4 ms·

The problem though is that humans can adapt, and the AI in general can't. A go AI can play Go perfectly, but if a player overturns a board, it can do nothing to

by Noos 10y ago

The problem though is that humans can adapt, and the AI in general can't. A go AI can play Go perfectly, but if a player overturns a board, it can do nothing to right this situation and resume playing. It's just following a list made for it, albeit a fairly substantial list. The worry is this, the cars will not fail like a human will. It may even be dangerous The car may not even be able to move in certain situations (like another car blocking them in front either purposefully or accidentally). Or it may not be able to even see certain conditions a human could.

iandanforth 10y ago
This. When these systems fail they will fail in surprising and inhuman ways. The accidents will appear to average people as easily avoidable which will give the impression of low quality engineering. e.g. Who doesn't slow down when a semi crosses their path on the highway?
- maverick_iceman 10y ago
  >they will fail in surprising and inhuman ways It's an improvement as long as those are few percent of the unsurprising and human ways human drivers fail.
- castratikron 10y ago
  Here's a good example. Self driving car side swipes a bus at ~25 mph. You can see the driver of the bus throw his hands up in the air in disbelief: https://www.youtube.com/watch?v=I9T6LkNm-5w https://www.youtube.com/watch?v=I9T6LkNm-5w From a human's perspective it's a stupid mistake. It seems stupid because computers think differently than people do. On the other hand a computer would probably think a human were stupid if it saw a human make a mistake when calculating the logarithms of every number between one and a million.
jdietrich 10y ago
>It's just following a list made for it, albeit a fairly substantial list. You're operating on a completely archaic understanding of AI. Modern systems can and do adapt to situations on-the-fly. What's more, they can adapt on aggregate - the experiences gathered by a single car can be shared to all cars. Very quickly, every car in the fleet will have billions of hours of cumulative driving experience. The overwhelming majority of motor vehicle accidents aren't weird and unpredictable edge cases. They're tragic but mundane events that result from a handful of root causes - inattention, excess speed, poor judgement and unnecessary risk-taking. "Driver/rider failed to look properly" is the key contributory factor in nearly half of road traffic accidents. Computers utterly dominate humans in this respect. For every accident caused by some bizarre and unpredictable set of circumstances, there are thousands caused by someone doing something obviously stupid. A computer can be programmed to be ultra-cautious in difficult situations. A computer can maintain 100% vigilance 100% of the time. Humans can't. Self-driving cars will undoubtedly fail in new and unexpected ways, but it's abundantly clear that they'll fail much less often than humans.
- projectorlochsa 10y ago
  The main approach of how AlphaGo went around the issue of never seeing the positions and playing good moves is search (monte carlo tree search). In the same way, if the world is modelled accurately and succintly, can a car predict the various scenarios and adapt, even if the current sitation was not seen before (extremely high speeds or unusual number of cars, etc.)
- Noos 10y ago
  I don't think so. I think when you introduce computers to anything, you add more edge cases, not less. They are just different ones though. And I'm also skeptical about the quality of this software once it hits an actual mass-market. People here are assuming high levels of rigor when for all we know, it could be similar to Android in the long term.
  - jdietrich 10y ago
    >I don't think so. I think when you introduce computers to anything, you add more edge cases, not less. That's sort of my point. Accidents are rarely caused by edge cases, but by predictable human failures. All the information needed to avoid the typical accident was plainly available, but the driver just didn't see it, process it or react to it appropriately. Driving is generally very predictable, but human attention and perception is hugely fallible. An AI that gets the basics consistently right but occasionally freaks out in an unpredictable situation would be a huge improvement over human drivers.
  - path411 10y ago
    Cars are already full of software.
- foota 10y ago
  One potential flaw here is that our current driving experience is based on humans. Perhaps humans are relatively good at handling unexpected situations, meaning they don't cause many accidents. Isn't it possible that by switching to AI we shift from mundane accidents to unexpected situation accidents?
maverick_iceman 10y ago
This is not true. DeepMind has built AI that can understand the rules of the game just by looking at the screen and then play it perfectly.
- Noos 10y ago
  When I was a kid, we used to have to replace chess pieces we lost while playing during lunch break with ersatz ones...pennies, nickels, etc. I'd like to see Deep Mind try to tell a game from that.

