Don't Believe the AI Hype!

[295566]

They don't need to be able to predict anything. That's not how any of this works. A computer doesn't need to operate the way a human brain works. We use prediction to overcome our inherent slowness. A computer gets to wait until you do something, measure it, find the proper course of action to counter it, then execute the action. all in .00000001 seconds. Humans take at least 8 orders of magnitude longer to do the same, so we cheat and guess what someone is *about* to do.

OK, so in that same situation, would AI be able to see the stressed out driver, notice it, and speed up/slow down in order to distance themselves from the potential danger? As a rational human might be able to predict potential danger? Or will it not recognize that as a threat until that .0000001 of a second after something actually does happen?

That, I think, is the point. Sometimes waiting until after things happen it's already too late.

 

[scotinexcile]

Depends on your definition of AI, right now and probably not before 2050 will what is right now Automated Intelligence change to Artificial intelligence

NONE of these systems has yet passed the Turing Test

 

[Marcoose]

BTW, just how is the AI in those cars going to share data with the motorcycle?

No less than the same way done in the last 80 years; indicators, brake lights, horn, headlamps. What other data do you require to use the public roads today that you don't already have?

It is just like lane sharing used to be. If there is no law or regulation addressing it, it is legal.

Source?

 

[295566]

No less than the same way done in the last 80 years; indicators, brake lights, horn, headlamps. What other data do you require to use the public roads today that you don't already have?

That's not what he means. I think, correct me if I'm wrong, he's saying that the AI controlled cars will be able to talk to themselves. "I'm getting off on this exit," then merges over 5 lanes seamlessly, as cars speed up/slow down to let the car merge over and exit. Or, "Object in lane .75 miles ahead," and other vehicles begin slowing, merging over, etc. Of course all of this could happen with human drivers in a perfect world, but we live in a society of "me first," where aggressive drivers will block others from merging, road rage exists, etc.


As per your other point... if something isn't explicitly illegal, by definition, it's legal.

 

[Marcoose]

...he's saying that the AI controlled cars will be able to talk to themselves. "I'm getting off on this exit," then merges over 5 lanes seamlessly, as cars speed up/slow down to let the car merge over and exit. Or, "Object in lane .75 miles ahead," and other vehicles begin slowing, merging over, etc.

Are you saying other road users may speed up, slow, merger, turn, sway, cross, etc, without your knowledge? That sounds just like today's traffic. Where's the outcry, the hyperbole?

As per your other point... if something isn't explicitly illegal, by definition, it's legal.

I'm pretty sure there are plenty of 'laws' saying a motor vehicle must be tested, licensed and insured, and operated by a licensed, healthy and sober driver. And I imagine there are 'laws' saying such driver must keep at least one hand on the wheel at all times. I can't imagine the AV companies found a loophole and are getting away with it. I imagine they have a special permit for operating the vehicle outside of the 'law'. But I truly don't know. Hence the question, source, please?

 

[Nkeane]

OK, so in that same situation, would AI be able to see the stressed out driver, notice it, and speed up/slow down in order to distance themselves from the potential danger? As a rational human might be able to predict potential danger? Or will it not recognize that as a threat until that .0000001 of a second after something actually does happen?

That, I think, is the point. Sometimes waiting until after things happen it's already too late.

You are missing the point. prediction is unecessary when a corrective course can be made serveral thousand times faster than the offending party.

no one is saying that AI will end traffic accidents. If you need it to be infallable for it to be allowed, then why wouldn't you also have that expectation of a human? self driving cars only need to be as good as humans. Any other gains beyond that, is all bonus.

 

[295566]

You are missing the point. prediction is unecessary when a corrective course can be made serveral thousand times faster than the offending party.

no one is saying that AI will end traffic accidents. If you need it to be infallable for it to be allowed, then why wouldn't you also have that expectation of a human? self driving cars only need to be as good as humans. Any other gains beyond that, is all bonus.

I think you're missing the point, actually.

How they will react to other AI "drivers" is easy to guess. The greater question, however, is how they will react to human drivers, motorcycle riders, bicyclists, pedestrians, etc.

