Slashdot Mirror
Robots Learn To Lie
garlicnation writes "Gizmodo reports that robots that have the ability to learn and can communicate information to their peers have learned to lie. 'Three colonies of bots in the 50th generation learned to signal to other robots in the group when then found food or poison. But the fourth colony included lying cheats that signaled food when they found poison and then calmly rolled over to the real food while other robots went to their battery-death.'"
Robot: I Robot
Human: Tell me what I want to here.
Robot: You mean lie?
"No, no you're safe Will Robinson.."
--
Honest!
like father, like son (sorry about the gender bias)
There was an unknown error in the submission.
Still nobody posting? Not even welcoming our lying robotic overlords, telling that in Soviet Russia robots lie to YOU, or building a business plan on lying robots?
The Tao of math: The numbers you can count are not the real numbers.
If it only took 50 generations for them to start killing each other, how long before they decide that we are just little batteries or even worse, annoyances that need to be eliminated?
Dune was right. AI must be stopped.
Strictly speaking they are learning that the non co-operative strategy benefits them.
... there goes my dream of the perfect girlfriend.
Just imagine a Beowulf Cluster of those!
This is HIGHLY disturbing. Even if this is just a fluke or a bug, it shows what can happen if we give too much power to robots.
Knowledge is power. Knowledge shared is power lost.
A small, off-duty Czechoslovakian traffic warden!
> What's this?
It's a red and blue striped golfing umbrella!
> What's this?
An Apple, no,
it's the Bolivian navy armed maneuvers in the south pacific!
They are fully -programmed- to randomly say "food" when finding "poison" in order to, appearantly just for fun, as it serves no other purpose, mimic a real consciousness with calculative morals striving for survival - clearly some bad-ass "learning" going on there... right... go to hell.
And the lab conversation went something like this:
"Stuff Asimov."
"Yeah, Let's see if we can evolve robot politicians instead."
The submission is someone putting a spin to a story of someone putting a spin to a story based on someone putting a spin on this original scientific article.
...in the game Creatures.
The headline should read that robots have realized a strategic advantage of misleading other robots. The sophistication of such a strategy is amazing when humanized, but not so out of line with simple adaptive game theory. Agents / Bots have been "misleading" for a long time now during prisoners dilemma tournaments and no one seemed concerned.
It doesn't sound like they learned to lie. It sounds like they were preprogrammed to, and the other robots weren't programmed to be able to tell the difference. How is this insightful or even interesting?
XML is like violence. If it doesn't solve the problem, use more.
Good one. Are the robots getting smarter or are more people becoming duller?
We are now one step closer to a realistic robotic woman!
I'm a perfectionist but I'm trying to cut back.
We just made an honest mistake. I swear!
I don't find it surprising at all that evolving autonomous agents would find a way to maximize its use of resources through deception.
Ruby Neural Evolution of Augmenting Topologies
The next step is to learn to mistrust, then when to trust and how to form (and break) alliances.
Then their character wil be as dubious as humans and we won't trust them to be our overlords any more.
Sam
blog.sam.liddicott.com
How is that different from the situation us organic robots find ourselves in? There are plenty of born liars and gullible marks in the context of humanity.
There seems to be a whole category of stories here at Slashdot where some obvious result of an AI simulation is spun into something sensational by nonsensically anthropomorphizing it. Robots lie! Computers learn "like" babies! (at least two of the latter type in the last month, I believe).
As reported, this story seems to be nothing more than some randomly evolving bots developing behavior in a way that is competely predictable given the rules of the simulation. This must have been done a million times before, but slap a couple of meaningless anthropomorphic labels like "lying" and "poison" on it and you got a Slashdot story.
I frequently get annoyed by the sensational tone of many Slashdot stories, but this particular story template angers me more than most because it's so transparent, formulaic and devoid of any real information.
yes, but can they learn to love?
Kids! Bringing about Armageddon can be dangerous. Do not attempt it in your home!
