Slashdot Mirror
iPhone Hacker Geohot Builds Self-Driving Car AI (bloomberg.com)
An anonymous reader writes: George Hotz, known for unlocking early iPhones and the PlayStation 3, has developed an autonomous driving system in his garage. "Hotz's approach isn't simply a low-cost knockoff of existing autonomous vehicle technology. He says he's come up with discoveries—most of which he refuses to disclose in detail—that improve how the AI software interprets data coming in from the cameras." The article has a video with Hotz demonstrating some basic autonomous driving similar to what Tesla rolled out earlier this year. He's clearly brimming with confidence about what the system can accomplish with more training.
"and a 21.5-inch screen is attached to the center of the dash. “Tesla only has a 17-inch screen,” Hotz says."
Now, THAT is innovation! It has a 21.5 inch screen! Not like the puny 17-inch on the Tesla. Obviously this guy is a big thinker.
Can someone please kick this guy in the nuts every day for the rest of his life?
Thanks.
Does he have insurance coverage for his selling idea? Seems very risky as he can be on the hook for big damages with stuff goes wrong.
Is he willing to have his code go under some thing like a FAA code audit?
How much redundancy is in that system?
Does his friends really want to take on the risk?
" “I live by morals, I don’t live by laws,” Hotz declared in the story. “Laws are something made by assholes.”"
We rebels don't need insurance! If we kill your family on the highway, well LOL dude. #hacktheplanet
We'll see. I do kind of hope that his youthful arrogance doesn't get him killed. It seems unlikely that one kid will be able to outdo the big-budget teams of researchers working on this problem -- but I don't think it's impossible.
Good luck with Michigan's unplowed roads and potholes! I doubt an autonomous car will ever be able to handle it here without killing people.
The article has a video with Hotz demonstrating some basic autonomous driving similar to what Tesla rolled out earlier this year.
Basic autonomous driving is (relatively) easy to do as long as you don't care much about it being actually useful in the real world. I suspect many good programmers and engineers could accomplish something functional (and dangerous) pretty quickly. It's not much more than an RC car with some sensors. Think Roomba on steroids. The problem is all the corner cases needed to actually make the system safe in real world driving. That is highly non-trivial.
People like you really disgust me.
With estrogen filled pussies such as yourself, we'd never make it to the moon, or never have any type of scientific advancement in the name of fear-mongering and paranoia.
I'm glad people walk this earth willing to take risk in order to advance mankind as a whole, without worrying about what the fuck geico thinks.
Please take your fear-mongering to a fox news board.
Thanks
FAA code audit is a joke so not sure what your are meaning by that, how much redundancy is in any system? i know how much redundancy competitors have built in!!! not much!
Why are we even talking about this?
He says he's come up with discoveries—most of which he refuses to disclose in detail.
So some guy may or may not have done anything worthy of note; he's not saying, so we don't know. I'm glad that he's "brimming with confidence" over what his system might be able to accomplish, but I fail to see why any shits should be given by the rest of us at this point. When he's actually accomplished it, and he's ready to disclose what he's done that's so great, send me a memo.
So the gist of it is that, after capturing data and training the AI, the car functions using only regular old cameras. The prohibitive cost in current self driving cars is the expensive sensors. So his system is much more economical. He envisions a kit to turn cars into self driving cars costing $1000 (obviously only certain cars with the ability to control steering digitally, etc, would be supported).
Better known as 318230.
...for without you, I wouldn't have root on my phone - Verizon would have taken it from me. I'd buy your car any day.
Make it rain!
most of which he refuses to disclose in detail
Snake, meet oil.
" “I live by morals, I don’t live by laws,” Hotz declared in the story. “Laws are something made by assholes.”"
We rebels don't need insurance! If we kill your family on the highway, well LOL dude. #hacktheplanet
Not only that but you can't tell us we're wrong
“For the first time in my life, I’m like, ‘I know everything there is to know’”
I am Slashdot. Are you Slashdot as well?
Exactly, but I think the key here is he's saying he's made "discoveries" to help the AI be much more robust. I don't believe him and you shouldn't either, but that's the take away. Autonomous vehicles existed, technically, back as far as ww2 (self guiding V rockets, can't recall if it was v1 or v2 but they were self guiding) That in itself is meaningless, if I wanted a car that can drive itself I could probably hodge podge it together quickly. It would result it my death and probably a few others because just making the car drive itself is not enough. It needs to not only react to things, it has to ANTICIPATE things. While I tend to root for the underdog and would love to see a guy tinkering in his garage make this happen, I find it dubious. And considering this is the guy who showed everyone how to jailbreak iphones and mod their ps3 to PIRATE CONTENT because information should be free, or some bullshit, I don't trust the fact that he's made great "discoveries" in regards to AI.
If killing people on the highway hacks out fixes, leading to quality AI for cars just a year or two sooner than a measured approach does, you will have net saved several million lives over a handful of people.
This is similar to arguments the FDA is a net killer -- the precautionary principle delays introduction of treatments rather than let some drugs get to market a little early, and dangerously.
