It's worth remembering that autonomous driving is a very young technology, rapidly-developed, with a lot of unknown elements. That it has a lot of flaws is but a given, so early in the R&D. It's...
It's worth remembering that autonomous driving is a very young technology, rapidly-developed, with a lot of unknown elements. That it has a lot of flaws is but a given, so early in the R&D. It's on its way to undergo lots of tests – which makes "hacks" like this so important for its maturity. Laboratory tests can only get you so far; the real world will provide a whole lot of different and unexpected challenges.
It does not help that the results of your progress will become very public if you're in the driverless-car business. And because of human nature (or maybe culture), you can make spectacular progress over a decade of hard work – and hit a bump that threatens to overturn your success when your car unexpectedly swerves into a bus stop full of people because of a bad pixel in the camera's matrix or something. Minor defects that would go quiet would, instead, cause tragic deaths and become a point of deep scrutiny.
"I expect this groundbreaking technology to work before it's released widely" is not a radical statement, but it barely stands the test of reality. Sure, you want to expect that the electric juice presser will not zap you every time you touch the buttons, let alone blow up and shower you with metal shrapnel for some fucking reason. Similarly, you'd want your autonomous car to be able to detect all regular traffic concerns: people running across the road where and when they aren't supposed to, other cars being driven recklessly nearby, a tree falling across the road, loss of grip on spilled oil or ice...
You could argue that being able to distinguish a patch of salt from a road line should be on the list, but you don't see that often. Weird shit is going to happen on the road, sure, and ideally, you want your car to be able to react appropriately – but it's not that simple, nor is it easy to rectify.
Which isn't to say that autonomous cars are naturally dangerous, or some technophobe stuff like that: it's to say that the technology that promises to relieve us of work is not going to be perfect – not initially, not eventually – so no matter how sweet the promise, you should keep your head about yourself, rather than mindlessly trust that the Machine God will guide your vehicle steadily.
Actually, exercises like this help improve the platform. The error was identified and correct so win-win all around- plus nobody got killed to find out the flaw.
Actually, exercises like this help improve the platform. The error was identified and correct so win-win all around- plus nobody got killed to find out the flaw.
This is how I feel about most emerging technologies. Computers were madly insecure in their initial versions, but have been slowly hardened over time. Other industries will need to go through the...
This is how I feel about most emerging technologies. Computers were madly insecure in their initial versions, but have been slowly hardened over time. Other industries will need to go through the same process; be it self-driving cars, IOT, or even medical devices.
Yes there is much to learn from existing lessons, but there are new problems unique to their respective industries as well. For instance self-driving cars have to answer the unique problem of civilians intentionally darting in front of the vehicles.
That video seriously felt like an anti-ad, there to smear the autopilots reputation. There is way too much production value for it to be just a bunch of researchers doing research.
That video seriously felt like an anti-ad, there to smear the autopilots reputation. There is way too much production value for it to be just a bunch of researchers doing research.
This really is no different than any other kind of hacking. At it's base level hacking is providing input to a system (usually unintended by the designers of that system) that produces an output...
This really is no different than any other kind of hacking. At it's base level hacking is providing input to a system (usually unintended by the designers of that system) that produces an output desired by the attacker. The only difference here is that the input devices are cameras and LIDAR sensors instead of keyboards. To me it seems that trying to call this "tricking" instead of hacking shows a lack of understanding of computers of the person that wrote this article.
If you read the article, you'll see that it's more nuanced than this. For example, the section titled "The law of tricks" discusses the difference between hacking and tricking a computer,...
To me it seems that trying to call this "tricking" instead of hacking shows a lack of understanding of computers of the person that wrote this article.
If you read the article, you'll see that it's more nuanced than this. For example, the section titled "The law of tricks" discusses the difference between hacking and tricking a computer, including the legal definition of hacking in the USA, and how that legal definition of hacking doesn't cover the activity described in the leading paragraphs of the article.
Your right, I only got about halfway through the article before I commented. I see why the author said what they did, however I do still hold that the distinction seems a strange one to me, even...
Your right, I only got about halfway through the article before I commented. I see why the author said what they did, however I do still hold that the distinction seems a strange one to me, even concerning the law. It just feels like the same thing.
Might I suggest that, in future, you wait until you've reach the end of the article before commenting? Or, do what I sometimes do, and open a comment box or text file to write notes in as you read...
Your right, I only got about halfway through the article before I commented.
Might I suggest that, in future, you wait until you've reach the end of the article before commenting? Or, do what I sometimes do, and open a comment box or text file to write notes in as you read the article, and then post the whole thing when you've finished reading.
I see why the author said what they did, however I do still hold that the distinction seems a strange one to me, even concerning the law.
Well, you can't charge someone with hacking a computer if the activities they undertake don't match the legal definition of hacking - which, as on of the interviewees indicates, could mean the legal definition has to be changed.