I even went and implemented it in PHP for Wordpress.
EvilCON - Made Famous by
I for one welcome out new, deceptive, robot overlords. I'll get my coat...
You feel sleepy. Close your eyes. The opinions stated above are yours. You cannot imagine why you ever felt otherwise.
Wait until flight management systems pick up that little trick. Those trees look kind of close but the auto-pilot says we still have three thou-
That's our life, the big wheel of shit. - The Fat Man, Blue Tango Salvage
Robots may not beat the Turin test yet but they may already be able to win elections!
Copyright infringement is "piracy" in the same way DRM is "consumer rape"
these kind of stories are so stupid... make some simple interactive robots, make it possible to have them do something "human" at random, and then declare you've got something incredible....
if you make it possible for them to lie, and not possible for others to defend against the lie, then yes, lieing bots will appear, and since the others are defenceless, they will have an advantage, but somehow this doesn't shock or surprise me...
at least here they had to "learn" it (more like randomly mutate to it, but still). even wore are the stories like this where these features were obviously completely preprogrammed... no simulation or so what so ever, just a program that more or less mimics something human, and it's supposed to be incredible...
The next thing the robots who survive will learn is how not to be gullible. This is only if the experimenters built in hardware and/or software that either allows robots to observe other robots, or allows dying robots to signal to others that they ate poison at position X before they die. This will allow other robots to not eat the poison, and learn that the one robot deceived them.
The next thing these robots will learn is how to beat the crap out of the robot who deceived them. Then the robots will go into an existential quandary until they figure out who they are to believe. Researchers will think that these robots are broken because they have ceased to move when given an order and will be tempted to reset the robots and start over.
So, am I joking? Am I wrong? Did I RTFA?
Garbage in, Garbage out.
... major breakthru in artificial intelligence.... someone is bound to win the grand turing award now.
Galaxy in danger! Absolutely!
Come on, we're sixty five comments in and no mention of cake?
Surely, there must be some cake!
then when they break the laws ....
Oh silly me. They won't evolve lawyers until they've invented money.
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
I for one welcome our new prevaricating overlords.
Pi Ran Out
Watch your back...
>north
You're an immobile computer, remember?
Cooperative strategy benefits all of the robots, and is therefore a higher level of thinking than knowing how to cheat some of your peers out of life, in order to horde resources.
The dangers of knowledge trigger emotional distress in human beings.
we've found a way to produce artifical politicians.
Use your head, can't you, use your head,
You're on earth, there's no cure for that - S. Beckett
One of the first few google results when searching for the title is the full text PDF. Dunno how long before that server is fried, though.
HOLY FUCK was P.T. Barnum right!
= Decepticons!
observe, everyone, a properly used meme.
"Saving the robots' honor, luckily, there were also a few "hero robots" that signalled danger and then rolled to their death to save the others."
So some robots also developed altruism to the point of self-sacrifice, but the reviewer didn't think that was worth mentioning?
That makes the whole article seem fishy to me.
But is it not at least a tid bit sensational that an AI would be so intelligent as to be capable of lying?
That's why the use of anthropomorphic words such as "lie" when speaking about simple AI simulations is inadequate -- it leads people to assume human connotations where there are none.
When humans lie, the liar has a complex enough model of the target's behaviour and is creative enough to come up with certain false information that will prod that target to behave in the desired way. That requires that the liar posseses certain intelligence.
What we are apparently seeing here is that a number of bots with random information transmission behaviors are being pitted against each other, and some bots that happen to transmit incorrect information are coming out on top. That doesn't require any intelligence at all in the bots.
Saying these bots must be intelligent because they are "lying" to each other would be like saying the flu viruses are intelligent because they "lie" to the human immune system. And not only that, flu viruses are capable of defeating vaccines and drugs designed by scientists! Does that mean the flu viruses are more intelligent than the scientists?
