The Sept. 26 news article “A flood of AI features just arrived — whether we’re ready or not” indicated that technology companies are pushing out their artificial intelligence programs to get user data for improving their software. This sounded analogous to the “full self-driving” option that Tesla has been years late on fully delivering. The companies make it “clear their AI is a work in progress” and are taking care “to build in guardrails.” Well, Elon Musk made “clear” that Tesla’s full self-driving mode is a work in progress, but the “guardrails” don’t stop people from having accidents — some fatal.
The guardrails in AI don’t work — so far. If an AI program is “trained” by going on the internet and getting all the information available, true and false, how will the program know what information is bad? It will find a site that says ivermectin is a good treatment for covid-19. Will it indicate that ivermectin isn’t approved by the Food and Drug Administration for treatment of covid?
Gathering all information from the internet reminds me of an idiom: garbage in, garbage out. Where’s the guardrail for that?
Robert Passman, Millville, Del.
