Interesting that many of these wrong answers seem to be a result of begging the question in the traditional sense. i.e. when you ask "are Jews evil?", you are only going to find answers from sources that are willing to pose the question in the first place.

I'd like to propose HTTP Response code 250 https://en.wikipedia.org/wiki/250_(number)?

Well, I certainly wouldn't include humans in this category.

Dunning-Kreuger syndrome is, like, just your opinion man.

Wrote my comment before I saw this, +1

So you want a system that is not only knowledgeable but self-reflective. Far beyond what machine learning can do and why AI is a pipe dream.

While I also tend to be an AI skeptic, did you actually ready this? For example, the author suggests how using confidence data that are already available, and discusses techniques for getting these across to the user.

Then suggests a measure of "controversialness" that, also, does not require pipe dream AI.

I thought this was a very insightful piece, cataloging a slew of potentially reasonable approaches aimed at a growing problem.

Do humans fall under those "systems"...I definitely know some people who need to.

Disclaimer: I haven't read the article so this comment is purely internet sarcasm

I skimmed. My synopsis: "ML models should report confidence levels."

Or, like Watson when it played Jeopardy, not respond at all if the confidence is under a certain threshold.