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
Freeway autonomous driving is doable. But on regular streets it's hard, maybe impossible given current roads and parking lots.
But a freeway-only fully autonomous vehicle is still very valuable. Long-haul trucks and RVs spend most of their driving time on freeways. If a trucker can sleep 8 hours while the truck drives itself on the freeway and then take over only when the truck exits the freeway, the trucking company can save huge amounts of money. You can basically double the productivity of a driver/truck combo, since you can operate it continuously instead of having to shut down for the night. Also it's a plus from a safety standpoint; tired sleep-deprived truck drivers cause a lot of accidents. It's worth doing.
I agree. I would like to offer you to be killed first in order for progress to be made. Obviously his approach of mounting a 21" Dell monitor and a joystick in his car is the way to go. Who needs research and closed testtracks when you have a Dell? #hacktheplanet!!!
Does he have insurance coverage for his selling idea? Seems very risky as he can be on the hook for big damages with stuff goes wrong.
Is he willing to have his code go under some thing like a FAA code audit?
How much redundancy is in that system?
Does his friends really want to take on the risk?
I think you're confusing engineering with science.
The difficult part of AI, or science in general, is getting something that works. Once you have a working demo, anyone can add the reliability, the redundancy, and do a code audit.
And indeed, visionary investors might examine the idea and think "I'll take on the responsibility for liability and development, because I believe that the value of your ideas will be worth more than the expense of dealing with those issues. Sell me your idea."
But it all starts with getting something to work.
I saved money by using an old Nokia phone. I can't share any of my secrets because they are secret.
If killing people on the highway hacks out fixes, leading to quality AI for cars just a year or two sooner than a measured approach does, you will have net saved several million lives over a handful of people.
So you're going to make sure to offer up yourself and your family to be the first people to be killed, right?
The end.
If killing people on the highway hacks out fixes, leading to quality AI for cars just a year or two sooner than a measured approach does, you will have net saved several million lives over a handful of people.
Then why aren't you already volunteering yourself to be one of the other drivers around him while he tests his system? You're all gungho about deaths from testing this being fine yet you don't seem to be mentioning that you'll sacrifice yourself first.
This is similar to arguments the FDA is a net killer -- the precautionary principle delays introduction of treatments rather than let some drugs get to market a little early, and dangerously.
It's not really similar at all. Drug trials done for FDA approval use people who volunteer and give consent to be part of the trial. They aren't just random people that get injected the drug without their consent.
So says anyone who's ever participated in those screw loose cannonball runs.
They could be pure media click bait - but are supposedly highly regulated. Why don't people driving an average of over 100mpg proveably by their own recordings for publically airing their accomplishments lose all driving privs permanantly for being such rampant arseholes?
Got me.
Is it just me or would anyone else love to PIT one of those while they're trying to whiz by you at recklessly insane speeds?
The problem is all the corner cases needed to actually make the system safe in real world driving. That is highly non-trivial.
Indeed and that is what seems to be quite interesting about his approach. Basically he is saying that he is developing a system that can generate all those rules and corner cases itself, without a human having to quantify each scenario and code the rules into the machine. He states in the video that the car has gotten to where it is now (basic highway driving) by teaching itself. If it has, and his approach is extendable, then this is quite an interesting solution to the problem, precisely because it may deal with the non-trivialities you describe in a, well, comparatively trivial way.
Unfortunately, based on the article and video, there isn't really any way to determine whether he will be able to extend his system to give better performance, or even whether his system is just one of those 'learning systems' that is actually so highly tuned to the problem domain that it is essentially just an obfuscated rule based program. I guess we will have to wait for it to either get better, or for him to release some more information.
The main thing that makes me suspicious is why he has gone to the media about this now. If he has gotten this far with a design that actually does use learning, then why not spend a bit longer and get it to the point where he can demo it in less predictable environments. That would get us all interested. As it stands his current system only works in very predictable situations, so without more information it is impossible to know if this is a scam or not.
"Amazing: Frankly, I think you should just work at Tesla,” Musk wrote to Hotz in an e-mail. “I’m happy to work out a multimillion-dollar bonus with a longer time horizon that pays out as soon as we discontinue Mobileye.”
“I appreciate the offer,” Hotz replied, “but like I’ve said, I’m not looking for a job. I’ll ping you when I crush Mobileye.”
Musk simply answered, “OK.”
The video makes it sound like geohot's big advance is preferring a learning algorithm over a rules based approach. Rules based approaches were popular in AI in the 80's, when researchers thought they could emulate reasoning with a system of logical rules. This approach has not been in vogue for 20+ years, so I'm wondering what is revolutionary about his approach to AI?
I suppose that's some *sort* of logic. Would you have applied it to AIDS victims or to Muslims if you had been given the chance? How about white people? They've probably killed more people in the past few hundred years. If we kill all the white people then that'll surely pan out as a life-saver eventually - with enough time.
We can save infinite lives if we just kill all the humans at once. We should do that. Think of the children!
So, yes, it was *some sort* of logic.
