NVIDIA’s CEO Sees Fully Autonomous Cars within 4 Years

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
Jensen Huang said today that artificial intelligence would enable fully automated cars within 4 years: “It will take no more than 4 years to have fully autonomous cars on the road. How long it takes for the vast majority of cars on the road to become that, it really just depends.” He also commented on the cryptocurrency situation, taming expectations of strong revenue growth from a wave of interest in cryptocurrencies.

“There are many tasks in companies that can be automated... the productivity of society will go up,” said NVIDIA’s Huang. But Huang joined peers taming expectations of strong revenue growth from a wave of interest in cryptocurrencies. AMD expected this week that there will be some leveling off of cryptocurrency demand. “Revenue for us in crypto is over $100 million a quarter. For us, it’s a small percentage... It’s obviously not a target market,” Huang said.
 
The biggest obstacle is going to be getting people to stop driving themselves. Stupid people driving would disrupt the autonomous cars too much.
 
I highly doubt it. AI does not scale linearly with increases in node processing power. The current best AI is at about an IQ level of a 5 year old. In 4 years that might be up to 7. Do you want a 7 year old driving your car?
 
Now, 4 years, or be it 10. The fact people will not be driving is a good thing.

The bad part of automation - which the politicians are blind to - is the fact that we'll see a huge swath of employed people become unemployed. Far larger than previous transitions.
 
PaulP said:
I highly doubt it. AI does not scale linearly with increases in node processing power. The current best AI is at about an IQ level of a 5 year old. In 4 years that might be up to 7. Do you want a 7 year old driving your car?
Click to expand...

Computers grow at an exponential rate. You are talking about a linear growth.

Easy example: AlphaGo. In one year's time it's gone from beating the human grand master to (with the new iteration) beating the original 100 games to 0.

The jump from village idiot to Einstein will only take an iteration or two (2-4 years)
 
travisty said:
Computers grow at an exponential rate. You are talking about a linear growth.

Easy example: AlphaGo. In one year's time it's gone from beating the human grand master to (with the new iteration) beating the original 100 games to 0.

The jump from village idiot to Einstein will only take an iteration or two (2-4 years)
Click to expand...
Wanna bet? I'll check back with you in a few years and we'll see who has to eat crow.
 
PaulP said:
Wanna bet? I'll check back with you in a few years and we'll see who has to eat crow.
Click to expand...

Just to be clear, the study saying that computers are at any IQ was not accurate IMO. I still hold onto the idea that the point of singularity - https://en.wikipedia.org/wiki/Technological_singularity - is around 2040.

As of this point though, the ability for autonomous driving is here and will be available for public use over the next 5 years. Laws will be put in place disallowing human driven vehicles in the not too distant future.

Ah found the video i was looking for:
 
dgingeri said:
The biggest obstacle is going to be getting people to stop driving themselves. Stupid people driving would disrupt the autonomous cars too much.
Click to expand...

No, that isn't the biggest obstacle.
 
PaulP said:
Wanna bet? I'll check back with you in a few years and we'll see who has to eat crow.
Click to expand...

Self driving cars are already better than human drivers, ALL human drivers. It will take longer for people to admit it and allow computers to drive for them than it will for technology to get far better than humans could ever be at such tasks.
 
dgingeri said:
The biggest obstacle is going to be getting people to stop driving themselves. Stupid people driving would disrupt the autonomous cars too much.
Click to expand...
They already disrupt other people.
travisty said:
The bad part of automation - which the politicians are blind to - is the fact that we'll see a huge swath of employed people become unemployed. Far larger than previous transitions.
Click to expand...
That can be a good part as well, if only we realized the enormous potential that gives us.
There is no point in making humans do work that can be automated. They can be freed up to pursue more meaningful careers. So the same job gets done with less human labour then more value is created. Therefore we should be able to afford to give the same or better living conditions to people even if they don't have to work 40 hours but only 30 or 20 or 10 or eventually zero. That's what I'd call technological singularity when human needs can be met without human labour.
 
dgingeri said:
Self driving cars are already better than human drivers, ALL human drivers.
Click to expand...