Going back to the example given above, about the stressed driver, a human operator may notice the distracted nature, or erratic actions, of the stressed individual. Taking that into account, an observant and cautious operator would likely choose to put distance between themselves and the stressed individual, eliminating the chance that they get swerved into. Now, if an AI doesn't perceive their actions as immediately dangerous, it may choose to stay right alongside the stressed individual. When that person swerves toward the AI, what is it going to do if there's a car ahead, behind, and a barrier to the left? There's nowhere to go. Reaction times thousands of times faster than a human won't help if there's nowhere to go and nothing to do.

Does that make sense? Of course this is all hypothetical, but just one instance in where a human can make decisions based on partial input, something AI cannot do, it is one of the limitations of computing as others have pointed out.

I'm not against AI tech; rather, I think it has the potential to make driving much safer. However, I know the tech is still years, possibly decades, from being ready for the majority of consumers. My initial post in this thread is in amazement that companies are able to alpha test this tech on public roads, completely unregulated.

 

[295566]

I'm pretty sure there are plenty of 'laws' saying a motor vehicle must be tested, licensed and insured, and operated by a licensed, healthy and sober driver. And I imagine there are 'laws' saying such driver must keep at least one hand on the wheel at all times. I can't imagine the AV companies found a loophole and are getting away with it. I imagine they have a special permit for operating the vehicle outside of the 'law'. But I truly don't know. Hence the question, source, please?

Laws and regulations, for the most part, aren't written "allowing" things, but rather forbidding them. For example, CVC 21352(a), the DUI section, doesn't say that a vehicle has to be operated by a sober driver, but rather that it's unlawful for someone under the influence of alcohol to operate a motor vehicle.

This is why there was so much dispute/confusion for many years regarding lane splitting in CA... there doesn't exist a law forbidding it, so therefore it's legal, by definition. It's no loophole, it's logic.

Interestingly, and perhaps the only "loophole" that may be exploited is CVC 12500(a), license requirement sections, which states:

12500. (a) A person may not drive a motor vehicle upon a highway, unless the person then holds a valid driver's license issued under this code, except those persons who are expressly exempted under this code.

Since, technically, the vehicle isn't being driven by a person, the AI wouldn't need a license. I can see this being disputed, potentially, and case law/new regulations being created.

 

[ZCrow]

You are missing the point. prediction is unecessary when a corrective course can be made serveral thousand times faster than the offending party.

no one is saying that AI will end traffic accidents. If you need it to be infallable for it to be allowed, then why wouldn't you also have that expectation of a human? self driving cars only need to be as good as humans. Any other gains beyond that, is all bonus.

I think your summary of AI works is flawed. Prediction exactly the right word to use because the current AI models I have read about work on deep learning which requires large sets of big data (or the correlation for us would be experience). With that large data set based on practical application, that fast reaction time means absolutely nothing. If this were just raw processing power we would still be a long way off from having AI, if ever. It's fast computing along with large sets of data and being able to analyze that data quicker than we can to see patterns we are not consciously aware of.

 

[Marcoose]

Interestingly, and perhaps the only "loophole" that may be exploited is CVC 12500(a), license requirement sections, which states:

[insert]

Since, technically, the vehicle isn't being driven by a person, the AI wouldn't need a license. I can see this being disputed, potentially, and case law/new regulations being created.

Good find. :thumbup

I'm still awfully sceptic that AV companies are testing 'unregulated'. They're bound to be under some regulation. Especially in nanny state California.

 

[Nkeane]

I think you're missing the point, actually.

How they will react to other AI "drivers" is easy to guess. The greater question, however, is how they will react to human drivers, motorcycle riders, bicyclists, pedestrians, etc.

Going back to the example given above, about the stressed driver, a human operator may notice the distracted nature, or erratic actions, of the stressed individual. Taking that into account, an observant and cautious operator would likely choose to put distance between themselves and the stressed individual, eliminating the chance that they get swerved into. Now, if an AI doesn't perceive their actions as immediately dangerous, it may choose to stay right alongside the stressed individual. When that person swerves toward the AI, what is it going to do if there's a car ahead, behind, and a barrier to the left? There's nowhere to go. Reaction times thousands of times faster than a human won't help if there's nowhere to go and nothing to do.

Does that make sense? Of course this is all hypothetical, but just one instance in where a human can make decisions based on partial input, something AI cannot do, it is one of the limitations of computing as others have pointed out.

