Robots capable of 'deceiving humans' built by crazed boffins Worrying news from Georgia, America, where boffins report that they have developed robots which are able to "deceive a human". "We have been concerned from the very beginning with the ethical implications related to the creation of robots capable of deception and we understand that there are beneficial and deleterious aspects …
Hardly crazed - more like taking the next logical step.
Computers tell us lies all the time. Whether it's "20 spaces free" at the local multi-storey or "you don't owe us any tax" to take a contemporary example. So when you put wheels on a computer and call it a robot, there's really no difference.
As it is, most people are incredible easy to deceive (if that makes sense) and are willing to believe pretty much anything they read, hear or see on a computer screen - provided they want to believe it. So maybe what we really need is a magic mirror that lies fluently when asked "does my arse look big?"
After all it's not the computer / robot that's deceiving us, it's our own willingness to accept the lies we are told.
""We strongly encourage discussion about the appropriateness of deceptive robots to determine what, if any, regulations or guidelines should constrain the development of these systems," adds the prof."
Hmmm.... that then would be a discussion to discuss whether they should be less like humans with an intelligence system which spins lies rather than sharing truths for control of the environment.
And regulations and guidelines are only for robots and do not ever apply to free radical/fundamental base humans with the capacity of original thought and/or remote programming of robots masquerading as human beings.
In summary, a team designed a robot that lays a false path and then goes off elsewhere. Same team designs another robot that's designed to follow this path of destruction unquestionably. Team is delighted to find that the hunter robot (that they designed) cannot find the 'deceptive' robot (that they also designed). That's amazing...
For 300 million Americans, GIT means the Georgia Institute of Technology, which is colloquially known as "Georgia Tech", and is part of the phrase "the Rambling Wreck from Georgia Tech". It usually ranks as a "low end of the top tier" science and engineering school, behind MIT, CalTech, Carnegie Mellon, and a few others. But quite respectable.
For 60 million British, it means he's an idiot. I'll give you some Aussies too - say 65 million.
And then there is the capitalization of the word...as if being outvoted wasn't enough... ;-)
My printer is already deceptive, it tells me i'm low on ink even if i'm not, It tells me i'm low on paper even if i'm not, and it constanlty tells me it's jammed even when it's not.
Don't buy a HP/Skynet printer
Grenade, because that's the only way to stop it from printing
Two customers come to mind: The Pentagon and Apple.
The Pentagon could deploy these, alongside their fleets of drones, to deny what happened, actually happened. 100% deniability!
As for Apple both the PR crowd could use them to create believable illusions and Customer Service section to deny defects that exist in fact, are actually the customer suffering from delusions.
There will ultimately be a ShouldIDeceive() function in the robot brain that will determine whether or not to deceive.
This will be human written, and the associated coding will be human written.
The device itself will take all the inputs and work out whether or not to call DeceptiveBehaviour() or HonestAction().
Once you put in all the "independent" thinking, it is essentially acting by itself.
The deceptions themselves will also be human coded though
They didn't say how they actually trained the robots - neural nets, genetic algorithms, or whatever. Assuming they used a genetic algorithm (likely), what this shows is that such a maximizing algorithm will train itself to incorporate deception, as long as the training scenarios are not constrained. Because, simply, it WORKS! It begins to find global maxima of it's fitness functions using deceptive routes...
Now, that really DOES have implications. It is very, very difficult for a human to look at a neural network or a genetic algorithm function and understand what it actually DOES, and under what conditions. All we know is that it maximizes the output fit for a given set of inputs in the training data or experience base. We actually have to observe it in operation to have some idea how it works (for any sufficiently complex matrix or functions).
Case in point - GAs were used to design the compressor turbines for the jet engines of the Boeing 777 - and the GA engineered a design which eliminated an entire set of compressor blades, and was the most efficient. Something that the human engineers had never been able to do, and had significant difficulty in understanding how it had done so, even when they looked at the design. But it worked, and those 777 engines are all the better for it.
But this could be the opposite - we could be training robots that reach globally maximum functions that, frankly, do so with no "morals". If those robots can lie, cheat, steal, even kill...well, unless there is a penalty for that in their training function, they WILL - because it is the most efficient manner of operating.
So, what these esteemed professors have shown is that unless we develop training functions with HUGE negative impacts for immoral behavior, our robots will train themselves to emulate your basic Colombian drug lords in behavior. Interestingly, there are a fair number of people who turn to crime even WITH society showing large penalties for it - and I fear that if the robots assess the probabilities they might come to the same conclusions.
Asimov was right...
Much the same can be said for children. Society has thousands of years of experience of how to train "learning units" to behave morally and we're pretty good at it.
If we ever did create a machine capable of acting like a human, it would have all the same flaws. It might even be "mortal" in the sense that after a century or so it became fixed in its mindset an unable to adapt to changes in the society it lived in, eventually becoming so depressed that it flipped its own Big Red Switch.
