Stories
Slash Boxes
Comments

SoylentNews is people

posted by Fnord666 on Tuesday April 11 2017, @09:11PM   Printer-friendly
from the do-you-know-how-you-think? dept.

Will Knight writes:

No one really knows how the most advanced algorithms do what they do. That could be a problem.

Last year, a strange self-driving car was released onto the quiet roads of Monmouth County, New Jersey. The experimental vehicle, developed by researchers at the chip maker Nvidia, didn't look different from other autonomous cars, but it was unlike anything demonstrated by Google, Tesla, or General Motors, and it showed the rising power of artificial intelligence. The car didn't follow a single instruction provided by an engineer or programmer. Instead, it relied entirely on an algorithm that had taught itself to drive by watching a human do it.

Getting a car to drive this way was an impressive feat. But it's also a bit unsettling, since it isn't completely clear how the car makes its decisions. Information from the vehicle's sensors goes straight into a huge network of artificial neurons that process the data and then deliver the commands required to operate the steering wheel, the brakes, and other systems. The result seems to match the responses you'd expect from a human driver. But what if one day it did something unexpected—crashed into a tree, or sat at a green light? As things stand now, it might be difficult to find out why. The system is so complicated that even the engineers who designed it may struggle to isolate the reason for any single action. And you can't ask it: there is no obvious way to design such a system so that it could always explain why it did what it did.

The mysterious mind of this vehicle points to a looming issue with artificial intelligence. The car's underlying AI technology, known as deep learning, has proved very powerful at solving problems in recent years, and it has been widely deployed for tasks like image captioning, voice recognition, and language translation. There is now hope that the same techniques will be able to diagnose deadly diseases, make million-dollar trading decisions, and do countless other things to transform whole industries.

[...] The U.S. military is pouring billions into projects that will use machine learning to pilot vehicles and aircraft, identify targets, and help analysts sift through huge piles of intelligence data. Here more than anywhere else, even more than in medicine, there is little room for algorithmic mystery, and the Department of Defense has identified explainability as a key stumbling block.

[...] At some stage we may have to simply trust AI's judgement or do without using it. Likewise, that judgement will have to incorporate social intelligence. Just as society is built upon a contract of expected behaviour, we will need to design AI systems to respect and fit with our social norms. If we are to create robot tanks and other killing machines, it is important that their decision-making be consistent with our ethical judgements.

https://www.technologyreview.com/s/604087/the-dark-secret-at-the-heart-of-ai/

What do you think, would you trust such AI even if you couldn't parse its methods? Is deep learning AI technology inherently un-knowable?


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by VLM on Wednesday April 12 2017, @12:20PM (7 children)

    by VLM (445) Subscriber Badge on Wednesday April 12 2017, @12:20PM (#492703)

    It'll be interesting to see how the AI deals with aggressive or semi-aggressive drivers.

    My guess is to prevent accidents the AI will be programmed or trained to be extremely skittish, and human drivers or more advanced later AI will take advantage of the skittish AI, cutting it off, refusing to let it merge, etc.

    Starting Score:    1  point
    Karma-Bonus Modifier   +1  

    Total Score:   2  
  • (Score: 2) by Unixnut on Wednesday April 12 2017, @02:26PM (6 children)

    by Unixnut (5779) on Wednesday April 12 2017, @02:26PM (#492761)

    Seeing as the main people looking forward to driverless AI cars are those who hate driving and want that time "freed up" to watch movies, fiddle on their phone, sleep, or do whatever else they want, why would it matter if the AI was risk averse and took the "slow and steady" approach.

    So ok, you might end up at your destination later, but as you can still do what you want inside the car, all you have to do is leave a bit earlier, and you spend more time entertained in a moving pod rather than the fixed pod which is your home/office.

    Personally, If I were the type to use an AI car, I would want it to be risk averse, rather than aggressive. Then while I have a nap or whatever, the car is taking its time slow and steady to take me to my destination. Makes no difference to me if my journey was 20% longer or not really, as I am already doing what I would have done when I was home in the first place.

