1
I ran a small test to see if an AI would flag its own bias, and the answer was a hard no.
3 comments
Log in to join the discussion
Log In3 Comments
jakeb8129d ago
What was the test you ran?
7
I tried something like that last week with a popular chatbot. I asked it three times in a row to check its own answer for political slant, and it kept insisting it was completely neutral every single time. It was a very clear example of not being able to see its own blind spots. The whole thing made me trust these systems a lot less, honestly. They just don't have the ability to self-critique in a meaningful way.
6
foster.tessa29d ago
Oh man, I did something like that too! I asked it to write a summary of a recent political debate, then immediately asked if that summary favored one side. It just rephrased its first answer and said it was balanced. Super frustrating because the bias was obvious to me. I had to point out specific words it used before it would even sort of admit a slant.
2