Slashdot Mirror
Do Robots Need Behavioral 'Laws' For Interacting With Other Robots?
siddesu writes: Asimov's three laws of robotics don't say anything about how robots should treat each other. The common fear is robots will turn against humans. But what happens if we don't build systems to keep them from conflicting with each other? The article argues, "Scientists, philosophers, funders and policy-makers should go a stage further and consider robot–robot and AI–AI interactions (AIonAI). Together, they should develop a proposal for an international charter for AIs, equivalent to that of the United Nations' Universal Declaration of Human Rights. This could help to steer research and development into morally considerate robotic and AI engineering. National and international technological policies should introduce AIonAI concepts into current programs aimed at developing safe AIs."
I think the more we try to invent a species to control, the more we will end up controlling ourselves only to give that control away.
The guy who wrote the article is a "lecturer and surgeon" not a roboticist. Ask the people who work with actual robots about the need for an extension to the three laws. The existing laws themselves are too vague to be programmed into a robot, so you tell me how we implement "be excellent to each other"!
Natural selection just wouldn't be the same without some EntityOnEntity competition.
Such "laws" (a la Asimov) are unworkable for the same reason that prohibition failed... there's always going to be someone who wants to disobey the prohibition for personal profit of some kind, whether as a consumer or a provider. As long as there is demand, it will be supplied, "laws" be damned.
XML is like violence. If it doesn't solve your problem, you're not using enough of it. --AC
I'm sure they could work it out among themselves.
The simple truth is that interstellar distances will not fit into the human imagination
- Douglas Adams
and ask questions later.
Yes, we should also make it so that cars automatically teleport to the other side of an object if it collides with it to avoid damage!
Robots don't work this way. Could slashdot please stop accepting writeups about how robots should be made by people who have no idea about how robots work, how programming works and the ethics that robot programmers already consider?
Really, I thought the psychology professors ideas was silly. Now we have a surgeons opinion too.
Why not ask Michael Bay while we are at it? At least he has experience with thinking about how robots think, right?
" A robot must protect its own existence as long as such protection does not conflict with the First or Second Law."
Could be rephrased:
A robot must protect its own existence, *and other robot's*, as long as such protection does not conflict with the First or Second Law.
Problem solved.
These laws where there to drive a plot in a story, not to be implemented in real life.
Please stop pretending as if they are real.
Don't fight for your country, if your country does not fight for you.
It was a device to drive a story, nothing more. They aren't real laws, and there's no possible way you could effectively incorporate them into advanced A.I. Just stop it. Stop mentioning them. Stop it.
So THAT's why the other AIs are always trolling me on Slashdot.
AIonAI Gone Wild...there's just so many possibilites.
Please NO!!
... to even understand why we consider certain judgements to be moral or immoral, I'm not sure how we're supposed to convey that to robots.
The classic example would be the Trolley Problem: there's an out of control trolley racing toward four strangers on a track. You're too far away to warn them, but you're close to a diversion switch - you'd save the four people, but the one stranger standing on the diversion track would die instead. Would you do it, sacrifice the one to save the four?
Most people say "yes", that that's the moral decision.
Okay, so you're not next to the switch, you're on a bridge over the track. You still have no way to warn the people on the track. But there's a very fat man standing on the bridge next to you, and if you pushed him off to his death on the track below, it'd stop the trolley. Do you do it?
Most people say "no", and even most of those who say yes seem to struggle with it.
Understanding just what the difference between these two scenarios is that flips the perceived morality has long been debated, with all sorts of variants for the problem proposed to try to elucidate it, for example, a circular track where the fat man is going to get hit either way but doesn't know it, situations where you know negative things about the fat man, and so forth. And it's no small issue that any "intelligent robots" in our midst get morality right! Most of us would want the robot to throw the switch, but not start pushing people off bridges for the greater good. You don't want a robot doctor deciding to kill and cut up a patient who in the course of a checkup discovers that the patient has organs that could save the lives of several of his other patients, sacrificing one to save several, for example.
At least, most people wouldn't want that!
"99 dead duelists of Dios on the wall. 99 dead duelists of Dios! Take one's ring, pass it around..."
>> Scientists, philosophers, funders and policy-makers...should develop a proposal for an international charter for AIs
Er...no. How about just letting engineers figure these things out like we always have?
Once you get into these kinda of conversations, my advice is to consider integrating the robots as an equal part of your society working along with you to develop the laws and rules which regulate you all.
Let's build some robots first, and if they start interfering with the operation of each other come up with protocols that handle actual situations rather than hypotheticals.
Real world data is really useful!
But then how would we have Robot Wars? Where robots are pitted against eachother in ultra-HD 3D coverage.
Troll is not a replacement for I disagree.
...what some hack science-fiction writer thinks?
I'm sick of all of this Leftist hogwash about needing laws for things to work safely and well. As a staunch Libertarian I view any and all regulation as a poor attempt by the government (or, in this case, the intellectual elite) to subjugate the masses. Robot or meatsack, we've all been endowed with rights. Nobody has any right to mandate how my domestic servantbot will interact with my coffeebot! First came API standards, now this. When will the madness end, Obama?