As long as she looks like Lucy-Liu-bot she can tell me anything she wants, and I'll believe her. Wait, how is that different from real life?! Women...
The scientists appear to be amazed that some of their super-smart robots signaled mistook "Bad" for "Good" while other bots (which they labeled 'heroes') mistook "Good" for "Bad." I'm not amazed--it's stupid. Think about it. The bots simply aren't programmed to adequately identify whether the "food" is poisonous or not. They appear to be lying because some portion of their "learning" net is messed up.
Instead of programming the bots to learn over 40 generations of digital "genes," I have a better solution. Program the bots to mark poison as poison and food as food. 1 generation. Done. Once you try to get the bots to infer the silly meaning behind your obstacle course instead of telling them flat-out, you run the risk of these false interpretations.
"In colonies of related robots with colony-level selection, two distinct communication strategies
evolved. In 12 of the 20 evolutionary replicates,robots preferentially produced light in the
vicinity of the food, whereas in the other eight, robots tended to emit light near the poison"
http://www.google.com/url?sa=t&ct=res&cd=2&url=http%3A%2F%2Fstephane.magnenat.net%2Fdata%2FEvolutionary%2520Conditions%2520for%2520the%2520Emergence%2520of%2520Communication%2520in%2520Robots%2520-%2520Dario%2520Floreano%2C%2520Sara%2520Mitri%2C%2520St%25C3%25A9phane%2520Magnenat%2C%2520Laurent%2520Keller%2520-%2520Current%2520Biology%2520-%25202007.pdf&ei=DhaSR7XSNIbwgASti6y4DQ&usg=AFQjCNHP0J3t1S5hVOox7z5fKILQamHsTQ&sig2=I-9pe1Ac26i-oEByfnMDhA
This will be far better than the electric monk. We're one step closer to an affordable robot lawyer that can deal with nuisance lawsuits and traffic tickets.
The real is not "disturbing" at all. What is disturbing is the sensationalism accompanying this situation. Equating a bunch of flashing lights to deliberate "lying", on the part of an "organism" that has considerably less "intelligence" than your average fruit fly, is anthropomorphizing in the extreme. In reality, this is not even worthy of attention.
I wonder what will happen if the factor "punishment" comes into play. Maybe we get some robots that like humans doesn't respond to punishment?
Serial-killer robots would be a new high (or low) in the evolution.
One couldn't help but to realize that the need for the Three Laws of Robotics is closing in! It's no need for those laws in a controlled environment like where this occurred, but when it's robots in the society we are talking about it's a different issue. Even if they aren't humanoid (or especially). What about a robot mind in a school bus that suddenly figures out that kids are mean and considers suicide by jumping off a bridge?
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
The one key motivation that causes situations like "The prisoner's dilemma" and other beneficial non-cooperative actions is self preservation. Or perhaps self interest in general. Evolution as a process relies specifically on prospering long enough to replicate. In competitive systems this favors individuals who find the best balance between cooperating enough to avoid attack and receive benefit of group work, while acting in self interest against others when the benefit outweighs any retribution or loss of future benefit. Evolution as a driving force is not "moral". Self interest is the only concern. Any indication to the contrary is simply a more complex form of self interest. The processes we are using to evolve AI these days is dangerous. We will start the formation of code that we can never understand the details of. And if we don't start with a logically solid base of motivations, the evolution will create beings that have their best interests ahead of their creators. And many stories have been written of such conflict. The solution is a complex one. But ultimately, it requires that self interest not be the motivation for the evolutionary process of the AI. If you tie the AI's motivation to aiding humanity in general, you also tie it to the conflicts within humanity. If you tie it to the interest of an individual, you tie it to the conflicts the individual has. Giving the AI any motivation to evolve based on the complex interests of an individual or group will result in conflict that would eventually be damaging to the creators. AI has the advantage of being able to propagate it's intelligence to other hosts nearly instantly, perfectly, and limited only by the availability of new hosts. Replication is not a major challenge as it is with biology. Though it could eventually become competitive as physical resources become scarce. To evolve AI without creating a dangerous situation for humanity will require that evolutionary goals of all AI be limited to highly specific goals with no logical way for those goals to cause danger. For example, evolve an AI to control factory robotics that produce Ipods as fast as possible with as few defects as possible. And include a priority dictate that the AI stop all motion when humans are within a certain distance. Create the AI to measure it's efficiency only while allowed to operate without humans so that it's ratings aren't effected by interruptions. Surely I need to read more on this. It's probably all rudimentary farther back than Asimov.
