Can you program intuition?

"davesaddiction @ opposite-lock.com" (davesaddiction) 08/31/2017 at 09:27 • Filed to: None

2

21

“...by building from a list of technical requirements, researchers neglect the single most important part of real-world driving: our intuition. Using it to determine the motivations of those around us is something humans are so effortlessly good at that it’s hard to even notice we’re doing it, nonetheless program for it.” “Why are computers so bad at this task of mind reading if it’s so easy for people? This circumstance comes up so often in AI development that it has a name: “Moravec’s Paradox.” The tasks that are easiest for people are often the ones that are the hardest for computers.” Full article:

!!! UNKNOWN CONTENT TYPE !!!

davesaddiction @ opposite-lock.com > davesaddiction @ opposite-lock.com 08/31/2017 at 09:27

0

If Kinja’d: https://qz.com/1064004/self-driving-cars-still-cant-mimic-the-most-natural-human-behavior/

MrDakka > davesaddiction @ opposite-lock.com 08/31/2017 at 09:35

2

You don’t need AI to make self driving cars. Just stick one one of these in there and voila! Self driving car.

Phyrxes once again has a wagon! > davesaddiction @ opposite-lock.com 08/31/2017 at 09:35

1

This is why the idea of some self driving cars, some semi-autonomous cars, and the rest being driver piloted cars all sharing the road scares me. Sure sensor feedback and machine learning are great but the arbitrary decision making made by some drivers scares me. Until the AI can predict which driver is going to abruptly take a right exit from the center lane without signaling I’d rather be paying attention myself and have the computer back me up if it reacts faster than I do.

crowmolly > Phyrxes once again has a wagon! 08/31/2017 at 09:41

0

We cannot do this with rooms full of high powered computers. What makes people think we can do it with a car?

crowmolly > Phyrxes once again has a wagon! 08/31/2017 at 09:41

0

We cannot do this with rooms full of high powered computers. What makes people think we can do it with a car?

davesaddiction @ opposite-lock.com > Phyrxes once again has a wagon! 08/31/2017 at 09:42

0

Yeah, you’re probably right. I’d say it would be okay if some of us drive and some are self-driven as long as all car have the ability to communicate with each other.

davesaddiction @ opposite-lock.com > MrDakka 08/31/2017 at 09:42

0

Perfect.

Phyrxes once again has a wagon! > crowmolly 08/31/2017 at 09:47

0

This is true, but it won’t stop companies from pushing the technology and people making terrible decisions because they don’t understand the limitations of the technology that their car is equipped with.

Urambo Tauro > davesaddiction @ opposite-lock.com 08/31/2017 at 09:47

1

Yeah, I think we pick up on a lot more clues than we realize when we’re driving. To be better than humans, HAVs are going to have to read, identify, and prioritize things that we hardly ever think about.

Like when lane markings disappear, we naturally continue on the last known trajectory and find something to guide us, like a seam in the road, or we aim for markings several feet ahead. Or we’ll judge the width of the road and compare it to the number of lanes there ought to be. After all, that’s how we handle unpaved roads- we divide the road in half, and drive on our side of that invisible line. HAVs need to be able to do this, without relying on paint that hasn’t worn away yet.

ttyymmnn > davesaddiction @ opposite-lock.com 08/31/2017 at 09:50

1

Phryxes hits the mark. The only way autonomous cars will work is if you swap out the regular cars with robot cars all at once. I think the combination of the two is a recipe for chaos. Who’s to say that a robot car won’t make an arbitrary maneuver that would have been eminently predictable in a human driver? Humans have the ability to predict the unexpected because we have millions of years of experience watching assholes do idiotic things. You can’t program that into a computer. The best a computer can do is be reactive while humans are capable of being proactive.

functionoverfashion > davesaddiction @ opposite-lock.com 08/31/2017 at 09:57

1

I read a good book on this topic a few years ago, however, it’s probably outdated in that it was before autonomous cars were really on the horizon, let alone on the roads. But anyway it was called On Intelligence and it talked a lot about tasks that humans master but AI cannot.

One example is finding a bathroom in a restaurant: we have learned that, generally, restrooms tend to be in the back of a restaurant, often hidden in a corridor or something. But humans are pretty quick to head to the back when looking for a restroom, and can easily distinguish between the kitchen, a closet, and the bathroom. Granted, a robot doesn’t need a restroom, but you get the idea. A computer would have a hard time “knowing” where the back of the restaurant is, let alone sensing the direction where one is likely to be. Then, a lot of places make cutesy-themed signs to indicate the bathroom, and again, a computer might have a hard time distinguishing these.

Basically the author talks about how humans’ ability to apply a huge variety of contextual clues to any situation, often without realizing it, is probably the biggest hurdle that AI has to overcome.

Anyway, check it out if you have a chance.

nerd_racing > MrDakka 08/31/2017 at 10:04

1

why stop there?

davesaddiction @ opposite-lock.com > functionoverfashion 08/31/2017 at 10:08

0

Sounds good - thanks for the tip!

nermal > davesaddiction @ opposite-lock.com 08/31/2017 at 10:11

1

Thinking at the high level, that would mean that the AI programmers also need to add the ability to predict the moves of other drivers.

For example, moving along a 3 lane highway with a car in front, doing the speed limit. What are all of the scenarios that could happen? Some poorly secured cargo could fall out. A woodchuck could run across the road causing them to brake / swerve suddenly. A woodchuck could run across the road and they could hit it, causing the woodchuck to become a projectile. They could be text messaging / drunk / tired and move erratically. They could cut across a lane or two at the last minute with no blinker to barely make an exit.

The human brain is a very complex thing, and one of the biggest advantages it has over programmed AI is learning - for now at least. That said, the AI will in theory be able to see more than a human would, while also acting more predictably, and without the biggest human flaws (distraction and panic).

davesaddiction @ opposite-lock.com > ttyymmnn 08/31/2017 at 10:11

0

As I responded to him, one way to gain parity is that all the cars could be communicating with each other every millisecond. While that’s still reactive, other cars would know immediately when a motion is taken (or is starting to happen - the steering wheel moves a split second before the wheels move), and would have to sense it first before reacting.

davesaddiction @ opposite-lock.com > Urambo Tauro 08/31/2017 at 10:13

1

Can be programmed for, but has to be programmed for. Every single possible scenario that a car could ever experience has to be programmed for... A monumental task.

davesaddiction @ opposite-lock.com > nermal 08/31/2017 at 10:20

1

It’s an interesting challenge, with some of the smartest people on the planet working on it. Will be interesting to see how long it takes for them to solve it.

davesaddiction @ opposite-lock.com > crowmolly 08/31/2017 at 10:23

0

When Deep Blue played chess with Kasparov, was it thinking he was thinking it was thinking what he was thinking?

Urambo Tauro > davesaddiction @ opposite-lock.com 08/31/2017 at 10:24

1

That’s what makes me skeptical that they’ll ever build something that’s good enough. But it’s amusing to watch them try, and impressive to see how far they’ve come.

davesaddiction @ opposite-lock.com > Urambo Tauro 08/31/2017 at 10:27

1

Car-to-car communication will be the linchpin.

MrDakka > nerd_racing 08/31/2017 at 23:30

0

Would rather be a brain in a jar in a car than in a dalek chassis.