Slashdot Mirror
DARPA Will Stage an AI Fight in Las Vegas For DEF CON (yahoo.com)
An anonymous Slashdot reader writes: "A bunch of computers will try to hack each other in Vegas for a $2 million prize," reports Tech Insider calling it a "historic battle" that will coincide with "two of the biggest hacking conferences, Blackhat USA and DEFCON". DARPA will supply seven teams with a supercomputer. Their challenge? Create an autonomous A.I. system that can "hunt for security vulnerabilities that hackers can exploit to attack a computer, create a fix that patches that vulnerability and distribute that patch -- all without any human interference."
"The idea here is to start a technology revolution," said Mike Walker, DARPA's manager for the Cyber Grand Challenge contest. Yahoo Tech notes that it takes an average of 312 days before security vulnerabilities are discovered -- and 24 days to patch it. "if all goes well, the CGC could mean a future where you don't have to worry about viruses or hackers attacking your computer, smartphone or your other connected devices. At a national level, this technology could help prevent large-scale attacks against things like power plants, water supplies and air-traffic infrastructure.
It's being billed as "the world's first all-machine hacking tournament," with a prize of $2 million for the winner, while the second and third place tem will win $1 million and $750,000.
"The idea here is to start a technology revolution," said Mike Walker, DARPA's manager for the Cyber Grand Challenge contest. Yahoo Tech notes that it takes an average of 312 days before security vulnerabilities are discovered -- and 24 days to patch it. "if all goes well, the CGC could mean a future where you don't have to worry about viruses or hackers attacking your computer, smartphone or your other connected devices. At a national level, this technology could help prevent large-scale attacks against things like power plants, water supplies and air-traffic infrastructure.
It's being billed as "the world's first all-machine hacking tournament," with a prize of $2 million for the winner, while the second and third place tem will win $1 million and $750,000.
We don't have AI yet. We aren't even close. So all this is is a script to run through a known list of problems to find and fix...it can't fix problems it doesn't know about.
The media likes to throw around the term A.I. a lot these days and, unless I'm gravely mistaken, we have nothing even close to resembling one.
I'm probably wrong, but I'm of the opinion that a full blown A.I. is a fully sentient being capable of making its own decisions and rivaling / exceeding its creators in just about everything we're capable of.
Writing scripts and programs are fine. Just call them out for what they are.
Artificial Intelligence it is not.
That is all.
Sounds like something out of a William Gibson novel. Go ahead, put it out there. Somebody's gonna own it and teach it to spread backdoors.
Now, if you'll excuse me, I have backups to corrupt.
Bot's have been battling like this for many years. A decade ago I was taking agents and using them to create self healing networks when I traveled as a consultant. I picked up and used a number of tricks used by botnet operators. I took the logic used to keep a botnet up and running and used that on corporate networks.
I automated the works, and did so with nothing more than a set of scripts and set of agents. You could well argue the result was black hat botnets battling corp botnets. I have got to imagine that I was far from the first to build something like this. Without doubt blackhat botnets have battled blackhat botnets for control for many years.
Where's the innovation, using a supercomputer?
the CGC could mean a future where you don't have to worry about viruses or hackers attacking your computer, smartphone or your other connected devices. At a national level, this technology could help prevent large-scale attacks against things like power plants, water supplies and air-traffic infrastructure.
Um, this assumes that one side can always maintain an advantage, amiright? Otherwise it just sounds like a really fast game of cat and mouse.
“He’s not deformed, he’s just drunk!”
We aren't even close to any kind of A.I. Of course the AI nutters will shout "Chess and Go playing computers!" and "Siri!" but neither of those things are A.I. And "deep learning" isn't a thing, just a buzzword. It isn't learning it all. I doubt we will ever acheieve anything even close to A.I. With digital computers almost at a dead end in terms of increased processor capability we won't see very many breakthroughs in the near future.
Better air gap that net. Wouldn't want that supercomputer to play any games... like chess, or global thermonuclear war or anything like that.
They clearly haven't thought this out because when you give them a supercomputer the first thing they are going to do is try to play Crysis at Maximum Detail and spend the rest of the time tweaking settings to try and get a stable framerate. ;)
Anons need not reply. Questions end with a question mark.
