Facial Recognition Glitches: Nothing New When Marketers and the Greedy Explains Technology
December 20, 2019
You can read “Federal Study Confirms Racial Bias of Many Facial Recognition Systems, Casts Doubt on Their Expanding Use” and get a semi coherent explanation about a nifty, much hyped technology.
A camera captures a picture. Software matches the image to a reference image. Software displays the identity of the person in the captured image.
Nothing could be easier, better, faster, and cheaper except when these systems return 25 to 60 percent incorrect matches. Close enough for horse shoes.
The write up states:
The National Institute of Standards and Technology, the federal laboratory known as NIST that develops standards for new technology, found “empirical evidence” that most of the facial-recognition algorithms exhibit “demographic differentials” that can worsen their accuracy based on a person’s age, gender or race. The study could fundamentally shake one of American law enforcement’s fastest-growing tools for identifying criminal suspects and witnesses, which privacy advocates have argued is ushering in a dangerous new wave of government surveillance tools.
I am not dragging my French bulldog into this pet store. I do want to point out a few things which are likely to make some people wish I would just go to the warehouse for the elderly and commence the dying thing:
1. None of the whiz bang technologies work in the real world. This means that enterprise search, content management, predictive analytics — don’t work like the marketing pitches say they do. The technologies work under quite specific conditions. When those conditions are not met, the systems go wonky. Clueless managers want to buy a silver bullet, preferably from someone with whom they can relate. When the tech nose dives into the ground, just call the lawyers and procure another system. There’s a reason liberal arts majors don’t take differential equations in college.
2. Engineering demonstrations take place in a hot house. You know. The kind of place that eccentrics use to raise orchids in Manhattan. Take the technology out of the hot house and let 23 year olds use the system, and the results are predictable. There are not enough dollars and people in the world to work through the data to figure out who is who and what is what. Why not guess? The results are likely to be more useful. Shocker. Come on. You know that random guesses can do better than a Bayesian based system which is not retrained on a continuing basis with carefully selected data.
3. Companies and stakeholders are so desperate for sales, opportunities to make presentations, and to convince people to give them money that the truth is squeezed from the engineers’ and developers’ actual statements. For example, the engineer says, “The training data must be updated every day, preferably in real time.” The marketer says, “Set it and forget it.” Yeah, right.
Net net: Facial recognition technology works under the right conditions. Unfortunately the right conditions are not the real world with people wearing sparkly sunglasses, a new hair style, a disguise, or a face that reflects one too many mojitos or a collision with a door.
Stephen E Arnold, December 20, 2019
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