    • (Score: 2) by kaszz on Wednesday April 12 2017, @02:43PM (2 children)

      by kaszz (4211) on Wednesday April 12 2017, @02:43PM (#492779) Journal

      Or we could go for Personal Rapid Transport right away and that means the vehicle steering can make do with a simple microcontroller. You type in your destination and go snoozing etc.

      (PRT = driving on rails essentially)

      • (Score: 2) by Unixnut on Wednesday April 12 2017, @03:41PM (1 child)

        by Unixnut (5779) on Wednesday April 12 2017, @03:41PM (#492813)

        That would require a lot of laying of dedicated roads with sensors, set up and configured for the cars. The infrastructure cost would be massive, not to mention the maintenance cost of all the software, bug fixes, securing against hackers, the elements, etc...

        While in the end, you would be left with, well, trains. Which we already have, and the benefit that rails are cheaper and easier to lay, requiring less maintenance and the only way to "hack" it is by explosively damaging the rails, something that can't be done remotely (yet). And all this while also lasting longer than tarmac.

        • (Score: 0) by Anonymous Coward on Wednesday April 12 2017, @05:30PM

          by Anonymous Coward on Wednesday April 12 2017, @05:30PM (#492908)

          the only way to "hack" it is by explosively damaging the rails, something that can't be done remotely (yet).

          Wrong.

          First, rails also have switch points and signals, which these days are typically operated remotely and using computers; also the modern trains are computerized and networked. So it is not inconceivable to cause physical damage to a train remotely; for example I'm sure if you switch a switch point while a train is currently passing it, it will not go well for the train. Also, giving two trains green light to the same track from opposite directions will likely lead to disaster.

          Second, these days explosively damaging the rails can certainly be done remotely. All you need is a drone.

    • (Score: 2) by VLM on Wednesday April 12 2017, @04:13PM (2 children)

      by VLM (445) Subscriber Badge on Wednesday April 12 2017, @04:13PM (#492847)

      I was worried more about the AI being extremely skittish meaning it would be nervous and jerky making for a rough ride.

      Part of it is if you're obviously reading the newspaper or sleeping or eating breakfast people would cut the AI car off as a game to see if they can spill your coffee or wake you up or scare the hell out of you.

      Also its unwise to cut people off who are playing on their phones because you'll get hit, but a self driving car has you on 10 cameras at once so you know it'll slam on the brakes 100%. As a human driver people don't trust me to slam on the breaks 100% so I'm not going to get messed with. The point isn't duration of travel or even getting messed with, so much as its being in a car that slams the brakes on 100% and panic swerves and all that.

      Its not so much the time factor as its the being shook up like a race car driver factor.

      Its an interesting thing to think about, if 100% of cars are automated in 2070 or whatever, there will have to be some kind of dial setting of aggressiveness and there will be argument over the exact ideal thermostat setting. I would not be surprised to see it as some kind of local vote protocol, if everyone on this segment of interstate is willing to drive like a maniac to save 5 minutes, its off to the races...

      As far as I know, this is a fresh hard sci fi story plot line. If someone already wrote a story along these lines it would probably be interesting to read.

      • (Score: 2) by Unixnut on Wednesday April 12 2017, @04:32PM (1 child)

        by Unixnut (5779) on Wednesday April 12 2017, @04:32PM (#492864)

        Ah I see. Well the only reason the car would need to slam on the brakes and jerk around is if it didn't read the conditions of the road correctly. This includes the drivers of other vehicles.

        When I was taught to drive, I was taught that having to swerve to avoid obstacles or slam the brakes on was a failure of myself as a driver. Had I read intentions of other drivers and the road conditions correctly, I would have anticipated and made changes earlier on (adjusting speed, distance, position) so as to not be in that situation. They are road users as much as me after all.

        Yes, reading intentions does include things like "That driver is an idiot", "that driver should not have ever been granted a licence", "that driver is doing 30 on the motorway", and "that driver is an asshole that cuts up everyone without indicating" (helps that the last type usually drives an Audi or BMW, so you can infer personality type by car somewhat).