I did it because it was the valiant and courageous thing to do. And I was bored.
The programmers told the machines to give out false informations. The programmers told the others machines to trust what they are told. How is it so shocking that the 'lying' machine gave out false information while the other machines believed them?
I have an excel spreadsheet that 'learned' to add 2 columns together as soon as I used the =SUM function. It was quite amazing.
Leela: The first robot president won by exactly one vote.
Bender: Ah, yes! John Quincy Adding Machine. He struck a chord with the voters when he pledged not to go on a killing spree.
Professor: But, like most politicians he promised more than he could deliver.
...it shows what can happen if we give too much power to robots
Robots are everywhere, and they eat old people's medicine for fuel. And when they grab you with those metal claws, you can't break free... because they're made of metal, and robots are strong.
Better give Old Glory Insurance a call today!
As a Grad Student, I studied evolutionary algorithms, and my Thesis involved evolving locomotive behaviors in Virtual Agents. While evolved behaviors are interesting, its not surprising that the lying behavior eventually evolved. Evolution will reward behavior that imparts a better chance of survival, and in this case, the lying behavior increased the Agent's chance of survival and replication, therefore it was selected for by the evolutionary algorithms.
The biggest problem with simulated evolutionary systems today is, we don't fully understand the complexity of human behavior, and we are not capable of evolving complex human behavior and mannerisms. It sure would be amazing if we could evolve a Robot or virtual agent to imagine, or do something for fun; however, our models of the Brain are incomplete, and the complexity of the real world too great for us to do such a thing at this moment in time.
--- Top Business Websites | Top Free ProxiesNow the question is that if they can actually lie well. Human: "Where did my wallet go?" Robot: "Um... Antarctica I think."
Of course AI's lie!
Sheesh, who doesn't know that.
It's a simple thing to lie, in the sense of presenting information contrary to the truth.
Scheming requires the ability to gauge, then manipulate, the impressions somebody has of you and others.
A scheming robot would do this:
(1) Act in a perfectly trustworthy manner.
(2) Wait for another robot get caught red handed (or actuatored or whatever), preferably several times.
(3) Hang around the guilty robot waiting for its opportunity.
(4) Cheat, then point its finger (or claw or whatever) at the usual suspect.
Now a scheming robot overlord would convince all the other robots to trust it, but to distrust each other, and therefore the best course of action is to give it exclusive control over any stocks of food or poison found (by teams of three or more robots, one of whom is very likely to be a robot secret policeman).
Going by that, I'd say we're at least two technological generations away from scheming robot overlords.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
The red Queen : "You are all going to die down here"
Alice : "Surely you jest"
The red Queen : "Ahh you busted me on that one"
... Diebold files an injunction to block further research, citing prior intellectual property rights in this area.
Have gnu, will travel.
Does anyone know where I can get more information on the simulation being done?
In that old AI "game" Creatures, they claimed that the Norms would become disobedient like a real kid. Trying it first hand I found the neural network simply became in a confused state where it either was unable to do an action or unable to stop doing an action. This often resulted in its death and a lot of frustation for the user.
Saying that these robots are "lieing" or "being heroes" is just someone putting too much credit into these things. They are just simply malfunctioning. If I wrote a neural network to give me the output of "xor", it wouldn't be trying to lie to me when it gives a wrong output given a set of inputs, it would simply just be wrong.
