AI Can’t “Lie”

Ars Technica published an article by Ryan Whitwam titled Testing suggests Google’s AI Overviews tell millions of lies per hour. The article itself is about a New York Times investigation into the accuracy of Google’s AI overviews. The framing in the headline matches the first paragraph: “…1 in 10 AI answers is wrong, and for Google, that means hundreds of thousands of lies going out every minute of the day” (emphasis added). While I am not a fan of “AI” replacing traditional search and have even published information about creating custom search engine shortcuts for DuckDuckGo’s No AI mode and Google Search, I am also not a fan of Mr. Whitwam‘s framing. Google’s generative AI implementation does not tell lies. It has no concept of truth or falsity. It returns purportedly responsive text to specific queries in accordance with its programming and index. It no more “lies” when the text is inaccurate than it “tells the truth” when the text is accurate. Assigning agency to agents confuses the issues. (I will add as a separate issue that for those of us who support “traditional” search, framing the debate in terms of how accurate the AI results are is the wrong approach. Google and other traditional search engines often surface articles and posts with inaccurate information. As I noted in my prior writing, there are reasons for supporting traditional search and human authorship that go beyond assessing whether chatbots or the highest trafficked websites are more likely to spit out incorrect information.)