There are assumptions behind every AI platform’s answer to our questions. AI tools don’t identify the assumptions as a standard practice. That’s why it’s critical to ask about the premises used to generate the answer.
Sterling Miller shared his solution in a webinar earlier this week on generative AI prompting for contract drafting. He suggests using a follow-up prompt after the tool gives its answer.
The prompt asks the tool whether its analysis was built on any assumptions, any incomplete information, or anything that’s uncertain. This technique surfaces the unsaid qualifiers and background used by the tool to generate that answer.
Unknown assumptions represent a significant problem in our ability to rely upon what the answer given. It happens because we aren’t giving the AI tool every piece of information it determines it needs to respond. When information the tool thinks it needs is missing, it may supply its own version of the facts and keep going.
Using a follow-up prompt makes the tool point to those search results or generated context. Knowing that, we can then check each assumption against the real facts and circumstances.
This kind of verification is not new. We’ve been using this technique whenever we review the work of someone with less expertise or experience on a topic. We ask that person for the reasons behind the decisions to include this fact or characterize what’s happening in a specific way.
We need to do the same with our AI generated answers. We need to seek out assumptions, correct or clarify the wrong ones, and send it back for a new work product based on those updated inputs. The conversation can continue until we are confident that the output is based on the right context for our question.
I learned a lot from Sterling’s discussions best practices for prompting. I’ll definitely be adding this assumption question to my AI workflows.