        I don't see why, when programming an AI, the designers/programmers would base their "AI personality" on anything other than what is considered "good driving". I would expect an AI to adopt driving courtesy, and do things like keep correct distance for speed, make way for merging traffic, and anticipate conditions and events and adjust for them to give its occupants the smoothest drive possible.

        > Its an interesting thing to think about, if 100% of cars are automated in 2070 or whatever, there will have to be some kind of dial setting of aggressiveness and there will be argument over the exact ideal thermostat setting. I would not be surprised to see it as some kind of local vote protocol, if everyone on this segment of interstate is willing to drive like a maniac to save 5 minutes, its off to the races...

        I doubt it. If all cars become automated, then why would you keep your own? Imagine it less like current and historic cars, and more like "pods with wheels". You would have an app on your brain implant or whatever, summon a "pod" that meets your requirements for amenities, space for occupants, distance to travel, trim level and cost, it will come pick you up and deliver you to your destination. You won't be able to set "aggressiveness" or much else that affects the driving.

        However I would imagine you would not care that much, after all, you are not driving, you are socialising with other occupants, watching something, playing a game, doing whatever you want to be doing really.

        And even if you wanted to own a pod, I suspect hacking them would be illegal (or at least really difficult and expensive). Not unlike how some modern cars had DRM to prevent modifications to its computers (at least until this practice was struck down).

        > As far as I know, this is a fresh hard sci fi story plot line. If someone already wrote a story along these lines it would probably be interesting to read.

        Reminds me of an Anime/Manga I used to watch when I was younger: https://en.wikipedia.org/wiki/%C3%89X-Driver [wikipedia.org] Not sure if it qualifies for your "hard sci fi" plot line, but it is something. People have made stories out of it, mostly dystopic in my experience (loss of control, total surveillance, government can track/control you where you go, "dissidents" and "undesirables" can be swiftly removed by virtue of "accidents" or "car malfunction" , etc...) .

        • (Score: 2) by VLM on Wednesday April 12 2017, @05:38PM

          by VLM (445) Subscriber Badge on Wednesday April 12 2017, @05:38PM (#492919)

          Automated or autonomous cars are often pitched with an abolishment of private property theme, sometimes the from each according to their ability and to each according to their need is stronger than the autonomous car message itself. I don't think its a useful or effective strategy. I mean, that kind of sales pitch never sold an automatic transmission, or a engine computer automatic choke (yeah, I'm old enough to have driven carb cars), or power windows. Why do you need the workers of the world to unite to enjoy this fine automatic electric hands free convertible top. I'm not signalling against autonomous cars or communism although I have in the past; I'm specifically countersignalling the method of selling "self driving cars AS communism".

          About this point I break out the analogy of I don't wear underwear at night (err when I'm asleep) so why not use a drone shipment so some other dude can share my briefs, I mean why would I want the unimaginable hassle of owning my own briefs when I could hire an expensive middleman to rent me temporary use of some shared briefs, kinda of like an underwear pimp but in trendy mobile phone app format. Speaking of pimps, my underwear is washed often enough to not be sentient or autonomous or automated (although that could be kinda cool...), but my wife is sentient and autonomous and I'm only enjoying her marital services for a couple minutes a day (whoo honey that was a fun 30 seconds) so naturally in our sharing gig economy I should be able use an app like wife-r where she could manage a redundant array of inexpensive husbands, aka a RAIH array. I mean, dude, wives are expensive, so some sort of timesharing threaded app is totally the way of the future. The UI is very modern, looks nice but not terribly effective. Yup I think we need to innovate legacy technology like marriage, get some apps in there, collect some rent as middlemen, yeah...

          A meme that is truly unappreciated today is the concept of the gig economy middleman as pimp. Or maybe the unappreciated meme is pimp as gig economy middleman. Either way I find it an annoying feature of depersonalized anti-socialized modern life that we rely on mobile phone apps to act as our pimps instead of fine upstanding human citizens such as seen in 1970s era blaxsplotitaion movies which used to fill UHF TV airwaves after 11pm or so on weekends in the good old days. Every time I hear about Uber I think its CEO should have a pink 70s caddy with furry steering wheel and 'fro haircut speaking the worst jive 70s hollywood could produce. That sounds like Uber to me.