I'll admit I've yet to research in depth, but have they moved on from the well paved areas in their testing? Last I read, around 2016, they were still having difficultly in poor weather conditions. I'm less concerned about the driving ability and more with the sensor reliability, durability, and accuracy.

I'm also concerned about practicality. Unless you can link it to your mind, you can't suddenly change your destination. As in you're driving down the street, see a store you just remembered you need to go to which is rapidly coming up and have 3 seconds to change lanes or exit the street. I suppose voice activation may remedy some of those situations, but certainly not all. If the business is not in a registry good luck getting the car to automatically changing destinations. Likewise, I can't imagine you specifically telling a car to stop next to a certain object or certain person.
 
Flogger23m said:
I'll admit I've yet to research in depth, but have they moved on from the well paved areas in their testing? Last I read, around 2016, they were still having difficultly in poor weather conditions. I'm less concerned about the driving ability and more with the sensor reliability, durability, and accuracy.

I'm also concerned about practicality. Unless you can link it to your mind, you can't suddenly change your destination. As in you're driving down the street, see a store you just remembered you need to go to which is rapidly coming up and have 3 seconds to change lanes or exit the street. I suppose voice activation may remedy some of those situations, but certainly not all. If the business is not in a registry good luck getting the car to automatically changing destinations. Likewise, I can't imagine you specifically telling a car to stop next to a certain object or certain person.
Click to expand...

For your first point "difficulty" with poor weather conditions means that the self driving cars become slower and sometimes stop entirely to avoid damage. I consider that light years ahead of human drivers. As we develop multiple sensor types, motion sensors and temperature sensors in addition to cameras, that can overlay for the computer's consideration, as well as sensor redundancy, things will pick up exponentially.

As for your second point, people don't handle that particularly well already. I see far, FAR too many people suddenly changing lanes from left to right with no warning over changing their minds. I think it would be best to force people to get away from that. I HATE it when people do that. It's stupid and reckless and endangers the people around them. Computers would take this into account and do what people SHOULD be doing, and that is patiently and cautiously getting turned around and to the new destination, WITHOUT endangering everyone around them.

That's just two aspects I think computers are doing things better than (stupid) people when it comes to driving.
 
travisty said:
Computers grow at an exponential rate. You are talking about a linear growth.

Easy example: AlphaGo. In one year's time it's gone from beating the human grand master to (with the new iteration) beating the original 100 games to 0.

The jump from village idiot to Einstein will only take an iteration or two (2-4 years)
Click to expand...

Not even close to the same.

One deals with a set of very few well defined rules on a very small playing field with minimal choices vs a known opponent with known programing and rule set . AI needed for a car deals with real world movements, and laws, the processing needed just to understand the world its working in doesn't even exist yet, all the systems we have to date go out the window when you get onto a road without nice fresh road markings, toss it into a construction area, no less a rural setting and it all comes to a stop. That is just dealing with set rules and bounds, not the other laws or even moral choices that take place while driving, such as avoid a crash but kill the occupants, or hit the jaywalker?

The issues they are facing are not small. And the problems stretch from hardware to software as well as moral and legal ramification.
 
  • Like
Reactions: PaulP
like this
dgingeri said:
Self driving cars are already better than human drivers, ALL human drivers. It will take longer for people to admit it and allow computers to drive for them than it will for technology to get far better than humans could ever be at such tasks.
Click to expand...

Got the stats?

Becuase last time I did them, which was with google's last report on their self-driving fleet (they are the only one publishing data), the self-driving cars were performing about 80% of the human populace. And they had yet to deal with bad rain, destroyed roads, winter, serious pot holes, driving drowsy, dwi, etc.

NVIDIA's demo self driving car litterally cut every left hand turn running through the oncoming lane in their demo video. That's shit.

