Molson Hart

Joking, but...I wouldn't put it past them. We will delete the algorithm. A driverless car killed a human recently. Supposedly, their rate of accident is much lower than human drivers. Nonetheless, we are distrustful. Is it because we intuitively know that algos lack SITG?

Yeah I heard, it's terrible :/ Good question. I think engineers would argue that the alg has a downside as these incidents are fed back to the alg. But. Maybe it's the engineers designing the algs who have no SITG and that's why we don't trust them.

I just read somewhere on twitter that computer science is a discipline that has yet to learn that actions have consequences. I think we are about to learn that soon and painfully.

I shared that. Word. Looks likely.

That's right it was from your feed.

I've spent a lot of time thinking about this problem the last few years. The current dominant engineering mindset (invent first fix it later) works very well for smartphone apps but not so much for self driving cars. Convex vs concave payoffs.

It reminds me a little bit of global warming. Suppose you have perfect regulations - they neither stymie development of AI nor allow for a "black swan" event, it's not enough because China and India are going to march to the beat of their own drum.

There's one crucial difference though. You can ban importing Chinese self driving cars into the USA by law but you can't ban the wind to blow over.