted an article today that stated the 99 percent statistic.

I wouldn't call that bullshit if I didn't have some experience with camera reading. I ran a license plate reader that could alert to stolen, wanted and amber alerts based on cameras mounted on a vehicle reading the plates. False or bad reads were not uncommon. I will say the technology is improving but 99 percent accuracy? I don't think so.

On February 18, 2018, The New York Times ran an article about facial recognition technology. It admits that facial recognition is improving by leaps and bounds. The 99 percent accuracy rate was for white men.

It turns out that there is up to a 35 percent false read rate on darker skinned black women. The article goes on to read that real world biases seep into artificial intelligence.

In modern artificial intelligence, data rules. A.I. software is only as smart as the data used to train it. If there are many more white men than black women in the system, it will be worse at identifying the black women.

One widely used facial-recognition data set was estimated to be more than 75 percent male and more than 80 percent white, according to another research study.

in 2015, for example, Google had to apologize after it's image recognition systems photo app labeled African Americans as "gorillas."

People a lot more educated about artificial intelligence than me predict that the biases in the facial recognition/artificial intelligence data systems need to be addressed now because the industry is growing.

I am writing about facial recognition technology because I predict that it will be introduced into the criminal justice system. I am not a fan of what I call junk science that may be capable of convicting innocent people.

The reason polygraphs are not admissible in court is that they may be only as reliable as the operator. That means in my opinion that the operator's prejudices and biases may be built into the outcome of a polygraph test.

Let's remember that some junk science that was used to convict people has already been thrown out. I'm referring to Bite Mark Technology and now questions are arising about the effectiveness of hypnosis.

I'm also wary of repeated interrogations. Investigators are looking for inconsistent statements that may have been made years apart. They have the video of the previous interview to rely on. The suspect only has their memory coupled with the nervousness of being interrogated by law enforcement.

If you think I'm wrong about memory, just ask yourself, have you ever walked down the hall to retrieve something and when you arrived, forgot what it was you were looking for?

Now try to remember the item sitting in a chair across the table from two detectives referring to you as a person of interest.

I rest my case.

No comments: