The Disinformation Dilemma: The Inevitable Degradation of AI Chatbot Data

Here’s the first problem: as the technology currently stands, AI bots occasionally get things wrong. We’ve all seen AI’s limitations when it comes to generating images based on text prompts: people typically (and creepily) possess six or more fingers and way too many teeth. For some reason, the AI isn’t able to grasp that human fingers need to stop at five and teeth at 32. These facts are lost; the correct generalization never manifests. Why? Because AI doesn’t understand “hand.” It’s only playing the odds, drawing from the common characteristics of its source material.

Perhaps the most embarrassing public example of this occurred on February 6th when Bard, Google’s “experimental conversational AI service,” made a factual error in its very first public demo. Bard was asked, “What new discoveries from the James Webb Space Telescope can I tell my 9 year old about?” One of Bard’s suggestions states the telescope “took the very first pictures of a planet outside of our own solar system.” But NASA says the first such picture was taken in 2004. DOH!

ChatGPT doesn’t fare any better. Not even according to Mira Murati, CTO at OpenAI, the…