Problem is that "human" here would mean outsourced idiot in a boiler room that isn't trained, paid or treated well enough to give a shit, so what you'll end up with will be no better than basing the ban decision on a boolean RNG.
From my experience around one in twenty rides the driver tried to pull some sort of scam on me. In Portugal the service was awesome, but in Austria, Hungary, Poland, South Africa and Argentina the experience was very mixed (ranging from great to poor).
- does the driver ever have a chance to interact with the person making this judgement? Do they have an opportunity to provide information which Uber was not in a position to automatically collect (e.g. a road was blocked due to construction or an accident)?
- what is the goal of the human reviewer and what incentives are put on them? Are they required to 'judge' a number of cases per hour, which places a limit on how much context they can actually consider? Are they responsible with arriving at a "fair" decision, or merely one which minimizes costs to the company?
- what outputs from the human reviewer inform future automated decisions? Are they merely adjudicating individual cases, or will their decisions help train the algorithms that automatically flag drivers going forward?
Merely having a human in the loop is insufficient. It's easy to create a system in which humans effectively rubber-stamp and amplify issues in the automated systems.
