pic: Google Streetview driver autograph

[techhelpbb]

Originally Posted by CNettles11

The same way we found the Google Street View car. I was hanging out with a few friends for my 16th birthday. We were having lunch at Taco Bell when our FRC team's website guy said "Is that a Google car?", so we tore off after it on foot. After we chased down the car and I got the driver's autograph, we went to a grocery store to buy some drinks. We ran into the lead mentor of team 3556 there and we showed her the autograph. We spent a few more minutes roaming the store and we were in the cracker aisle when I asked our FRC team's lead programmer to take a picture of me with the autograph.

Great so now every image that the Google Street View car took has you in pursuit of it?

[CNettles11]

Sadly, no. They weren't taking pictures. The car was en route to it's next picture taking location.

[bassoondude]

Originally Posted by CNettles11

Sadly, no. They weren't taking pictures. The car was en route to it's next picture taking location.

Aww, that would make an awesome easter egg for the world to find.

[CNettles11]

It would be publicity for team 3556, two of us had our team's shirts on.

[Whippet]

I envy you so much right now.

[JesseK]

I can't find the transcript, but I read the black box voice recordings from the final minutes of the AirFrance 447. It essentially came down to a confusion between what the autopilot ("fully autonomous") was supposed to and not supposed to do. It also came down to 2 co-pilots being unable to sync flight controls with each other due to an automated 'smart' algorithm that replaced a physical link between the flight sticks.

Let's take that premise and think long term, on the order of decades. It's difficult to do for those of us who are only a couple of decades old. Yet let's suppose autonomous cars do hit the streets for the common person next year and gains popularity so that 20 years from now about half of them are autonomous (literally, millions of cars).

I seriously doubt Google overlooked the AirFrance case -- where the human in the loop gets complacent because the software works "99.999% of the time". Yet last time I checked, that 0.001% failure rate still means the autonomous vehicle could be at fault for several thousand incidents because the human didn't take over in time. That 1 in 100,000 ratio is enough to cost someone their job. It's also enough for the programmers in question to remove "developed and integrated Google's autonomous vehicles" from their resume. Finally, if any of the incidents were fatal, would that one person's death still be honored by vehicle autonomy, or would there be an acceptable threshold so long as the overall fatal accident rate in the country fell?

If anything I would hope Google's end goal is autonomy that augments or constrains a human's ability rather than completely supplanting it. I conjecture that if we constrained most incidents of aggressive driving then most fatal accidents wouldn't occur.

Mind you, I wasn't referring to the hoards of cab drivers who may be out of work in 50 years. Nor was I referring to "true AI", where the code really is intelligent and can re-write itself to change its behavior. (Ironically, the programmer would then go from being a computer programmer to being a computer psychologist...).

[The Lucas]

Originally Posted by Tom Bottiglieri

Yes, I think that's kind of the point. A manually controlled car is just a broken autonomous car.

"Manual Mode" is available if you dont mind the occasional car crashs

It will be interesting to see how communication between cars, street lights, central highway controllers, etc... develops if auto cars become the norm.

Now if only Tacocopter was legal (and real) Google wouldn't have to autonomously drive blind people to Taco Bell

[neshera]

Just as a thought exercise:
If someone develops a non-invasive cure for cancer, many nurses and surgeons will be underemployed/unemployed. Would this be a bad thing?

[Ian Curtis]

Originally Posted by JesseK

I can't find the transcript, but I read the black box voice recordings from the final minutes of the AirFrance 447. It essentially came down to a confusion between what the autopilot ("fully autonomous") was supposed to and not supposed to do. It also came down to 2 co-pilots being unable to sync flight controls with each other due to an automated 'smart' algorithm that replaced a physical link between the flight sticks.

This is a gross oversimplification of an extremely complicated flight controls and human computer interactions issue. Just because the Airbus side stick is not physically linked like a Boeing yoke and does not have force feedback does not mean that the controls were not "synced." Please do more reading.

[JesseK]

Originally Posted by Ian Curtis

This is a gross oversimplification of an extremely complicated flight controls and human computer interactions issue. Just because the Airbus side stick is not physically linked like a Boeing yoke and does not have force feedback does not mean that the controls were not "synced." Please do more reading.

The pilots weren't in sync with each other, as demonstrated at the 02:13:40 mark here. The AirFrance 447 was used as a mere example of the exact complexities you're talking about. I don't think I 'grossly simplified' it.

The issue at hand in a car isn't whether 2 drivers need to be in sync or whether steer-by-wire is 'safe', 'important', or 'whatever'. It's a fundamental tendency for humans to stop paying attention to the dynamic, constantly-changing situation the vehicle is put into since 'autonomous' can do it 'better'. The Airbus situation, where the pilots had to take over in an emergency, is easily multiplied on the road because there are astronomically more interactions for a computer to be unable to handle while driving.

For the thought exercise -- doctors who cure cancer (but which one of the hundreds that have been identified??) are still fundamentally doctors. They will always have inifinite patients.

[Aren Siekmeier]

Originally Posted by JesseK

The ... driving.

