With all the talk about Tesla's autopilot (or assisted driving), a thought occurred to me as I drove down the freeway and suddenly came across a blown truck tire in pieces. My instinct was to slow down a little and gently swerve to avoid the chunk of rubber without cutting anyone off. Luckily there was nobody in my blind spot and the minor swerve was smooth and successful. What would I have done if there were cars blocking me? Likely I would have just run the tire over at whatever speed. Mentally the calculation would have been one of damage minimization (bumper, possible more? vs. cutting someone off) rather than avoidance (there was no time to stop as the rubber beast emerged over a hill).
The decision I think was 'instinctive' and depending on variables such as other car risk, my binary decision would be different.
For a few miles I pondered this and the process that went into my action...the coulda shouldas, and the question: what would AI have done? I would love to a simulator that would show us, and would be curious if the same result repeats, or adjusts based on learning. The only thing is that as this was the first time this particular situation happened to me, we need a decision from the machine for the unknown. I'm sure eventually the swarm intelligence would eventually capture similar scenarios, and learn from it, but how does a net new situation get addressed? Is the value of a bumper weighed against the potential negative (not definite) of swerving gently into the next lane? Machine intelligence defiantly wins when it comes to the routine things, but how about the outlier events?