Ford admits that despite an array of sensors, they are having trouble making it work well/reliably in foul weather. In large part because one or more sensor types get fouled due to the weather, and without all of them there are critical elements the algorithm ceases to be able to deal with properly.
 
dgingeri said:
For your first point "difficulty" with poor weather conditions means that the self driving cars become slower and sometimes stop entirely to avoid damage. I consider that light years ahead of human drivers. As we develop multiple sensor types, motion sensors and temperature sensors in addition to cameras, that can overlay for the computer's consideration, as well as sensor redundancy, things will pick up exponentially.

As for your second point, people don't handle that particularly well already. I see far, FAR too many people suddenly changing lanes from left to right with no warning over changing their minds. I think it would be best to force people to get away from that. I HATE it when people do that. It's stupid and reckless and endangers the people around them. Computers would take this into account and do what people SHOULD be doing, and that is patiently and cautiously getting turned around and to the new destination, WITHOUT endangering everyone around them.

That's just two aspects I think computers are doing things better than (stupid) people when it comes to driving.
Click to expand...

So what you're saying is, autonomous cars are not yet ready for prime time as they are unable to do what people can regularly do very easily. Clearly, they're not yet good enough.

When they're capable on doing things people do regularly on a similar level I'll certainly be happy. Seems like in 5-8 years, they may very well be an ideal choice for your daily route on the freeway around an urban area. Everything else, best to stick to manual until they can develop reliable sensors. And of course, until they can develop a good interface system it will be best to use manual if you're doing anything but your daily route.
 
  • Like
Reactions: PaulP
like this
dgingeri said:
For your first point "difficulty" with poor weather conditions means that the self driving cars become slower and sometimes stop entirely to avoid damage. I consider that light years ahead of human drivers. As we develop multiple sensor types, motion sensors and temperature sensors in addition to cameras, that can overlay for the computer's consideration, as well as sensor redundancy, things will pick up exponentially.
Click to expand...

This statement alone proves you don't know how these things work at all.
 
  • Like
Reactions: PaulP
like this
raz-0 said:
Got the stats?

Becuase last time I did them, which was with google's last report on their self-driving fleet (they are the only one publishing data), the self-driving cars were performing about 80% of the human populace. And they had yet to deal with bad rain, destroyed roads, winter, serious pot holes, driving drowsy, dwi, etc.

NVIDIA's demo self driving car litterally cut every left hand turn running through the oncoming lane in their demo video. That's shit.

Ford admits that despite an array of sensors, they are having trouble making it work well/reliably in foul weather. In large part because one or more sensor types get fouled due to the weather, and without all of them there are critical elements the algorithm ceases to be able to deal with properly.
Click to expand...

LIDAR is the best form of object sensing available in cars. Because it works off lasers, it reflects, and defracts in things like snow, sleet, and rain. IR and Visible spectrum are cameras which can be fouled with dirt, snow, ice, mud, poop, cracks, etc.

For AI to work it has been be trained first. And certain road conditions are so unique that it would be impossible to train for all of them. For example: Outdated GPS. Exits & removed, redesigned or removed, construction sites, merges during accident scenes, bad road markings, malicious vandalism or poor road signs.

Even Ford said at Level 5, they would require a trunk load of computing power. I figured out the FLOPs they were aiming out based on their power specs. Even if you went to the smallest transistor size possible and ignored growing tunneling and leakage issues, the additional power would require a much larger alternator to power it. That cuts into fuel economy.

Identifying a vehicle, it's direction, position, and speed in clear day conditions is easy. Identifying everything else like: Is that pedestrian going to dash across the street? What does that old road sign say? Where are the lanes on this unmarked road? My car feels like it's hydroplaning. All this requires dynamic thinking and intuition also. I could name dozens of situations like this.

It's a pipe dream.
 
  • Like
Reactions: PaulP
like this
He hopes they will in 4 years so he can make all that money back he just spent on developing the chips for them.
 
I know I'm going to get scourged for this but I honestly do not understand the fascination with autonomous vehicles.

Am I the only one that likes to drive?
 