Actually, what these so-called hackers did reminds me of something I used to do when I performed testing on software systems: give the systems unexpected inputs and see how they deal with it. I once crashed a system by inputting a negative number because a developer had assumed the system would only ever have to deal with positive numbers (which was true, but users are unpredictable and humans make errors). I wasn't hacking the system by inputting a negative number, I was testing it. To me, that feels like what these people did: by inputting white stickers instead of road markings, they were testing how the system deals with unexpected inputs. The system should handle them gracefully, without crashing - either metaphorically or, in this case, literally.
Your are correct, I was in the wrong for commenting like that without getting through the entirety of the article. Thanks for the advice about taking notes as I read, I'll try that out.
Your are correct, I was in the wrong for commenting like that without getting through the entirety of the article.
Thanks for the advice about taking notes as I read, I'll try that out.
Does that make them white hats? R&D? public testers?..
To me, that feels like what these people did: by inputting white stickers instead of road markings, they were testing how the system deals with unexpected inputs. The system should handle them gracefully, without crashing - either metaphorically or, in this case, literally.
Does that make them white hats? R&D? public testers?..
It's worth remembering that autonomous driving is a very young technology, rapidly-developed, with a lot of unknown elements. That it has a lot of flaws is but a given, so early in the R&D. It's on its way to undergo lots of tests – which makes "hacks" like this so important for its maturity. Laboratory tests can only get you so far; the real world will provide a whole lot of different and unexpected challenges.
It does not help that the results of your progress will become very public if you're in the driverless-car business. And because of human nature (or maybe culture), you can make spectacular progress over a decade of hard work – and hit a bump that threatens to overturn your success when your car unexpectedly swerves into a bus stop full of people because of a bad pixel in the camera's matrix or something. Minor defects that would go quiet would, instead, cause tragic deaths and become a point of deep scrutiny.
"I expect this groundbreaking technology to work before it's released widely" is not a radical statement, but it barely stands the test of reality. Sure, you want to expect that the electric juice presser will not zap you every time you touch the buttons, let alone blow up and shower you with metal shrapnel for some fucking reason. Similarly, you'd want your autonomous car to be able to detect all regular traffic concerns: people running across the road where and when they aren't supposed to, other cars being driven recklessly nearby, a tree falling across the road, loss of grip on spilled oil or ice...
You could argue that being able to distinguish a patch of salt from a road line should be on the list, but you don't see that often. Weird shit is going to happen on the road, sure, and ideally, you want your car to be able to react appropriately – but it's not that simple, nor is it easy to rectify.
Which isn't to say that autonomous cars are naturally dangerous, or some technophobe stuff like that: it's to say that the technology that promises to relieve us of work is not going to be perfect – not initially, not eventually – so no matter how sweet the promise, you should keep your head about yourself, rather than mindlessly trust that the Machine God will guide your vehicle steadily.
Actually, exercises like this help improve the platform. The error was identified and correct so win-win all around- plus nobody got killed to find out the flaw.
This is how I feel about most emerging technologies. Computers were madly insecure in their initial versions, but have been slowly hardened over time. Other industries will need to go through the same process; be it self-driving cars, IOT, or even medical devices.
Yes there is much to learn from existing lessons, but there are new problems unique to their respective industries as well. For instance self-driving cars have to answer the unique problem of civilians intentionally darting in front of the vehicles.
That video seriously felt like an anti-ad, there to smear the autopilots reputation. There is way too much production value for it to be just a bunch of researchers doing research.
This really is no different than any other kind of hacking. At it's base level hacking is providing input to a system (usually unintended by the designers of that system) that produces an output desired by the attacker. The only difference here is that the input devices are cameras and LIDAR sensors instead of keyboards. To me it seems that trying to call this "tricking" instead of hacking shows a lack of understanding of computers of the person that wrote this article.
If you read the article, you'll see that it's more nuanced than this. For example, the section titled "The law of tricks" discusses the difference between hacking and tricking a computer, including the legal definition of hacking in the USA, and how that legal definition of hacking doesn't cover the activity described in the leading paragraphs of the article.
Your right, I only got about halfway through the article before I commented. I see why the author said what they did, however I do still hold that the distinction seems a strange one to me, even concerning the law. It just feels like the same thing.
Might I suggest that, in future, you wait until you've reach the end of the article before commenting? Or, do what I sometimes do, and open a comment box or text file to write notes in as you read the article, and then post the whole thing when you've finished reading.
Well, you can't charge someone with hacking a computer if the activities they undertake don't match the legal definition of hacking - which, as on of the interviewees indicates, could mean the legal definition has to be changed.
Actually, what these so-called hackers did reminds me of something I used to do when I performed testing on software systems: give the systems unexpected inputs and see how they deal with it. I once crashed a system by inputting a negative number because a developer had assumed the system would only ever have to deal with positive numbers (which was true, but users are unpredictable and humans make errors). I wasn't hacking the system by inputting a negative number, I was testing it. To me, that feels like what these people did: by inputting white stickers instead of road markings, they were testing how the system deals with unexpected inputs. The system should handle them gracefully, without crashing - either metaphorically or, in this case, literally.
Your are correct, I was in the wrong for commenting like that without getting through the entirety of the article.
Thanks for the advice about taking notes as I read, I'll try that out.
Does that make them white hats? R&D? public testers?..