What the dataverse hackathon presents is neither an engineering problem nor an AI problem. It is a human behavior problem. Lives are at stake: whatever methods we use has some flexibility. We can't know what's for sure true, but we can be certain of lies we make ourselves.
What it does
The operator will begin the log as soon as the repair starts. As they are making the repair, they are mandated to input data as they go. At (semi-)regular intervals, the system will randomly change one or more input fields while keeping the original inputs. It will then be the operator's job to find any and all inputs and change them as necessary. During training, we can run this program on any pre-existing or future incoming input fields to teach operators how they should be inputting data. In deployment scenarios, we can use this as a validation test to see whether we can trust our operator's logs.
How I built it
This should be a simple backend plugin that runs on low-spec military laptops.
Challenges I ran into
Marketing! I mean if you start with: "Let's make the mechanics' lives harder" it can be a hard sell. (Laughs self-deprecatingly).
Accomplishments that I'm proud of
Thinking of this. Using empathy to put myself in the shoes of business managers who have fucked me over in the past.
What I learned
Presentation skills. The mechanic repair hierarchy and day-to-day operations from primary sources.
What's next for Sabotage!
User testing! Test in a technical school! If it works, try on live scenarios at increasing stakes.