M76 said:
That can be a good part as well, if only we realized the enormous potential that gives us.
There is no point in making humans do work that can be automated. They can be freed up to pursue more meaningful careers. So the same job gets done with less human labour then more value is created. Therefore we should be able to afford to give the same or better living conditions to people even if they don't have to work 40 hours but only 30 or 20 or 10 or eventually zero. That's what I'd call technological singularity when human needs can be met without human labour.
Click to expand...

I totally agree that it can be good and hope to be in a world where everyone can choose how they spend their day. It's the transition that's going to be a mess if things arnt done now
 
jwcalla said:
I know I'm going to get scourged for this but I honestly do not understand the fascination with autonomous vehicles.

Am I the only one that likes to drive?
Click to expand...
I'm with you, I enjoy driving too - outside the city. I can see how these could be revolutionary in that regard, on paper at least.

What floors me is how quickly people are willing to jump into it head-first in an age of security flaws and products manufactured for cheap as possible. Call me overly pessimistic, but I'll gladly eat crow if in 10 years I'm sleeping on my way to work... and heck, playing a fully realized Star Citizen that meets or exceeds everyones expectations when I get home. (I may as well go big!)
 
dgingeri said:
Self driving cars are already better than human drivers, ALL human drivers. It will take longer for people to admit it and allow computers to drive for them than it will for technology to get far better than humans could ever be at such tasks.
Click to expand...

How long will it take for you to admit that you're simply wrong. Autonomous cars are not better than human drivers. They can't drive in inclement weather. They can't drive well or at all on roads that don't have high resolution 3D maps (which is majority of the roads in the US). I'd love to see how any of them would handle snow covered roads. I don't mean where any of the asphalt peaks through, I mean literally the whole road is covered. Something extremely common for my home state of Idaho. That or dirt trails that aren't on any maps.

There are instances where the self driving car is better than a human driver, but it's way too small a percentage for them to be sent into prime time. In 4 years time, I see the hardware being at a point that we could have autonomous vehicles, but I don't think the software would be there yet and it'd probably be cost prohibited.
 
  • Like
Reactions: PaulP
like this
Well autonomous or semi-autonomous can't come soon enough. Increasingly I'm having to travel long, boring stretches like LA to Vegas or LA to San Francisco - and just holding the wheel with cruise control on is displacing time I could be more productive, or reading/watching movie.

Even if sleeping during isn't feasible yet, would be great to be able to work on a laptop or phone while the car "holds the wheel" and handles itself, and only notifies me with options on where to stop for gas or food. I *think* Teslas already do this but not sure.
 
Last edited:
Our entire highway system would have to change. There would have to be lanes dedicated to only allow autonomous vehicles or human drivers would be causing way too many wrecks vs. the AI one's. There's also the huge problem of economics. How many people do you think will be able to trade in their 1992 Toyota for a new self-driving car just because you mandated they have one? There are around 250 million cars in the US.... not to mention all the roadway changes that need to be made to make self driving cars work right in traffic. At just $30K per vehicle (assuming they only cost that much), that's $10 Trillion just to replace all the cars alone. This is going to take a lot longer than a few years to work through.
 
DPI said:
Well autonomous or semi-autonomous can't come soon enough. Increasingly I'm having to travel long, boring stretches like LA to Vegas or LA to San Francisco - and just holding the wheel with cruise control on is displacing time I could be more productive, or reading/watching movie.

Even if sleeping during isn't feasible yet, would be great to be able to work on a laptop or phone while the car "holds the wheel" and handles itself, and only notifies me with options on where to stop for gas or food. I *think* Teslas already do this but not sure.
Click to expand...
Editted for honesty :)

Sad fact for me is that if I'm not driving so I'm not engaged in the vehicle's behavior, I will get car sick if I try to read. I can probably do video for a little bit, but top out at 20 minutes. That is time I don't get back.

OTOH, the laws are going to require you to be engaged and ready to take over in an instant. No way you can do that if you're reading. That level of autonomous and more importantly the trust in it is way far down the road.
 
You must log in or register to reply here.
Back
Top