Slashdot Mirror
IBM Technology Creates Smart Wingman For Self-Driving Cars (networkworld.com)
coondoggie quotes a report from Network World: IBM said that it has patented a machine learning technology that defines how to shift control of an autonomous vehicle between a human driver and a vehicle control processor in the event of a potential emergency. Basically the patented IBM system employs onboard sensors and artificial intelligence to determine potential safety concerns and control whether self-driving vehicles are operated autonomously or by surrendering control to a human driver. The idea is that if a self-driving vehicle experiences an operational glitch like a faulty braking system, a burned-out headlight, poor visibility, bad road conditions, it could decide whether the on-board self-driving vehicle control processor or a human driver is in a better position to handle that anomaly. If the comparison determines that the vehicle control processor is better able to handle the anomaly, the vehicle is placed in autonomous mode," IBM stated. "The technology would be a smart wingman for both the human and the self-driving vehicle," said James Kozloski, manager, Computational Neuroscience and Multiscale Brain Modeling, IBM Research and co-inventor on the patent.
So in the event of potentially unsafe driving conditions, something with the intelligence of a 2-year-old and the strength of 1000 gorillas randomly grabs the wheel from you. I can't see anything that could possibly go wrong with this.
I was trying to drive thru my garage door that was done because the house was on fire and I was locked in. My car decided that was too dangerous, applied the brakes, and let me burn to death.
Perhaps IBM is patenting this technology to allow other companies to recognize and adopt this safety technology. How altruistic.
>"If the comparison determines that the vehicle control processor is better able to handle the anomaly, the vehicle is placed in autonomous mode," IBM stated
So I could be in self-driving mode and the computer rips control from me because it thinks it can do a better job? No thanks! Maybe if it detected I was somehow impaired, but the idea of my control being removed randomly is not attractive. Sounds like it would be OK if it were an OPTIONAL mode/setting (3 modes- computer drive, human drive, or hybrid/auto-switching).
>"co-inventor on the patent."
Really? That is worth a patent? Seems like a pretty obvious idea to me. Oh, since I thought of allowing it to be 3 separate modes, could we patent that??
Will it throw itself onto a Fiat multipla?
Fact is _I'm_ a reliable wingman. Total darkness is your friend, I'll take that grenade.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
You! You are still dangerous. But you can be my wingman anytime.
Imminent death predicted. You have 4 milliseconds to take control. Thank you for choosing IBM. Would you like Watson to choose a casket for you ?
Nullius in verba
Just a couple of seconds before the collision the AI releases control of the vehicle back to the human and wakes up Leroy because he ain't never seen no accident like this one. :)
Seriously, this looks like the perfect way for the AI to not be held responsible for any accidents. What are they going to think of when there is no option (no steering wheel, pedals, etc.) for a human to take control?
So while other companies are actually developing self-driving cars, IBM is looking at what they'll need and throwing patent mines into their path.
AI: "WAKE UP! Take control now!"
Me: "AAAAAHHHHH!!"
Car: *crash*
#DeleteChrome
When driving, one's brain is immersed in the steam of visual data - preempting problems.
To suddenly be asked to context switch between being a passenger and driver is dangerous.
Requiem for the American Dream
Shove your Personal Restraining Capsules up your asses, human hands will remain on our car wheels TYVM.
Don't mix humans and AI's. Give them their own roads. If they want to build an underground system and overhead network that meshes with existing roads that is well planned. Even someone like me is likely to try it. I might even like it!
human driving and is given the choice of swerve into the path of an oncoming car or hit a pedestrian, AI takes over and hits the pedestrian instead of leaving the human to take the 3rd option and put the car in the ditch and avoid the other choices.
do not trust a computer with choices of this magnitude, leave it to people.
...and it is, according to nearly every engineer in the autonomous vehicle business, including the head of Google's autonomous vehicle project, unsolvable. It is at the core of the current regulatory conflict between legislators, who want to keep a human in the loop, and most autonomous vehicle makers, who want humans out of the loop because of the unsolvability of the hand-off problem. Google has already stated they will not produce their autonomous vehicles until the government agrees to remove the human-in-the-loop requirement for operating autonomous vehicles on public roads.
