I can't think of a counter example, but trivially I'd put a rule into "Interact-hoominz.h" to the effect that "hoomin saying 'guess' is indistinguishable from 'return any one of the three highest probability results you can estimate (be prepared for a second, different guess".Commander_X wrote: ↑Mon Mar 30, 2020 12:07 amI'll take a quick chance at this.
I think there is a word that always sets the robots (and AIs in SF, to a wider scale) to always promptly answer "I was not programmed to do this". The word in this case would be "guess". Whenever they are pressured to overcome this part of their "programming", they either give an estimate, or prognosis, but when just asked for their "best guess" they decline.
No competent programmer would get caught by such an "edge case" a second time. And any competent pre-field tester would have such edge cases in their first box of "break the new version" tools. Certainly, when I was field testing work's systems, I'd consciously spend hours trying to break the new version with edge cases, "evaluates to divide-by-zero", and that sort of thing. Which is why, with a field staff of nearly 60 people, a third of the bug reports came from me. The programming manager would throw new hires at me, for their code to receive a polite mauling.