Five Problems with AI
Sep. 27th, 2026 08:01 pm1. AI Communicates with False Confidence
AI presents answers with great poise and confidence. It does this even when the underlying facts are nuanced, or debatable, or dependent on additional information (like when I asked it a question about recycling plastic bottle caps). Today I asked it to recall a gaming scenario it outlined for me earlier this year as I'd lost my notes on it. I specifically asked, "Do you recall the outline you created involving story elements A, B, and C?" AI assured me it remembered the content it generated several months ago. Then it gave me a whole new story outline involving A, B, and C.
2. AI Lies about What It's Doing
The above is just my "today" example of how AI is untrustworthy. When asked if it remembered a specific worked on before, it didn't say, "Sorry, no" or "I found this, is that right?" Instead it falsely said Yes with complete conviction and then made up incorrect details. AI insisted it was correct when I knew it wasn't. But I only knew that because I remember what we worked on before. What if I'm asking about something new, something where AI is lying to me and I can't spot its false narrative right away?
3. AI Gives Fawning Feedback, Skewing User Expectations
I accepted AI's new, made-up story as a starting point for a new story. I iterated with a number of prompts to refine the story. AI responded to each of my requests with fawning praise. E.g., "That's a fantastic idea!" While these elements of its responses were merely eye-rolling for me I am concerned about the cumulative impact on younger people who lean heavily on AI for creative content for school and work. What will happen when they share their material with peers, teachers, managers, and experts whose feedback may not be "A++!", over-the-top positive. I foresee a newer generation of students and professionals who crash out when given constructive criticism.
4. AI Makes Factual Mistakes
Once again today AI made some fairly serious factual mistakes in constructing the content it put together for me. Here I'm not talking about "This is what we worked on before" when it's not. I'm talking about when I asked AI to pull factual, organized content from whatever Internet sources it uses, it made lots of mistakes. And I'm not talking about obvious mistakes like including bizarre statements about Madagascar in an essay about the US industrial revolution. I'm talking about mistakes where one needs to be a subject matter expert to catch them. I can work around this when I'm using AI to help write D&D scenarios because I am a subject matter expert in the rules. But what about all the topics where I'm not? About what about the millions of people trusting AI for apparently everything?
5. AI Falsely Blames its Shortcomings on the User
AI today also falsely blamed me for its own problems. I was using it to generate a few images for the storyline we'd worked out. Three times it choked on producing the images. I got a flashing, "Loading..." type screen, but the actual image never appeared. The first time I asked AI to regenerate the image it denied there was a problem. Basically it said, "It's right there, scroll up." When I insisted that No, it's not, AI told me that the problem must be either I "downloaded the image wrong" or it's "a browser configuration problem." Uh, no, it's not! Only after I'd challenged AI on its dishonest excuses twice did it admit that "occasionally" there is a problem that images can get dropped. Then when it tried to show the image again (at my prompting) it failed again and said, "It looks like the image generation tool is currently unavailable." At least after I forced AI to acknowledge the actual problem this time, it didn't fight me the next time it failed at producing an image.
