The major exception to this received wisdom is Elon Musk at Tesla Motors, who pretty much believes that no problem is unsolvable. Yay for him; we need incurable optimists. If we are ever going to think our way out of the crap-sack world we are currently headed for, it will be because of technological optimists like Musk.
I realize that betting against Elon Musk is probably not a good strategy, but I do think the hand-off problem is not solvable, for any value of "solved" you care to assert. Human task switching is not fast enough at pedestrian velocities, let alone autobahn velocities. Dragging a human from their porn/spreadsheet/email/phonecall/whatever and expecting him/her to correctly grok a traffic problem in a fraction of a second is not realistic.
Or even several fractions of a second, if IBM's cognitive modelling of human drivers, which is what this expert system is actually doing, successfully expands the window for a driver to react. I am not sure that replacing a human with a robot that is programmed to learn to drive like a human is a win, but we'll have to see what happens. If we don't have a spike in traffic fatalities in BMWs (IBM will be rolling this package out in BMW models first, because they have already announced they are giving Watson to BMW) I will happily revise my estimation of the hand-off problem's solvability. :)
This idea is a total non-starter.
It's less safe than human-only full stop.
First, any revolutionary rather than evolutionary self-driving capability, anything that can't be spun as "beefed up cruise control," will be blocked by gov't and proles on safety. How will you spin this well enough to get past safety concerns? It's like letting drunk people drive as long as there's a sober passenger available to grab the wheel. It really does not pass common sense.
Second, there is now significant competition among programmers. Eventually, one of them will probably get far enough along to do full-autonomous, meaning:
- handle exceptions by pulling over to the side of the road, either for towing or handoff, depending on whether there's a steering wheel or not.
- match human safety end-to-end with this system
If there's competition between a luxury that's a little more expensive but kills your neighbors less, and a cheaper discount one that occasionally murders bystanders, do you think both will go on sale?
Basically, there is no competition, unless you can match the safety numbers of the best player so that the difference between you is in the statistical noise. IBM's approach will not get you to market. It's a money-pit.
Thank you for playing, IBM, please drive through now.
But what if the human is deemed to BE the anomaly?
If a working model of a invention was either required or, more likely, resulted in a form of stronger patent then a lot of these hand waving type patents could be avoided. A patent should only apply to a working model, not an "idea". As anyone who actually does something for a living can tell you the idea is not the hard part.
Computer: situation has occurred that human intervention is required, please respond.
Human: zzzzz
Computer: response acknowledged. Thank you
Because it's not like EVERY SINGLE AUTONOMOUS CAR currently being tested already does exactly this, forcing the driver to take over when it encounters a situation it can't handle. Or like the driver assistance features already available in cars do this the other way, such as hitting the breaks for you if you're about to run into something.
But it's ok, they added "with machine learning", so that makes it new. I guess that's the new version of "on the internet". Take anything that people are already doing, add the words "with machine learning", and now you can patent it.
"I'm too busy to research this and form an educated opinion, but I do have time to tell everyone my uninformed opinion."
Turn onto the gravel road, HAL.
I recall learning, oh, something like 30 years ago, that the Space Shuttle had multiple computers, running at least two different operating systems, managing all vital systems on a space shuttle.
With all the concern about self-driving cars being cracked, or otherwise running into problems, why is no one demanding something similar? The computers themselves are pretty cheap these days - and will be cheaper by the time we start putting this in every car. Just have a minimum of three computers running a minimum of three different operating systems, determining what the car does. One of those computers can have priority for decisions about where to drive and such (with a human override, which could be as simple as changing which computer got priority for these decisions), but the other two computers would monitor every move for safety. If a single computer returned results outside of safety parameters, the car would shut down until the problem could be resolved - and control of safely slowing down and stopping would be according to the majority of the computers.
Cracking the automobile's control system would still be possible, but it would require that at least two different kinds of systems be cracked, almost simultaneously.