I'm not against AI tech; rather, I think it has the potential to make driving much safer. However, I know the tech is still years, possibly decades, from being ready for the majority of consumers. My initial post in this thread is in amazement that companies are able to alpha test this tech on public roads, completely unregulated.

your very specific situation is an outlier. yes, there will be a collision. no one says that AI will stop 100% of collisions. If you want to say that becuase of this, that humans>robots at driving, I would say you are being puroposefully obtuse.

 

[295566]

your very specific situation is an outlier. yes, there will be a collision. no one says that AI will stop 100% of collisions. If you want to say that becuase of this, that humans>robots at driving, I would say you are being puroposefully obtuse.

You focused on one point of my whole post and ignored the rest, then start calling names. :thumbdown

 

[ontherearwheel]

First time in a long time, I rode Mines rd from Livermore to the Junction and back. I forgot all about the one lane parts of the road.

Going toward the Junction, I came up on two semi dumps going the same way, in the one lane. I got around them, went to the Junction, turn around and headed back.

On the return, I kept the fact that I had passed them in my mind and that I would be heading toward them. I rode around blind turns far more cautious than normal and far more to the shoulder of the road.

Also, since it is twisty, you can "see" around corners...kinda. There are times you can see what's on the road heading into a blind turn in the opposite direction before you get to the blind turn.

First.....would AI remembered those trucks on the return trip?

Second, would AI anticipate the vehicle in the second example as a potential threat?

 

[295566]

First.....would AI remembered those trucks on the return trip?

Second, would AI anticipate the vehicle in the second example as a potential threat?

I think that's another version of the same question I posed. Would AI be able to predict future outcomes with only limited information? All current evidence points to no, as it seems to operate based on the facts presented before it immediately. And, in some instances, even if reaction times are hundredths of a second, sometimes it may be too late.

Again, I think AI has the potential to greatly reduce traffic collisions and make the roads overall safer... but there are still challenges to overcome, and, IMO, the tech is still a long ways out from being publicly "ready."

 

[NeilInPacifica]

I think your summary of AI works is flawed. Prediction exactly the right word to use because the current AI models I have read about work on deep learning which requires large sets of big data (or the correlation for us would be experience). With that large data set based on practical application, that fast reaction time means absolutely nothing. If this were just raw processing power we would still be a long way off from having AI, if ever. It's fast computing along with large sets of data and being able to analyze that data quicker than we can to see patterns we are not consciously aware of.

In the long run the AI models will behave the same way a good driver does, it will recognize danger and avoid it because avoiding accidents will be one of its goals. It will have both predictive powers and fast reaction times; in the long run far superior to a human driver and with the added bonus of an intelligent mesh across all the vehicles on the highway, perhaps including bicycles and pedestrians. Eventually society will ban human drivers because they will just slow things down. Might be a while before all this comes about, but it will eventually.

 

[Schnellbandit]

I think saying computers will be predicting what humans will or won't do is quite a ways off. Lets say a system had the sum of all data available. Just how does the system predict that little jonny's runny nose is about to cause him to blow a snot clod all over his sister and that mom is going to turn around to quickly wipe his nose and in doing so, wipe you out? The computer can react but predicting what a person is about to do? Way way off.

When I ride, I'm pretty sure I'm doin what other riders to, pay some attention to what the driver is doing in addition to what the car they are driving is doing. I can see if the driver is looking around, fumbling with their phone, just ducked down to grab that something that probably fell out from between their legs etc and I adjust accordingly. The computer can't do that, it can only react after that driver, in a fit of frustration moves over.

As a rider, seeing the car just ahead of you with a driver doing who knows what inside, don't you get away from them or wait until they do the stupid thing? The AI car is just going to sit there, unaware until the stupid happens. Granted, it can react very quickly but predict anything beyond what data shows, not a chance.

I think perhaps a lot of possibilities are being attributed to AI (which really isn't AI) as actual capabilities or within reach over the next even 10 years.

As a dedicated transportation system, AI controlled cars would be a miracle or time saving and safety but the approach isn't like that, it's a slow mix that is being mistaken for giving development time to work things out.

For car drivers, AI seems like a great thing but we're talking about motorcycles here and knowing how they are ridden, AI cars seem like a vinegar and oil affair. In the beginning everything seems okay and then wham, we realize they were never meant to exist on the same roads. I sure hope that isn't the case but when it comes to motorcycles, the not working out part means us meeting pavement, not a sideswipe or a fender bender.

