“You can’t see me” is the famous catchphrase and hand gesture of WWE legend John Cena, where he waves his hand in front of his face
John started it as a joke dance with his younger brother who did to a hip-hop beat, meaning his rivals were not on his level. It turned into a huge internet meme that he is invisible.
If you’re a WWE fan like me, you’re probably humming the tune to the song right now.
I thought of John Cena recently, not because of his wrestling accreditations, but more so of him being invisible and his opponents not being able to see him.
The BBC recently reported that a Sainsbury’s store in London, which is using AI to identify unwanted customers has recently made some mistakes! The software identified one gentleman as someone who should be escorted of the premises. So as he was going about doing his normal shopping, he went to the self-checkout and then two store managers said to him that he’s got to leave.
Shocked is an understatement.
He left, raised it up with Sainsburys, who later said they had made a mistake. Well, it was them and their AI.
The company who has provided the AI face recognition said that the software does face recognition of people who have been shop lifting/stealing and/or caused trouble in a store. Now mistakes can happen. And then last step in their process is to have a human verify.
Let’s be honest, we humans at best, are no where near as good as machines.
So a mistake was made. Actually, this wasn’t the first time. As the story broke out, a lot more people came forward saying they had experienced something similar.
Sainsbury’s are not halting the roll out of their AI facial reconigiton system.
This got me thinking, a) what is the accuracy of the AI system (can they really see the face clearly or is this a John Cena moment? and b) human intervention – if this is meant to be the final sanity check before an action takes place, then why isn’t it working?
The process itself sounds logical, but the steps within don’t fill Sainbury’s with confidence.
So do we blame AI?
No. AI is just doing what it is being told. It has been given tasks, parameters, data, information, etc… and asked to go ahead and generate an output.
Is it the human?
Maybe. Should the human double check, verify and what capabilities does the human have to make the right decision?
Who do we blame?
The process.
The process should have ‘quality’ checks. If AI identifies a person, provides a high level percentage rate that it is them (against stored photos, data, etc…), then the human intervention should be to go to the person and talk to them. Could the human wear a camera that has a close up of the person which feeds back to the AI system to verify their original decision?
It comes down to confidence of the system.
I think it’s a matter of time before you will not need human intervention. AI will be able to do an initial check, close up check and continue to verify, up to the point where a decision is made with 99% confidence and then for a human to talk to the individual.
There’s more and more camera’s in more and more public spaces. Facial recognition is happening everywhere. Your face is on a computer somewhere with some tags of data.
What that data is, only you and the computer know.
Oh, and maybe AI will be using it for some purpose.
Wonder what AI would make of John Cena….