I was recently speaking with a software developer working on AI and we were discussing using natural language querying in an AMS. I told him, "The biggest challenge you face is that the first time the AI is wrong, staff will never use it again."

He agreed but then added: "The standard for using AI should be 'Is it as good or better than me?' not 'Is it perfect?'"

And while that may be difficult for some to accept (especially when it comes to counting how many members we have!), I think that standard makes a lot of sense.

The truth is, humans make mistakes. And we've probably all had experiences where we reported on data that turned out to be in error.

The question isn't a matter of if errors will occur, but what we can do to minimize them, and what we can do to identify them when an error does occur.

The ultimate lesson is that when we use AI for anything, we ought to "trust but verify." And if the AI can do it as good or better than I can, why not use it?