[yummyfajitas]

Do you really think that a highly competent organization that's hired hundreds of data scientists can't estimate/discount the base rate of fat fingered/random bad ratings?

(Fun fact: I don't know about Uber, but I know Ola does exactly this. They use "how would you handle this exact problem" as a warmup question on their interviews.)

[TeMPOraL]

They could, but do they have incentive to?

It's good for them to keep drivers who committed (e.g. those who took Uber's help in getting a new car) on edge. It's not that they have shortage of candidates, or that they actually care about them.

[yummyfajitas]

Lets think about their incentives.

Uber, Ola and others are desperately in need of more good drivers. They are growing rapidly, and literally will pay their competitors in order to attempt to recruit their drivers.

http://www.forbes.com/sites/ellenhuet/2014/05/30/how-uber-an...

http://www.theverge.com/2014/8/26/6067663/this-is-ubers-play...

"Uber has been aggressively poaching Lyft drivers for months, offering them huge bonuses just to do a few rides on Uber. This week, Lyft started fighting back with similar bonuses..."

There are two ways to have an extra driver next week. One is expensive recruiting methods. The other is not kicking them off the service. Remember, growth = new acquisitions - churn.

Lets think now about the costs. Handling the base rate of bad ratings would cost a data scientist a month or two of time - approx $20-60k. Assuming those huge bonuses are $500/driver, you'd need to retain 120 drivers for that python script to be worthwhile.

So if Uber has no incentive to solve this problem, it's only because it's a tiny problem (affecting < 0.1% of their workforce).

Are you really claiming that Uber hires hundreds of data scientists, but doesn't actually have them answer questions like "which drivers should be fired"? What do you think they do with them?