Mercator
New Member
Discussing Toronto's approach to the introduction and integration of self driving vehicle technologies.
The issue of bad weather is obviously especially pertinent in our climate, and I think it is no accident (ha!) that almost all the autonomous vehicle work is being done in sunny California.
- Moral -- Cars that are faced with an unavoidable fatality decision are going to decide whether to save the occupants or pedestrians. Picture the scenario of a baby stroller suddenly running in front of the car at the last second, from behind roadside newspaper boxes (unavoidably unseen by the car's sensors until too late; now a fatality has to happen). Car must instantly decide to crash into baby stroller OR suddenly veer into a parked car/lamppost 1 meter to the side. Legally solve this. Now consider the sole occupant of car is your child being soccermomed unaccompanied to school. Whose life goes? Whose Responsibility? Legislative? Insurance? Etc
It's a though experiment which in practice doesn't seem very relevant. Autonomous vehicles, like humans, won't ever be making their decisions based on full information. They'll never be able to know everything about their surroundings. Maybe that women's stroller is just being used as a shopping cart and there is no baby in it, meanwhile maybe there are two kids in the parked car.
If an autonomous vehicle detects an obstruction that will likely cause an unavoidable accident, the best solution will always be to just apply the brakes instantly. Even if a collision is unavoidable reducing speed from ~40km to ~20km and informing trailing vehicles to brake immediately to avoid further collisions would be the morally preferable course of action.
Expecting cars to perform some moral-kamikaze is more likely just to cause automated versions of that accident were a woman caused a pileup trying to avoid hitting some ducks.
Your example is the easy one for the computer. But I read a better example in an article a month or so ago. Say a car is on a mountain road and all of a sudden a person is blocking the road. There are 2 options (and only 2). One is to hit the breaks but still hit the pedestrian and there is a 100% chance they will be killed. The other option is to steer off the cliff and kill the people in the vehicle. What would the computer be programmed to do?
How will it affect public transit? Rapid transit corridors will probably boom (e.g. subways, commuter trains, high-traffic bus/LRTs, high speed trains) with self-driving minibuses/Ubers/transit/taxis/etc becoming one of the many last-mile connecting options, while low-traffic suburban buses may die (e.g. UberPOOL style services may be more profitable for lower fares per person, and TTC/GO may run UberPOOL style dial-a-bus services much more efficiently than a fixed-route suburban bus, for example -- assuming they keep dynamic pickup/dropoff routes as ultraefficient as possible, and number of occupants per vehicle low -- then it can probably beat the transit economics and transit performance performance of a low-usage suburban bus route).
.....
Sheer numbers and road capacity pretty much guarantees continued need for rapid transit options. It definitely will impact transit, it will strengthen the raison d'etre for rapid transit along central corridors, it will restructure inefficient transit (e.g. circuitous suburban bus routes with few passengers), but it isn't going to kill transit as a whole.
The default of "only always try to brake in front of imminent collision" isn't going to clear-cut it in 100% of all situations.That's a moral and utilitarian jackpot that would always outweigh some kind of outlandish attempt to avoid collision by executing James Bond maneuvers.