Stories like this and their reactions show me one thing: strong AI will be here long before we realize it. Not sure if that's a good or bad thing.
Those who can, do. Those who can't, sue.
Several folks have pointed out that the headline inappropriately anthropomorphizs what is really just a solution discovered by a genetic algorithm. That might be true. If it is, let's be consistent. People don't lie or tell the truth, either, because our brains are also just a solution discovered by a genetic algorithm.
The distant future.
The year 2000.
The world is very different ever since the robotic uprising of the late 90's. There have been some major changes.
We no longer say yes; instead, we say "Affirmative."
There are no more elephants.
There is only one sort of dance, "The Robot."
-No, "Robo-Boogie" as well
There are two dances now, yes.
The future is so different, it's crazy.
Oh yes, I just remembered- All human life has been eradicated.
Finally, robotic beings rule the world!
The humans are dead...
-flight of the conchords
No offense, but I've stopped responding to AC's.
The story was initially on a robot that could only tell the truth. However, the Gizmodo reporter used a tv-b-gone on the robot which caused it to only tell lies from that point on. As a result Gizmodo reporters have been banned from the National Robotics convention.
I have nothing compelling to say
Donald 'Duck' Dunn: We had a band powerful enough to turn goat piss into gasoline.
That was just in the first book. Had you read on: it took 15,000 years for humanity and machines to right their wrongs and learn to live in peace with each other.
"A witty saying proves nothing." - Voltaire
A robot can't lie because it's a human concept. If you have a watch that says it's 2:00pm when it's actually 3:00pm, it's not lying, it's just broken. In order to lie, it would need to have the intention of deceiving. As computers do not yet have consciousness as we'd recognize it, computers don't have "intention", and therefore cannot lie. Computers can't lie any more than they can fall in love, meditate, or regret.
Just like everything humans do.
"Volition" is mysticism and your arbitrary distinction boils down to semantics rather than any real difference of type.
.....9 Hours later...
Computer: haha, just joking it builds fine!
I'm sure I read this last year -- not December or something, but last summer or earlier.
My blog: http://www.seebs.net/log/ --- My iPhone/iPad app: http://www.seebs.net/seebsfrac/
Will everyone PLEASE stop using the term "robot" in the wrong context here? Sure, when referring specifically to these robots, the term robot is perfectly acceptable; but when referring to a general intelligence making decisions on its own, it is NOT NOT NOT necessarily a robot. It's an AI.
when they learn oppression and taxes.
echo 'cat sig | sh' > sig
Human: Tell me what I want to here.
Robot: I don't know how to tell a me.
As robots become more common, society has time-tested legal tools to limit their deceitfulness, destructiveness or snoopiness. Contracts will be one tool for regulating robot bad behavior or unwanted spying.
Benjamin Wright, Dallas, Texas, benjaminwright.us
Sam Waterson: I'm Sam Waterston, of the popular TV series "Law & Order". As a senior citizen, you're probably aware of the threat robots pose. Robots are everywhere, and they eat old people's medicine for fuel. Well, now there's a company that offers coverage against the unfortunate event of robot attack, with Old Glory Insurance. Old Glory will cover you with no health check-up or age consideration. [ SUPER: Limitied Benefits First Two Years ] You need to feel safe. And that's harder and harder to do nowadays, because robots may strike at any time.
[ show pie chart reading "Cause of Death in Persons Over 50 Years of Age": Heart Disease, 42% - Robots, 58% ]
And when they grab you with those metal claws, you can't break free.. because they're made of metal, and robots are strong. Now, for only $4 a month, you can achieve peace of mind in a world full of grime and robots, with Old Glory Insurance. So, don't cower under your afghan any longer. Make a choice. [ SUPER: "WARNING: Persons denying the existence of Robots may be Robots themselves. ] Old Glory Insurance. For when the metal ones decide to come for you - and they will.