by Noos 10y ago

iandanforth 10y ago

This. When these systems fail they will fail in surprising and inhuman ways. The accidents will appear to average people as easily avoidable which will give the impression of low quality engineering. e.g. Who doesn't slow down when a semi crosses their path on the highway?

maverick_iceman 10y ago
>they will fail in surprising and inhuman ways It's an improvement as long as those are few percent of the unsurprising and human ways human drivers fail.
castratikron 10y ago
Here's a good example. Self driving car side swipes a bus at ~25 mph. You can see the driver of the bus throw his hands up in the air in disbelief: https://www.youtube.com/watch?v=I9T6LkNm-5w https://www.youtube.com/watch?v=I9T6LkNm-5w From a human's perspective it's a stupid mistake. It seems stupid because computers think differently than people do. On the other hand a computer would probably think a human were stupid if it saw a human make a mistake when calculating the logarithms of every number between one and a million.

jdietrich 10y ago

>It's just following a list made for it, albeit a fairly substantial list. You're operating on a completely archaic understanding of AI. Modern systems can and do adapt to situations on-the-fly. What's more, they can adapt on aggregate - the experiences gathered by a single car can be shared to all cars. Very quickly, every car in the fleet will have billions of hours of cumulative driving experience. The overwhelming majority of motor vehicle accidents aren't weird and unpredictable edge cases. They're tragic but mundane events that result from a handful of root causes - inattention, excess speed, poor judgement and unnecessary risk-taking. "Driver/rider failed to look properly" is the key contributory factor in nearly half of road traffic accidents. Computers utterly dominate humans in this respect. For every accident caused by some bizarre and unpredictable set of circumstances, there are thousands caused by someone doing something obviously stupid. A computer can be programmed to be ultra-cautious in difficult situations. A computer can maintain 100% vigilance 100% of the time. Humans can't. Self-driving cars will undoubtedly fail in new and unexpected ways, but it's abundantly clear that they'll fail much less often than humans.

projectorlochsa 10y ago
The main approach of how AlphaGo went around the issue of never seeing the positions and playing good moves is search (monte carlo tree search). In the same way, if the world is modelled accurately and succintly, can a car predict the various scenarios and adapt, even if the current sitation was not seen before (extremely high speeds or unusual number of cars, etc.)
Noos 10y ago
I don't think so. I think when you introduce computers to anything, you add more edge cases, not less. They are just different ones though. And I'm also skeptical about the quality of this software once it hits an actual mass-market. People here are assuming high levels of rigor when for all we know, it could be similar to Android in the long term.
- jdietrich 10y ago
  >I don't think so. I think when you introduce computers to anything, you add more edge cases, not less. That's sort of my point. Accidents are rarely caused by edge cases, but by predictable human failures. All the information needed to avoid the typical accident was plainly available, but the driver just didn't see it, process it or react to it appropriately. Driving is generally very predictable, but human attention and perception is hugely fallible. An AI that gets the basics consistently right but occasionally freaks out in an unpredictable situation would be a huge improvement over human drivers.
- path411 10y ago
  Cars are already full of software.
foota 10y ago
One potential flaw here is that our current driving experience is based on humans. Perhaps humans are relatively good at handling unexpected situations, meaning they don't cause many accidents. Isn't it possible that by switching to AI we shift from mundane accidents to unexpected situation accidents?

maverick_iceman 10y ago

This is not true. DeepMind has built AI that can understand the rules of the game just by looking at the screen and then play it perfectly.

Noos 10y ago
When I was a kid, we used to have to replace chess pieces we lost while playing during lunch break with ersatz ones...pennies, nickels, etc. I'd like to see Deep Mind try to tell a game from that.