"So long and thanks for all the fish."
It was the V-2 and was. for some limited definition, self-guided. The V-1 was just good until the power went away and then it went silent and fell where physics told it to. The controller in the V-2 was a little more complicated but not *vastly* so IIRC. It still couldn't go much more than where you pointed it at but it was able to stabilize itself a bit better. It basically followed a predetermined path using accelerometers and gyroscopes and wasn't all that controllable but it was autonomous, to some extent.
They couldn't just say, "Go hit Antwerp!" And it might not even land on Antwerp but out in the ocean or not even make it that far. By the way, Antwerp got more V-2 visits than anywhere else if I am remembering correctly. It wasn't all that accurate, I think it had something like a 2 (maybe it was 6) km accuracy rating. I'm drawing from memory (i'm a wee bit of a fan of history) so I've not made a scholarly study or anything and this might be mistaken but I think you'll find it's factually correct or pretty close.
"So long and thanks for all the fish."
It's Geohot. He's a known uncontrolled dumbass with little forethought and lots of hyperfocus.
Support my political activism on Patreon.
We may joke around and laugh at this guy, but he is basically a one-man Google in terms of self-promotion, ethics, and respect for others.
An interesting issue (and it applies to not just this guy but to the big guys too) is the recent research that showed pretty much all neural nets will incorrectly classify input that is almost identical to input that gets classified correctly (including deep nets that are showing best results in image classification etc.). They chose to tweak a very small amount of input that maximized error causing mis-classification and found that they were able to do this consistently across nets of varying architectures and training methods.
It's a pretty significant finding and one that shows we have a ways to go before we can trust systems like these for safety related activities.
“‘If’ statements kill.” They’re unreliable and imprecise in a real world full of vagaries and nuance. It’s better to teach the computer to be like a human, who constantly processes all kinds of visual clues and uses experience,
I think he meant to say that "They're reliable and precise in a real world full of vagaries and nuance." The statement is insightful - if statements are perfectly accurate and do exactly what they were coded to do, no more no less. Normally, this is what makes computers really powerful and reliable. But when dealing with humans, it is what limits them. It's an analog world out here.
https://www.youtube.com/watch?...
Except for perhaps the learnability, which is more of an AI than an automated driving advance? Though, in any case it would certainly be impressive to duplicate this independently.
It sounds like the gas and brake controlers are fairly commonly built in, but I was not aware of the steering controls, and he didn't mention adding any motors.
Bukowski said it. I believe it. That settles it.
OTOH, it's worth remembering that people make that same mistake.
I think we've pushed this "anyone can grow up to be president" thing too far.
No limited definition. It was fully self guided as was the V1. Technically it is not difficult, especially if one were to make it with modern day off the shelf components.
If killing people on the highway hacks out fixes, leading to quality AI for cars just a year or two sooner than a measured approach does, you will have net saved several million lives over a handful of people.
The alternative is that his hack kills some people which causes regulations to lock out future testing, which in-turn holds back the technology by decades, thus dooming even more people.
Who should be the person/people that make that decision?
Ok, the kid is bright, but he's also arrogant, reckless and probably a bit insane. But set aside his personality, I don't understand the negativity on this forum towards another geek who hacks things together to make it work. There are a lot of hard problems in what he is working on, and if he can come up with a new way to do computer vision, I would be really happy too. The current start-of-the-art in this field is convolutional neural network (CNN), which, basically, is just a kind of brute force pattern matching. I have been working on a robot that "can see, listen and understand, and climb tree" (the climb tree part is to design some mechanics flexible enough to climb tree, then it's flexible to handle any terrains), so I understand the difficulty of computer vision and speech recognition. What Hotz said sounds like snake oil, but who are we to judge when we haven't seen the details? I'll keep my mind open for now, and hope that we could have a better way than CNN.
There are many true stories about inventors being killed by their invention. I'd put money on this being another one of those stories.
Yeah, except that getting self driving cars onto the road a couple of years earlier won't save anybody. The people who are the most likely to cause automobile accidents are the last people who will be buying self driving autos. The initial effect of self driving cars on accidents will be essentially zero.
I would buy one immediately. In 20 years of driving I have never had an accident, never so much as a fender bender or backed into someone or something. The last "driving accident" I had occurred in a go-cart when I was 4. Will it save lives if I can get a self driving car a couple years earlier? Not really.
Joe six-pack, who always buys SUVs because "I feel safer when I get in an accident" and drives pedal to floor start & stop between every stop light, he's not going to buy a self driving car, because "I know how to drive better than some machine!"
Self driving cars won't do anything to save lives for years after there are many different makers and models available, so that people can still be "individuals" instead of just being safe.
Killing people to get a product onto market which will save nobody initially is a waste, and it's short sighted.
Not sure if anyone else caught this but I thought it was funny and decided to share. If you go to his site http://comma.ai and take a look at the source code, you will see the line below in a comment
till everyone who doubted me is asking for forgiveness
if you ain't been a part of it, at least you got to witness
bitches
The line was taken from the song Forever by Drake