Originally Posted by Philphine (Post 19043829)

I'm waiting for the riderless skateboard. it's gotta be able to do the gnarliest park tricks and make me look good though.

Originally Posted by Philphine (Post 19059618)

I wonder if in the future you'll see threads on driverless car forums complaining about riderless bicycles.

Originally Posted by dabigboy (Post 19065654)

On a serious note, I firmly believe that driveless cars will *never* be truly workable in the current real-world environment until/unless artificial intelligence makes some big breakthroughs. Philphine brings up a great point about the driverless car figuring out if a motorcycle or other smaller object is occupying a lane. Without really awesome (and not-yet-attained) AI, any attempt to handle these sorts of situations individually are just limited-scenario hacks.

Matt

Originally Posted by dabigboy (Post 19065654)

On a serious note, I firmly believe that driveless cars will *never* be truly workable in the current real-world environment until/unless artificial intelligence makes some big breakthroughs. Philphine brings up a great point about the driverless car figuring out if a motorcycle or other smaller object is occupying a lane. Without really awesome (and not-yet-attained) AI, any attempt to handle these sorts of situations individually are just limited-scenario hacks.

Matt

Originally Posted by ThermionicScott (Post 19067468)

It's not a great point, it's a stupid point. The cameras and software on driverless cars are much better able to see and keep track of the traffic around them than we humans are.

https://www.youtube.com/watch?v=tiwVMrTLUWg

Originally Posted by Imnotchinese (Post 19134995)

Last year, AI made a huge leap forward, by passing the "three wise men" test, in which three robots programmed with the same AI, but two of them were muted. Then each one was asked who could speak. Two of them said nothing, and one said "I do not have enough information to answer that question" and then corrected itself by saying "wait, actually I believe I have a voice" this is a huge step forward, as it marks the first time a robot has officially acknowledged it's own existence.

Originally Posted by dabigboy (Post 19156876)

Come on ThermionicScott, tell me how you really feel about my opinion. :)

Originally Posted by TheLibrarian (Post 19157767)

As another article said about Hotz and Tesla's automated driving to make it 99% accurate is easy to make it 99.9999% accurate is much harder and these systems need to be that accurate. Also cars just stopping even 1% of the time would clog up every highway. There's an easy 200 cars going by at any time and if 2 are always stopped thats problems. But you're just pulling these numbers out of the air, don't think you mean 1% is ok. Further cars stopping in the middle of the road is not safe. If the error is on the person we can blame thats one thing but if the error is on the product that's endless litigation.

For riderless bikes not sure what the new thing is but certainly seems unnecesary and would have all the same problems as cars. As for the statistics driverless cars have not been out in any significant usage to determine if they are safer or not. Controlled studies with engineers babysitting them aren't representative.

Originally Posted by dabigboy (Post 19156876)

Come on ThermionicScott, tell me how you really feel about my opinion. :)



That's certainly a very interesting response, but from a programmer's perspective, its significance is hard to quantify without knowing HOW it arrived at that answer.

Hope springs eternal, but I can't imagine an AI ever eclipsing the reasoning and decision-making ability of a human. The thing about humans is that we have something an AI can never have: imagination. It follows that we therefore also have insight. Because of this, no matter what situation arises, no matter how extreme the circumstance, no matter how unplanned the event, a human always has the capacity to evaluate the situation and develop some sort of rational response. It may not always be the best response, and it may not even be the correct response, but there will be a plan of action that takes the immediate factors of the situation into account.

I'm sure anyone who's ever spent any time with a computer or other electronic gadget (particularly if you've had to support, fix, or develop software) can relate to the concept of "babysitting" the software. You may have something that works very well, but every now and then there is that unforeseen scenario in which the software loses its "mind" and does something completely unpredicted or silly. Yet, a human might look at that scenario and know immediately what to do (or, at the very least, know that what the software is doing is NOT a good response).

