If there's one company who hasn't demonstrated it's learned from its mistakes it's Uber.

But let's extrapolate. Say one day there are 20 self driving car companies. Should they be required to share what they learn so the same mistakes aren't repeated by each company or does the competitive advantage outweigh the public benefit from this type of information sharing?

Nice case for a "Reductio ad Absurdum":

Should "not killing pedestrians" be a competitive advantage? No. It should be mandatory for every vehicle.

Therefore,we should make sure that everything works towards this goal, including sharing data!

Federal regulation demanding shared accident databases wouldn't be the worst thing...

What mistakes has Uber not learned from recently? They changed their leadership. That shows some learning.

"recently" is a pretty big caveat to throw in there. Historically Uber hasn't changed any of their behaviour until they've been caught doing something. Let's wait and see how they react to something like this.

Good point, but I added that in because I think recent is the most important for trying to know if something learns from its mistakes.

They seem to be adjusting quite a bit in the last 1-2 years so that is the definition of learning. Otherwise, just showing how companies were once dumb is usually not very useful in knowing how adaptable they are.

I agree that we’ll know better looking back how they react to this event.

I think there is a case for open sourcing the decision making code for this exact reason.

Or does the prospect of onerous regulation outweigh the competitive advantage?

Airlines are the same way. Every time there’s been a crash they’ve learned something and changed procedures to prevent it again. It’s made flying pretty safe, unless you’re an animal flying on United...

Uber's algorithms might learn from this. None of the other self driving car companies will.

I disagree. I would be astounded if Waymo, Tesla, and every other player in the field isn't already trying to craft 1000 new test cases about this incident. It's now the second(?) known death involving this technology. The first question their bosses will be asking them in their next meeting would be "would our cars have done this, too?"

That’s where the ntsb steps in and regulations created.

I'm having a hard time imagining that kind of regulation of ML algorithms.

Making sharing of all sensor data of an accident and the preceding couple of minutes mandatory could be a simple regulation that would be very helpful for all companies trying to improve their algorithms.

That isn't correct. Culture is learning from other people's mistakes. The only question is whether the human can and is willing to accept the lessons (and whether the lessons are correct).

Culture is comprised of many things, and some similar things, but not that. By definition, culture is merely a collective persistence, and therefor inherently superficial. Culture cannot capacitate “learning” or even lessons. Culture is limited to memetic abstractions, with a lot of alleged noise. Insofar as culture supports persistence of worthwhile norms, it would be limited to evolutionary adaptation (survival of the fittest). It’s important to distinguish to avoid the fallacy of argumentum ad populum.