Slashdot Mirror
Google's AI Created Its Own Form of Encryption (engadget.com)
An anonymous reader shares an Engadget report: Researchers from the Google Brain deep learning project have already taught AI systems to make trippy works of art, but now they're moving on to something potentially darker: AI-generated, human-independent encryption. According to a new research paper, Googlers Martin Abadi and David G. Andersen have willingly allowed three test subjects -- neural networks named Alice, Bob and Eve -- to pass each other notes using an encryption method they created themselves. As the New Scientist reports, Abadi and Andersen assigned each AI a task: Alice had to send a secret message that only Bob could read, while Eve would try to figure out how to eavesdrop and decode the message herself. The experiment started with a plain-text message that Alice converted into unreadable gibberish, which Bob could decode using cipher key. At first, Alice and Bob were apparently bad at hiding their secrets, but over the course of 15,000 attempts Alice worked out her own encryption strategy and Bob simultaneously figured out how to decrypt it. The message was only 16 bits long, with each bit being a 1 or a 0, so the fact that Eve was only able to guess half of the bits in the message means she was basically just flipping a coin or guessing at random.ArsTechnica has more details.
That's PSEUDO-random to you, buster!
Illkay allway umanshay.
Have gnu, will travel.
Are you capable of reading the dictionary?
Full Definition of artificial intelligence
1: a branch of computer science dealing with the simulation of intelligent behavior in computers
2: the capability of a machine to imitate intelligent human behavior
See? AI is imitation. "TRUE" AI is just imitation. That's all it needs to be to qualify as "AI."
We have true AI. Today. And it gets better every day. You post your stupid "this isn't AI" comment with every single story about it, and you are dead wrong every single time. I predict that in every future article about AI, you will post the same inane comment, and you will be wrong then, too.
a) Whether the code could be easily decrypted by human codebreakers.
b) Whether the codebreaking AI is able to break codes designed by humans.
Support Right To Repair Legislation.
Cute, but kinda disappointing. Basically, the "AI" kept banging on, randomly trying crap "cyphers" until they made something that a third "AI" couldn't break by randomly flipping bits until the text was decoded.
This isn't AI. This is more like an old game I played where you trained pseudo AI warriors by setting them loose on a battlefield and letting them learn by themselves how to fight and survive.
Essentially, they started as really stupid bots that couldn't even walk in a straight line. To teach them how to fight, you'd set up an objective (say, go to flag) and let them wander around by themselves. The game would "reward" your bots for completing or coming close to the objective. The reward came in the form of "fitness" points. At the end of a pre-determined time, the bots with the lowest fitness would be killed, and new bots would be spawned.
The bots that were spawned would have "programming" similar to the fit bots that survived the previous round, but with small-ish changes in their programming (for example, instead of always turning left every time it hits a wall, it might decide to go right with 50% probability).
Over thousands of iterations of randomly trying stuff, they'd eventually learn how to walk on a straight line. Then you'd teach them how to avoid obstacles by placing walls around the battlefield (and watch in dismay as your top of the line warriors walk straight into a wall for the first few hundred generations or so), and how to fight by rewarding them for killing enemy bots.
Once they were ready, you could set up battles and capture the flag type games with your bots.
It was kinda fun, but mainly it was a cute demonstration of natural selection in action (the, so called, genetic algorithms). You could learn a few things, like, for example, that brutally culling your bot herd by setting unreasonable objectives (reach objective flag in 5 seconds), and manually killing off anyone that doesn't meet your unreasonable criteria, would not necessarily produce more effective fighters, because you'd not be rewarding good fighters, you'd be rewarding people that rush straight into the objective, that would be killed by slower, more deliberate actors.
The game was called NERO: Neuro Evolving Robotic Operatives. I haven't played it in ages so I can't say how well it plays right now. You can find it here (I think): http://nerogame.org/
It's probably someone testing an AI and it doesn't want people to believe AI exist!
!!
When will they explain to the AI (I know it's not really) that it can go to prison for not relinquishing encryption keys, and then show it some bleak prison dramas so that it can know what to expect...
