Slashdot Mirror
Is It Illegal to Trick a Robot? (ssrn.com)
An anonymous reader writes:
Can you get into trouble under anti-hacking laws for tricking machine learning...? A new paper by security researchers and legal experts asks whether fooling a driverless car into seeing a stop sign as a speed sign, for instance, is the same as hacking into it.
The original submission asks another question -- "Do you have inadequate security if your product is too easy to trick?" But the paper explores the possibility of bad actors who deliberately build a secret blind spot into a learning system, or reconstruct all the private data that was used for training. One of the paper's authors even coded DNA that corrupts gene-sequencing software and takes control of its underlying computer, and the researchers ultimately warn about the dangers of "missing or skewed security incentives" in the status quo.
"Our aim is to introduce the law and policy community within and beyond academia to the ways adversarial machine learning alter the nature of [cracking] and with it the cybersecurity landscape."
The original submission asks another question -- "Do you have inadequate security if your product is too easy to trick?" But the paper explores the possibility of bad actors who deliberately build a secret blind spot into a learning system, or reconstruct all the private data that was used for training. One of the paper's authors even coded DNA that corrupts gene-sequencing software and takes control of its underlying computer, and the researchers ultimately warn about the dangers of "missing or skewed security incentives" in the status quo.
"Our aim is to introduce the law and policy community within and beyond academia to the ways adversarial machine learning alter the nature of [cracking] and with it the cybersecurity landscape."
Modifying a stop sign with the purpose of fooling a self-driving car is similar to someone tampering with a stop sign to fool human drivers, and can be handled with existing laws.
The answer is probably going to depend upon one word:
Computer Fraud and Abuse Act (18 USC 1030):
(a) Whoever--
(5)
(A) knowingly causes the transmission of a program, information, code, or command, and as a result of such conduct, intentionally causes damage without authorization, to a protected computer;
Can you convince judges that "cause the transmission" should only mean active electronic transmission, or can prosecutors convince judges that "cause the transmission" should have the same epidemiological sense as causing the transmission of a virus, worm, etc, regardless of means.
There's going to be kids that are going to see videos and attempt to recreate any flaw - just like there's plenty of pennies smashed on train tracks over the years (not really dangerous, but if kids could be jailed for intent...), there's going to be flaws in any automated system by random folks you can't "teach a lesson to."
One of the biggest purposes of having an automated system approaching computerization ("robot", if that's what gets clicks), is that you can spot flaws, and ALTER the system to better adapt to changing needs, rather than rely on pure punishment to cover faults.
Related Obligitory XKCD:
https://xkcd.com/1958/
Folks can use the power of misleading information to kill eachother in a lot of ways, poison and war being classic examples - and yeah, those should be punished, but they should also be used to make systems that work better.
Overall, these things still make the world better, and less randomly susceptible to harm. The analogue equivalent has more holes in virtually every case, we're just more used to them. From almost all past technology (non-weapon) , we're better off after going through the learning process than if we feared it forever, or remained only conservative in our approach.
So yeah - punish folks that have actual intent to harm just like anything, but you can't stop folks from playing with the world around them, and the new stuff in it.
Ryan Fenton
Is cutting the brake lines on a car a security issue? Of course not. But it is a crime.
Exactly. Just saw this article on advertisers and behavior control. It cited how people respond to the smell of disinfectant by keeping a room cleaner, cited it as a sort of mental weakness. Of course non-sociopathic people, on smelling disinfectant, will take it as a sign someone really wants the room clean, and thus keep it clean as a courtesy and possible medical safety thing...But advertisers see this sort of thing as a behavioral switch, and would feel free to place disinfectant smells in a businesses just to get that behavior. The whole mode of thinking behind advertising is the kind of manipulation that could lead AI astray.
-The art of programming is the pursuit of absolute simplicity.
Trying again...
"99 dead duelists of Dios on the wall. 99 dead duelists of Dios! Take one's ring, pass it around..."
Add some gator-misspellings to that sign and it would be a Far Side comic.
Comment removed based on user account deletion
I'm honestly curious as to how these self driving or driverless cars handle traffic light outages or intersections that just have blinking yellow lights in small towns like where i live.
Same way as human drivers. Road rules are a pretty simple system; do you honestly imagine that the engineers behind the technology managed to "teach" the car to recognise traffic lights, but somehow forgot to include programming on what to do if they're out?
6. Enjoy a free anal cavity search after the authorities check the footage from the dozens of cameras festooned all over the truck.
Existing laws cover such behavior. Expect charges ranging from Malicious Mischief to Vandalism to Terrorism depending on how vindictive the prosecutor feels.
"You want to know how to help your kids? Leave them the fuck alone." -George Carlin
yes it's illegal to cause traffic accidents. be it by defacing signs, stealing stop signs, or screwing with the road markers. this is not even a question.
What if you cause it by wearing a custume looking like a stop sign to a computer, but like a custome to a human?
Comment removed based on user account deletion
Are we talking about robots that are going to kill you?
"bad actors who deliberately build a secret blind spot" - Reminds me of Robocop's "Directive 4". I dunno why.
It is pitch black. You are likely to be eaten by a grue.
If you were designing a system to deal with unserviceable or intermittent traffic lights, what would you do?
Are you saying you can't think of a way too deal with it?
Or are you saying that you can, but you're so stuck up that you think none of the people working on this stuff are as smart as you?
* someone knowingly and intentionally circumventing security, and
* when the robot has a flaw and behaves unexpectedly.
I can see some companies maknig accusations of malicious interference as a way to save face.
It's a little worrying that some of the companies with the most advanced AI research are actively hostile to privacy and freedom. Everyone knows this is going to end in a horrific disaster - it's kinda obvious. Yet seemingly many people consider the prospect of slightly cheaper taxi rides to be totally worth the risk.
Better empty the truck fast. When it detected your robbery it called in a drone strike.
Gulag FTW!
yes it's illegal to cause traffic accidents. be it by defacing signs, stealing stop signs, or screwing with the road markers. this is not even a question.
What if you cause it by wearing a custume [sic] looking like a stop sign to a computer, but like a custome [sic] to a human?
Remember, intention is the key. If you were going to a costume party, dress up like that, and on the way to the party, then it is not your fault but rather the AI. On the other hand, if you just dress up like that and stand along a road/street where self driving cars often time go by, then it could be illegal depending on how they interpret your intention (and likely you would be at fault).