The aviation industry has been into this field for a long time. Airplanes with the right equipment can practically fly themselves. But, we really can't trust the systems entirely. This point is driven into pilots throughout their training, and is reinforced through embarrassing incidents and tragic accidents. There are numerous instances in which an autopilot or flight control system will "throw its hands up" and simply quit flying the airplane (Air France 447, for instance), at which point the pilot (who has hopefully stayed in the loop, but often is not) must assume control. There are also other instances in which the computer does something strange and dangerous, something a human pilot would never do (though in most such cases, the computer will just give up and disengage, thanks to fail-safes built into the software).

Now, compare this to cars: quick, appropriate responses are FAR less important in an airplane, where a pilot (or computer) has the luxury of wandering around the sky a few thousand feet here or there, heading the wrong direction for a bit, not properly controlling airspeed, etc. That's a lot different from the world of driving, where we are regularly mere inches away from another car coming the opposite direction at a differential speed of 130mph+.

That's not to say I don't see driverless cars making a big impact in the future. If our road system is modified a bit and made more uniform, if the car is only expected to go on relatively improved roads, if GPS is always available and/or some sort of computer-friendly road marker system is devised, if extremely twisty or hilly roads can be properly dealt with by the computer, if specific types of inclement weather are avoided, and IF the driveless cars are kept in good repair (and we know some won't be), then they may be reasonably workable.

And, who knows, maybe there will be a paradigm shift in AI and we'll see something that completely bypasses the current AI pitfalls, and can truly challenge the human mind's ability to reason. I'm happy to be proven wrong. :)

Matt

Originally Posted by dabigboy (Post 19158628)

ThermionicScott, I wouldn't say I have a huge amount of faith in the ability of humans to make good decisions on the road, I just have faith in the tendency of them to make reasonably predictable decisions (vs completely wild and counter-intuitive decisions). Plus, while humans may not always make the best decisions, by taking on the responsibility for the drive into your own hands, your level of risk is more closely tied to your own personal decisions. And I think that is always a good thing. I like known risk: my own skills, condition, and decision-making ability. I don't like unknown risk: the minds of the programmers who designed the car's AI to foresee or somehow allow for every possible circumstance that could ever arise, which of course is impossible.

I'm not really disagreeing with you or trying to invalidate your point of view, I'm just trying to clarify my approach to the whole idea. We differ in where we want to allocate risk.

Imnotchinese, you're hitting on a very interesting paradox of computers vs the human mind. There are certain procedural, logical operations that computers are far superior at. For instance, if an AI-controlled car has information on all nearby vehicles, a sufficiently powerful computer can, in a matter of milliseconds, simulate a number of potential outcomes depending on different courses of action, and choose the one that would seem to have the lowest risk based on pre-programmed conditions and the laws of physics. Furthermore, if the computer is able to accurately measure things like distance, relative speed, direction, etc, it can do very accurate calculations for things like required stopping distance and likely path of travel. But the human mind can do other things in near-instant time, things that continue to baffle our understanding (such as highly abstract associations, inference, etc). Simple example: two people who know each other very well are talking. One says something rather vague and poorly worded, like "it's going to be worth it", when such a statement has no bearing on the current conversation. But the other person will, in their mind, instantly go to something they know the first person is working at, or a goal the two friends were recently trying to accomplish, and make a pretty accurate guess that that's the thing the first person was talking about. Or it may be as simple as a comment that goes back to a conversation the two had some minutes ago. Oftentimes, the mind is somehow able to pick up that comment and associate it almost instantly with one of perhaps many recent conversations.

These truly disparate strengths of humans vs computers are why, while I'm very skeptical of an AI-controlled car, I think a computer-aided car (as we are already seeing) has huge potential. Imagine having graphical vectors of other traffic's movement projected onto your windshield, or markers showing where you'll likely stop if you apply maximum braking!

Ah, this is interesting stuff. :) I've been interested in AI and software failure modes for some time. My career is in computers, and I've been programming for fun since 2002 or so. Being "passionately curious" is a good thing, I think it's the most important component of learning.

Matt