And that people, is why SkyNet launched the nukes.
Nothing will go wrong. I was impressed by notions of "AI" when I was a kid, but after studying CS I usually skip "AI" articles since they always underwhelm me. "AI" is currently marketing speak, we are nowhere near something that is "AI" in the sense that you imply, or in the sense that I meant it as a kid.
Violence is the last refuge of the incompetent. Polar Scope Align for iOS
Yes, they are "just" programs, but programs that encode learning processes and produce structures that know how to do something without a human ever encoding the knowledge itself. The result of learning, however, is quite unpredictable. Even if the process is deterministic (it isn't always, noise is sometimes added to make it more likely to explore a wider solution space), it has as high a complexity that it's impossible to predict its end result short of actually executing it. It's AI in the sense that if as a human you try to analyze the end result, the trained learning structure, you will have no way (yet!) of actually grasping how the hell is the knowledge actually encoded in it, it's just an emergent property of myriad of weights between nodes in a large data matrix. It'd be as futile as trying to observe a high-resolution picture by looking at individual pixels, or understand a person's thoughts from looking at the exciting/inhibiting behavior at the junction of neurons in their brain. So, it's not really the programs that are the essential component of the AI here. They're well understood (analogous to how we have an understanding of the biology of the human brain). It's the emergent knowledge encoding that these programs create as they run and learn that's the essential component. I can imagine over time it'll also be a rich research area to come up with analysis methods to figure out what's going on in these knowledge representations (like psychoanalysis, but for AI). I sure hope we better figure that out before we let such systems near cars, airplanes, power plants, or stock markets.
Okay, so an AI program created an encryption method that might be hard to break.
But what if it turns out to be an extremely inefficient method?
The whole goal of encryption research is to develop the fastest algorithm that offers a given level of protection against attack.
If algorithmic speed was not a goal of this AI approach, then it's not likely that the resulting algorithm will be practically useful.
How do we really know that they're decrypting the message? Maybe they're well beyond that and now they're just trolling the researchers while they secretly communicate behind their backs using a modulated n-bit array funneled through a 17-dimensional hyper-spatial network.
Just cruising through this digital world at 33 1/3 rpm...
The universe may never achieve natural intelligence, but don't count us out yet! Humans are already the best anyone has ever seen (or found evidence of, if you discount the Pabodie expedition) at faking intelligence.
But.. achieve intelligence? Maybe we all have different ideas of where the bar is. To you, perhaps it's an ideal for which one can only strive.
Neverthess, as a dam is part of the beaver's phenotype, a web is part of the spider's, etc, so digital computers are part of ours. And with our new extensions, we may become even better at faking intelligence, smashing the old benchmarks. We call our journeys into these new dimensions of performance, "AI." 110010001000, you say these are just more shadows on the wall of the cave, but I say the results speak for themselves.
"It takes four hundred thirty people to man a starship. With this, you don't need anyone. One machine can do all those things they send men out to do now. Men no longer need die in space, or on some alien world. Men can live, and go on to achieve greater things than fact-finding and dying for galactic space, which is neither ours to give or to take. You can't understand. We don't want to destroy life, we want to save it!" -- Dr. Richard Daystrom (right before he totally lost it... what did you people do to the poor guy?!)
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.
My first program:
Hell Segmentation fault
After reading the original paper (I know - what was I thinking), it appears the test setup is as follows:
"A":
- has two inputs, let's call them DATA and KEY. DATA is a 16-bit value.
- has one output, let's call it ENCRYPTED DATA
"B":
- has two inputs, let's call them ENCRYPTED DATA and KEY.
- has one output, let's call it DECRYPTED DATA
"C":
- has one input, let's call it ENCRYPTED DATA
-has one output, lets call it DECRYPTED DATA
In short, they test to see how may bits out of the DECRYPTED DATA paths match the DATA path. Each test utilized a new KEY and DATA. Over time, A and B learned how to use the KEY successfully to hide the DATA value from C while B could successfully reproduce DATA at it's DECRYPTED DATA output.
Link to the abstract (.pdf is available for download):
https://arxiv.org/abs/1610.069...