You definitely raise valid concerns about software's potential shortcomings. I agree that we probably shouldn't make any quick move towards fully automating something like driving (especially when there are others humans on the road, whom we certainly can't expect software to be able to predict). I was mostly pointing out that the blind protection of jobs for the sake of keeping employment up and not displacing people tends to be counter productive.

However I also agree that (from the little I have read about AirFrance 447 from your posts and elsewhere) the AirFrance 447 situation isn't exactly relevant. It seems to only be a case of the pilots not communicating. They weren't aware, as they certainly should have been, that the two sticks could be in different positions. And they didn't communicate their differing intents. And yes, secondary to those problems, the software took the average, which certainly didn't help the situation, but it's hard to perfectly handle the edge case of getting two very different inputs.

[Aren Siekmeier]

Originally Posted by neshera

Just as a thought exercise:
If someone develops a non-invasive cure for cancer, many nurses and surgeons will be underemployed/unemployed. Would this be a bad thing?

Absolutely not, is obviously the answer you're looking for. As JesseK suggested, doctors would still have plenty to do treating patients. Cancer researchers will still have plenty to work on. Even with some non-invasive cure, there will be research that can be done into improving the treatment, or finding alternatives, or even if cancer becomes a complete non-issue, there will always be more to research. Similarly, the nurses and surgeons will continue to have surgeries to carry out, or if the demand for invasive procedures really does dwindle they can adapt to the developing demands of carrying out these procedures (already, more and more surgeries are being carried out via ultrasound or endoscopy). It's akin to the field of physics. I often wonder why people are so driven to discover a "grand unified theory of everything," as if after that we will know everything there is to know about the universe and our "toil" as physicists will be done. There will always be more that we don't know.

[Tristan Lall]

Originally Posted by JesseK

The pilots weren't in sync with each other, as demonstrated at the 02:13:40 mark here. The AirFrance 447 was used as a mere example of the exact complexities you're talking about. I don't think I 'grossly simplified' it.

The issue at hand in a car isn't whether 2 drivers need to be in sync or whether steer-by-wire is 'safe', 'important', or 'whatever'. It's a fundamental tendency for humans to stop paying attention to the dynamic, constantly-changing situation the vehicle is put into since 'autonomous' can do it 'better'. The Airbus situation, where the pilots had to take over in an emergency, is easily multiplied on the road because there are astronomically more interactions for a computer to be unable to handle while driving.

I think Ian was getting at the fact that the AF447 situation wasn't a failure of the automation itself, it was a failure to understand how the aircraft would behave with the automation in "alternate law" mode. In "normal law", which is how the majority of the pilots' training was conducted, a reasonable procedure for an emergency climb is to pull straight back on the stick, and engage maximum power (takeoff/go-around throttle settings, known as TOGA). Instead, due to pitot icing (i.e. neither the pilots nor the automation were able to get accurate airspeed indications), the system defaulted to alternate law, a more manual mode, and the pilots didn't respond correctly. One pushed forward to gain speed, one pulled back to gain altitude, when in reality they needed both, but needed speed first to avoid stalling, then altitude to avoid crashing. The pilots did not understand the implications of those opposing control inputs.

On a fly-by-wire Boeing, the pilots would have been fighting each other, due to the interconnected yokes, and probably would have realized what was going on. The autopilot would still not have worked, and the pilots would not have had an airspeed reference (due to pitot ice). Therefore the aircraft would not have been able to deliver the stall warning (though it probably would have said something about angle of attack).

In other words, these were conditions that would incapacitate any autopilot, and seriously diminish the ability of any pilot to fly manually. A more correct analogy to make with the Google cars is to wonder whether they would let the cars continue to operate in conditions that were unusually snowy, or with a malfunction in the drivetrain or steering. I think the answer is clearly no. (Of course, a car can pull over and wait; a plane has to land first.)

[techhelpbb]

I've made quite a few handicap retrofit components over the years. Including gizmos to drive cars with joysticks and the like without actually removing the driver from the equation. Mostly for people that have physical issues that would prevent the operation of the vehicle otherwise.

The odds that Google will succeed in convincing people that their automated driving cars are perfectly safe, so safe that no further mechanisms are required are slim.

There are so many very real factors to consider. For example, if your car stalls you loose your power brakes and power steering. How many people here have tried to drive an old car without these features? If you do you'll discover why often handicap retrofits require auxiliary electric hydraulics in the event of a stall. No simple retrofit like you see on MythBusters turning the wheel will accommodate this situation.

To make matters worse, you have to consider the transmissions. Automatic transmissions usually get worse gas mileage than manuals, and with the exceptions of the hybrid styles of the 2, you're likely to end up with something less fuel efficient and possibly worse on the wear and tear of the brakes or clutch.

Course if you make the car electric you might not even need the clutch (electric motors have extremely high torque output so they don't need the clutch if the electrics can handle the load at gear change, they don't stall when they loose RPM usually).

This is all a very clever idea, and I support all the safety warnings that drivers might gain from this experiment (there are so many safety features car makers just don't use that already exist or could easily be put into production). However, I can't see anything Google puts on the road today getting anywhere near the requirements to put into mass production.