Should we really trust AI for medical advice and air traffic control?
This item is available in full to subscribers.
Please log in to continue |
Register to post eventsIf you'd like to post an event to our calendar, you can create a free account by clicking here. Note that free accounts do not have access to our subscriber-only content. |
Are you a day pass subscriber who needs to log in? Click here to continue.
To the editor:
I recently had an incidental demonstration of how extremely wrong AI answers can be.
I texted my niece a “Find the Differences” puzzle based on a cartoon image from a television series. She responded immediately with a list. I asked where she got it. She replied “I used AI mode.” Not a single item on the list was precisely accurate.
The first six on a list of 15 were just wrong, no differences whatsoever. Some very obvious differences were missed, such as a bunny figurine’s ear from erect to floppy and different style faucets on the sink.
Some were maybe close possibilities such as “Easter basket.” There were no changes to the basket but one of the eggs it contained was a different color. Another egg on the counter close to the basket also changed color. AI identified the egg as “soap.” AI identified a throw rug as “floor.”
If there were several changes on the same item, AI lumped them together as one difference. The bath tub had a stain on the top rim and a change in the design of the claw foot. AI counted them as one, “Bathtub leg & stain.” It lumped four changes to the woman’s face as one, her “expression.”
Should we really trust AI for medical advice and air traffic control?
Saundra McMillan
Riverside