How to ensure that AI and automated decision tools are traceable?
My company want to introduce AI for automated decision making on insurance claims. They are asking me to make sure that any tool that we use is traceable and algorithms can be explained in case of a review/audit and to have strong governance practices in place. I am leaning towards using Natural Language Processing and intend to stay away from LLM due to their limitations with auditing etc. Any suggestions from HN?
I'd suggest hiring a consultant to help with this. I mean, I'm the last person typically to engage in "gatekeeping" but on something like this, a mistake could represent an existential threat to the entire firm. You don't want to be "learning as you go" under those circumstances, IMO.