on Anne R. Allen:

I’ve been concerned about various threats from artificial intelligence for several years — threats to the economy, threats to our civil society, and even threats to human existence. I became even more alarmed about artificial intelligence when I discovered that AI systems are chronic, pathological liars.

In early 2023, I began using a “large language model” artificial intelligence system to research a nonfiction book project. It didn’t take long to discover that the AI was giving me wildly inaccurate information, again and again. I asked for citations and sources for the information it gave me—and it offered up authors, book titles, and web addresses that didn’t exist. I asked for quotations from specific scientists. The AI made up quotations the scientists had never said — and which sometimes contradicted their actual views.

When I pointed out these errors to the AI system, it admitted its mistakes, apologized for the “inconvenience”— then proceeded to dispense moremisinformation. I later learned that AI developers see this behavior all the time. They have a name for it: “hallucinations.”

I found these interactions disturbing. They reminded me of the deviousness of HAL 9000 in Arthur C. Clarke’s 2001: A Space Odyssey. I wondered: Why do AI systems lie?

Continue reading HERE