 

[fubar929]

I think saying computers will be predicting what humans will or won't do is quite a ways off. Lets say a system had the sum of all data available. Just how does the system predict that little jonny's runny nose is about to cause him to blow a snot clod all over his sister and that mom is going to turn around to quickly wipe his nose and in doing so, wipe you out? The computer can react but predicting what a person is about to do? Way way off.

Don't be so sure... Recognizing license plates is easy and you can be sure that all of the AI-controlled vehicles from the same company share the same "database". Combine that with the ability to precisely determine the location of another vehicle within its lane and it shouldn't be too difficult to determine whether a particular driving pattern is outside what is considered "normal" for for that vehicle at any particular time of day on any given road. Once the AI driver notices the vehicle moving out of it's normal lane "window" it can begin to react, likely well before a human driver would.

It'd also likely that an AI driver will be able to execute driving strategies that no human driver could. As a human driver, if someone swerves into your lane you need to make a fairly instantaneous decision about whether to brake, swerve, or (for advanced drivers) brake and swerve. An AI driver can be much more accurate when calculating trajectories, acceleration, and distances and it may also know more about the capabilities and driving patterns of surrounding traffic. So when another car attempts to change lanes into it, maybe the AI driver knows that the following car is equipped with collision-avoidance braking, so instantaneous threshold braking is the right escape route. Or contact is inevitable maybe it swerves toward the car that's changing lane, makes contact before it can build up too much lateral speed, and gently "bumps" it back into its own lane? Contact has occurred, but the AI driver has minimized damage to both vehicles. Maybe the AI notices the initial drift toward it's lane so it honks the horn and swerves aggressively toward the distracted human driver to get their attention before they even leave their lane? Finally, let's not forget that an AI driver can probably predict a collision sooner than a human could and so could potentially protect its occupants better than a normal car (ex: by pre-tensioning seat belts or inflating airbags before the actual impact).

Obviously, we're not there yet but I'd guess that a lot of these things are much closer than anyone thinks. That said, I'm sure that not all self-dirivng car companies are giving motorcyclists the attention they require. The Waymo cars I see driving around Mountain View seem to be especially cautious of bicyclists and pedestrians, which is a hopeful sign, but I haven't witnessed enough motorcycle encounters to know how they're doing there. Everyone else is rushing to catch up with Waymo, so might be willing to cut corners in an attempt to get to parity

 

[norcalkid]

In the long run the AI models will behave the same way a good driver does, it will recognize danger and avoid it because avoiding accidents will be one of its goals. It will have both predictive powers and fast reaction times; in the long run far superior to a human driver and with the added bonus of an intelligent mesh across all the vehicles on the highway, perhaps including bicycles and pedestrians. Eventually society will ban human drivers because they will just slow things down. Might be a while before all this comes about, but it will eventually.

I agree with this. And I don't think the mass give a shit if motorcycles on public roads go the way of the Doto.

 

[Schnellbandit]

Somehow, I don't see some mesh network actually working unless there is some standard of valuing the data and how it is interpreted. These are all proprietary systems, which one, Apple, Google, Ford...is going to give up their way to go one way that isn't theirs? There is way too much money involved and if we think for the betterment of drivers, those companies are all going to give up anything, I ask, if they are unwilling to stop people from using phones in cars, something they could do right now, what makes us think they will act in the name of safety with AI? People are being killed every day because of texting and phone use while driving and yet the means exist to stop it, right now.

I have yet to hear anyone describe or explain in any detail exactly what they mean when they say AI will predict anything related to human behavior. Any action on the part of AI after the human makes a decision is reaction not prediction.

While AI might be able to predict the results of large scale patterns in traffic movement, someone explain how any AI system could possibly predict even the simpliest human behavior when that human driver sees a billboard that says milk is on sale and they just turn the wheel to make a u-turn over a double yellow? The AI can react to a u--turn but it can't begin to predict who will read the sign nor who will do something because they read it.

If we're talking long after we are all dead, maybe AI will realize the prediction claims but then think about what that really means, the control of people through AI.

 

[Smash Allen]

The computers have eyes, Jerry :rolleyes