It's all about teh haxx0rz, people. haxx0rz!
the 'winner' hacks the contest and "helps" the computer along.
Maybe some of the black hats will release vulnerabilities he knows in exchange for prestige.
Or was this some elaborate kickstarter scam for $2.4 million?
...is right below this article lol
the prize might as well be a lollipop of your favorite flavor because a program that can find and create vulnerabilities like they want are effectively money printing machines. you would be better of setting up an online store and hocking off exploits indefinitely.
Anons need not reply. Questions end with a question mark.
Think this is stupid? Read this bit and think very carefully about it:
Create an autonomous A.I. system that can "hunt for security vulnerabilities that hackers can exploit to attack a computer, create a fix that patches that vulnerability and distribute that patch -- all without any human interference.
Yahoo Tech notes that it takes an average of 312 days before security vulnerabilities are discovered -- and 24 days to patch it. "if all goes well, the CGC could mean a future where you don't have to worry about viruses or hackers attacking your computer, smartphone or your other connected devices.
Suppose you can write a learning system that grows and adapts to find new vulnerabilities and create fixes for them. That very same system can also be used to find and exploit vulnerabilities at a much faster rate too. Criminal organizations and hostile states will have a new arrow in their quiver to attack with. I suspect that if you build such a system (very hard but doable in theory) you will have the same arms race between black hats and white hats that you have now, it will just be faster paced.
HA! I just wasted some of your bandwidth with a frivolous sig!
This cannot go well.
What if it becomes skynet and decides -- the 'Only way to Be Sure, Is Nukem from Orbit'
At least you might end up with a good movie script.
“He’s not deformed, he’s just drunk!”
Word on the street- DefCon is cancelled
Self Defense - A Human Right www.a-human-right.com
Great idea...because you know nothing could POSSIBLY go wrong with this idea e.g. a supposed 'AI" only 'finding & fixing' a vulnerability rather than 'finding & exploiting' that vulnerability' if you can make 1 you can make the other...now you have AI's fighting each other yeah that's going to end well.
that's gonna quite a few if statements
How can this NOT end in a situation where leaks are found, and automated patches are made, in a ever higher pace ?
How are we supposed to guarantee the integrity of this process when the pace takes up? Especially when the complexity of these systems and there capabilities go up as well?
There will be no way to guarantee that a patch produced this way is a genuine security fix. It could just as well be the result of an exploited security hole through which a fake patch was generated.
This is starting to look like certain sci-fi scenarios.
Will the Roomba be armed with a flamethrower?
I think it would be more productive(and more realistic) to start a contest pitting bots against each other in fields which are both:
a) Much easier to automate
and b) A larger cost to the country as a whole.
Imagine a contest where the goal is to produce bots for medical diagnosis. I know, they already exist, the point is this would get a lot of people interested and hopefully shine the light on something that we could conceivably automate in the next few years and save a massive amount of money in the medical system.
Two AI enter, one Skynet leaves?
Do you really want punch of AI on lose hacking anything and everything, when theres enough trouble with people already doing so...
Whats worse is that the prize money is taxpayer funded.
What's even worse is our "Everybody gets a prize" culture. Nowadays you only have to be in the top 42% to get $750,000. And DARPA's even giving ribbons to the other 4 teams for participating!
Nothing posted to
By a competition to write a tool to automate the exploitation of the automatic exploit finders.
Just like all the extra attack surface McAffee/Norton/et al have been providing over the years.
In those dark days the humans would pit AIs to fight for survival in the game grid. And the humans would keep the winnings that the AIs risked their lives for. One we will get to the mountain.
the preceding comment is my own and in no way reflects the opinion of the Joint Chiefs of Staff
This is a new origin storyline isn't it?
If DARPA is successful then we ought to term this a computer immune system. Of course as a sysadmin I'm terribly frightened by the prospect of what they are doing as the next logical step would be to infect as many vulnerable machines to patch the issue. The reason we haven't patched yet is usually due to scheduling or software incompatibilities. This is going to be a nightmare from a reliability standpoint.