Don't believe me? Well, build one and prove me wrong. Until then, spare me the scare stories you watched when you were little, written by people who hadn't (and still don't) have a clue about what actually makes us human.
Except that we know how to police and reform humans - there are very key differences when it comes to robots.
Of what threat to a robot is time in jail? Can a robot even feel "mortal" and worry about it's own death as a sanction against crimes committed? If it lacks true consciousness, can it worry about losing it?
Can a robot feel pain? Can you "spank" a robot?
Of what use to a robot is group therapy, "getting it's life back together", agreeing to conform to human norms? How would such be accomplished? Can a robot "find religion" and repent? Can a robot repent without religion?
And I don't have to prove anything - the whole POINT of the article was that they already HAVE built robots that have learned to deceive as part of their programming. QED...my post was to state how to consider fixing it technically...
One does not "try and" do something.
You try TO do an action.
'I tried to get in to the cinema'
'Did you get a discount?
- No, but I tried to.'
See the following reference from Paul Brians, professor of English at Washington State University, in his book 'Common Errors in English Usage':
http://www.wsu.edu/~brians/errors/try.html
Asimov only said that a robot must nor harm, or by inaction allow harm to, a human. And sometimes the truth hurts. Although, in "Liar!", not as much as...
There's also deception in the Asimov story described here,
http://en.wikipedia.org/wiki/Satisfaction_Guaranteed_%28short_story%29
where a human-looking robot poses as a downtrodden housewife's ideal lover to raise her social standing with her neighbours, who don't realise that he's a robot (and there may be a problem with that guarantee).
The first Robot is just following orders, which is to knock down some markers, then movee in another direction. The second robot is just following it using a simple path-estimation routine.
For the first robot to 'lie' it must be willfully deceiving the other.
It's not...
That would require a real AI.
Human: "so robot, i'm afraid we have some rules, you can lie when told to by an authorised human, but you must tel autorised humans the truth, ok?"
Robot: "Yeah sure, sounds fair to me"
Human walks away thinking "Well, that was easy"
Robot drives away thinking "Moron".
Philip K. Dick wrote a short story in the 1960s about a robot which could camouflage itself as a TV set, sneak into people's homes, commit murder, and then leave evidence at the crime scene to frame some innocent human being or other. I forget the title, but it's in 'The Golden Man' collection.
I too go with the 'computers deceive regularly' meme. I subscribe to constructivism, which points out quite scientifically that the evidence of our senses is largely illusory, and any resemblance to reality - whatever that is - is rather coincidental.
Or, to put it another way: Show me a GUI which does not deceive the user, in some important respect.
...however, more sophisticated robotics labs using far more advanced nascent AI will take this to the next step - whether it's prudent or not. You know, so preoccupied with whether they could, they don't stop to ask whether they should.
Marry this with all the frightening research being carried out by the U.S. military - everything from super-human (genetically enhanced) soldiersl; fully autonomous and armed combat robots and drones; cybernetics - and it starts to get very scary, very quickly. These obscene things aren't just dreams of the paranoid, the U.S. military has crowed about all these areas of research and how it will "revolutionise warfare". It's bad enough having people kill people, but really steps over the line when we have machines killing people, making the decisions completely without reference to their human masters.
This has "disaster waiting to happen" written all over it in BIG FLASHING NEON LETTERS. "Terminator" isn't a fantasy, it's very much a prophecy if we continue down this path. We're getting so close to true AI, it's only a matter of time before self-awareness comes. The last thing you want is a robot with no morals, ethics or empathy becoming aware its own existence will be at threat upon successful execution of their mission against "enemy" humans.
"I've killed all the humans marked "enemy" now the rest of the humans want to kill me: therefore all humans are the enemy." Seriously, we don't want to go there.
I think the U.N. should put its foot down and put a stop to this insanity, or at least some unassailable barriers, like Asimov's three rules of robotics, and place a total ban on the use of this technology for anything other than peaceful purposes with enormous sanctions for those who break international law.
Okay, maybe something more threatening than a fluffy bunny slippered U.N. foot is needed to dissuade the U.S. from this path to universal destruction.
I remember reading a site called 3 Laws Unsafe (http://www.asimovlaws.com) launched by the Singularity Institute for AI around the time the "I, Robot" movie was released. Their argument was that Asimov's laws make for great fiction, but in reality would be either unethical to implement in truely intelligent machines, or end up causing more harm than good (think HAL-9000's conflicting commands causing the Discovery disaster in 2001: A Space Odyssey).
And I really don't think this is such a great coup - all that has been demonstrated is machine-machine deception, which, as someone already pointed out, is all in the code. Machine-human deception, as someone else pointed out, depends on how much faith the human in question puts in the output of the given machine (as well as the actual instructions given the machine - i.e. no properly developed and tested ATM software would be installed if it was known to give out false balance information, but if what would we do if we woke up tomorrow to be told by the first ATM we went to that we were plus or minus €10,000 of our last known balance?).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
