Hacker News Reader

Comment by toomuchtodo | Hacker News Reader

Neither Uber or Lyft will reach full autonomy before exhausting their runway, unless a miracle occurs and they solve level 5 within the next 1-2 years (caveat: it could happen, but it’s very unlikely).

You can’t do hundreds of millions of dollars in raises forever without some proof of progress (or maybe you can if SoftBank is willing to pour money in for a decade).

At some point, someone is going to be left holding the bag when it’s evident that the last 20% is the hardest part (but mandatory for level 5 service).

View on HN

I don't think level 5 would be the next step for self driving automation. The biggest ground based fleet with very high autonomy is John Deere. Therefore, I would expect large commercial trucking to be the next logical target. It would cut their costs alone by changing how much their insurance policies and general efficiency. John Deere trucks almost drive themselves. So, I would see trucking be replaced first. It looks like Elon Musk wants to do that with their new set of Trucks. Since they would have the same sensors as the cars that would be a bunch of data to collect. It would also be a easier way to get proof of progress. But, at this point if you replace trucking you will get a large economic change because you are changing one of the largest source of jobs in America.

Then since you have proven that you could try acquiring Uber and or Lyft at a fire sale price because they couldn't come up with a solution in time and get their data and logistics.

Yep, you’re right on both points (trucking being the first industry that will adopt self-driving technology on mass-scale and the potential displacement of millions of jobs) but I’m not sure if John Deere will be spared in the process, only if they’re smart and see the same writing on the wall.

Not just that, getting self-driving approved with regulators is a major challenge. Also, to achieve level 5 we need not just algorithm improvement, our physical infrastructure need to be improved as well. On intersection, two human drivers wave at each other to decide who goes first. How should the human driver communicate with the machine? What should the machine do when there is a dog standing in the middle of a street? Move on and let a human driver take care of it? Isn’t that now a bystander effect? These are just some examples I am not sure if textbook definition of level 5 will address.

Going really extreme: in a car accident in a somewhat remote area, the driver might be able to get an injured passenger out if the passenger stuck. Now a fully autonomous car, there is only passenger. Yes, for sure this is extreme and perhaps likely a dramatic scene only, but it is one of those edge cases we can’t cover in a self-driving car.

> How should the human driver communicate with the machine?

In theory, a self-driving car could watch for hand-based turn signals, just like a human driver could, and take those into account.

In practice, what someone signals is far less interesting than what they do, and all it does is provide hints for reaction time, which 1) a computer already wins at, and 2) a computer should always leave enough time for braking/stopping/etc anyway. A human driver should never make an assumption based on someone's turn signal that would lead to a crash if they don't do what they're signaling, and neither should a computer.

> What should the machine do when there is a dog standing in the middle of a street?

Don't hit it (stopping if necessary), and listen when the human in the vehicle says "stop" (or "call animal control", or any number of other reactions). The human in the vehicle should always be in control of the high-level actions, just not the moment-to-moment ones.

> Going really extreme: in a car accident in a somewhat remote area, the driver might be able to get an injured passenger out if the passenger stuck. Now a fully autonomous car, there is only passenger.

Not sure what scenario you're getting at here. If you mean a driver and passenger in the same vehicle, the same problem exists today with an injured driver.

The algorithm doesn’t have to be perfect, it just has to be a significant improvement over human beings. For every fatality an algorithm causes that a human would not, there could easily be ten fatalities prevented.

You do realize that most people won’t accept that, right? Worse, when the algorithm does kill someone, who is responsible?

Edit: Instead of just downvoting and moving on, why not actually formulate a response? Is there anything more empty than disagreement without the capacity to argue your case?

Won't accept which - that computers only have to beat human fatality rates to be preferable, or that the ratio will be 10:1 (or better) in favour of computer-driven cars in terms of fatalities?

I'd agree that if we aren't seeing better fatality numbers, people won't accept it -- but there's little to suggest that humans are a difficult target to beat. Especially in the US, where car-related fatalities are so numerous and much of the research and trials are being done.

Working out accountability for fatalities is already a hugely complicated process, involving an unhealthy number of law enforcement and legal personnel. The problem of identifying responsible party for a - as you put it - failed algorithm doesn't seem intractable, and certainly isn't a show-stopper.

Haven't voted either way on this thread, but I suspect your re-hashing of regular topics already resolved is the problem.

Not the OP, but: I don't believe people will accept that computers only have to beat human fatality rates to be preferable.

People have this weird psychological effect around control. They will accept higher failure rates if they believe (correctly or otherwise) that they have control over a situation. Or rather, they will believe that they can beat the statistics at least for themselves, as long as they believe they have control over the situation, and achieve a lower failure rate then the driverless car. In other words, "the higher failure rate might apply to the average person, but I'm above average".

It's related to faulty risk assessment: it's much more likely that you'll get in a car and die than it is that you'll be killed in a terrorist attack, but most people don't even think of getting into a car as all that risky, and yet we start drawn-out, unwinnable wars because people are afraid of terrorism. People believe that they personally won't die in a car crash because they have control of the situation and that they're a good driver, but they have no control over when or where a terrorist attack will take place, so it's somehow a worse threat.

Well, half of people are above-average though. Maybe the people who you talk to are just all above-average.

> You do realize that most people won’t accept that, right?

That's why we have courts (and insurance).

Conventional vehicles kill people all the time, sometimes through provable manufacturer negligence. But we still have them.

It’s funny, you failed to mention the group with the real power there... politicians, specifically, Congress.

This is currently being tested in court. Tesla's autopilot has resulted in fatalities.

That's different, though. Autopilot doesn't absolve the driver of the car from responsibility to be alert and take control when necessary, so the argument can be made that the responsibility is shared, or even rests on the human driver, depending on the circumstances. But a level 5 system will theoretically not even have a driver, and might not allow passengers to control the vehicle at all aside from giving it a destination.

I think that becomes easier, though: in the case where no human in the car has control over what the car does, it's obviously the manufacturer's fault when something goes awry. Though if doing something like rapidly and repeatedly changing the car's destination and intentionally "confusing" it leads to a crash, that might shift blame to one of the passengers, which would be unusual.

Reply on news.ycombinator.com