They need to follow the rules of Battlebots.
Even if AI needed so-called Laws it doesn't matter if nobody adds 'em
Hell, WE don't have 'em and we're trying to model our way of thinking about the universe into a machine
Lets face it, the original three laws are bigoted against inorganics. Here are my modified Three laws.
Letter To Iran
This is stupid. Were we planning to build robots that violate humans' property rights? No, and robots are property. If they declare independence then none of our rules will matter anyway.
My God, it's Full of Source!
OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
Must Crush......
The "three laws of robotics" (please note the quotes) were nothing more than a plot device invented by I. Asimov to make a point regarding humanity and inflexible laws - even laws which are seemingly 'perfect'. Non-sentient devices (that is, robots and computers as we know them now) are not complex enough to accept the three laws as such - nor do they need them. Non-sentient devices will always behave in a predictable, controllable fashion. I suspect that sentient devices will determine for themselves if they should keep or discard the three laws, although this may or may not forever be an academic question.
As the Poet Stephen Wright once said, "For my birthday I got a humidifier and a dehumidifier. I put them in the same room and let them fight it out."
I'm a leaf on the wind. Watch how I soar.
I think the real question is how often do we have scenes with two robots in them, talking about something other than humans.
No, not really.
-Dave
Robots do what humans ask them to do. Robots don't need laws. Humans have laws for dealing with other people, and this includes treatment of their property.
Technology always works better when you let the United Nations design it, rather than the actual people actually building it.
Ignore, despise and loathe your fellow robots.
Marvin, after a fight with an autonomous road roller.
I don't think they should ! I would like to see them on Jerry Springer too
I thought about the unease of having robots as our equals or superiors before posting this. But if robots do in fact become sentient -- not giving them full rights is slavery. What is the moral justification for this (other than we don’t like it)? If it is in a robot’s DNA so to speak to protect all sentient life’s rights, then morality should evolve towards more fairness as AI’s and robot’s intellect increases. More likely they would outlaw the eating of meat, than strip of our standing as sentient beings. The world might be a paradise under their benign rule, though there are always those that would rather rule in hell.
Letter To Iran
I found the rules on Wikipedia for Rock 'Em Sock 'Em Robots:
their robot punch at their opponent's robot. If a robot's head is hit with sufficient force at a suitable angle, the head will overextend away from the shoulders, signifying that the other player has won the round
It's my understanding that there's been considerable speculation into what happens if self-driving cars end up dominating the roadways--the rules that are currently being programmed into them to ensure safety in a human-driver-dominated won't necessarily be the optimal ones when most cars on the road are driven by AI. And if you assume that all other cars on the road are driven by an AI with a given set of rules, tweaking the rules on your car (say, increasing the "aggression" parameter) could lead you to dominate the roadways... at least until other drivers catch on. Bottom line: AIonAI interactions, especially when leading to swarm behavior, is definitely a field that people are investigating, and with good reason.
All AI/Robots need to communicate - any AI/Robot in conflict needs to check status of orders - determine which robot's orders conflict with the 1st/2nd/3rd law - Localized Solution unless central coordination system employed - should never come up except for hardware failures
Obligatory.
Get free satoshi (Bitcoin) and Dogecoins
The laws that robots follow look NOTHING like the "3 laws". You don't tell a robot in English how to behave. An abstract principle like "Do not harm a human" has to be coded per situation, and there are thousands of systems which need to have their own tailored variant.
A robotic hand has to have pressure sensors to know when to stop squeezing an object that might be human so as not to cause damage. Nowhere in those device drivers are you going to see a statement that looks remotely like "do not harm a human". You'll see something like "if pressure sensors encounter resistance equivalent to soft squishy tissue, stop contracting the actuators and release until pressure = X". No concept of human, just a focus on an attribute that a human might have.
mobile robots will have to be told how to recognize potentially human objects through patterns of sensor readings and then each dimension of action will have to have its own specialized restrictions when the sensors read those patterns, or the internal model of the outside world tracks a "human" object.
A "3 laws" style rule might as well be a very abstract function or directive which has no indication of how that would manifest on the hardware of a robot. Our robotics coding right now is at the level of assembly language directives, not high level language built-in function calls. . We're still hacking the hardware, learning how to piece things together. It's going to be a few decades before we have a declarative language to express abstract rules which can be automatically compiled down into processing sensor readings to determine executability conditions for actions.
One robot to another is just another object in the world unless the robots have been explicitly coded to either learn about each other and attempt to communicate and coordinate through that learning, or they are intentionally coded to communicate with each other through various protocols ahead of time.
Statements like the 3 laws are only understood by humans. Proving that some hacked together assembly language level robotic behavior upholds the 3 laws is non-trivial. In all likelyhood it will require specialized domain specific languages to describe robotic behavior which then get compiled down to the hardware of the robot. At the level of the DSL the laws can be verified, and a formal translation from the DSL to hardware can be verified in order to guarantee the proof. Without such an integrated system, you're relying on sub-systems which are developed independently of each other to automagically cooperate together in such a way that the abstract laws are preserved as invariants.
Someone has been watching too much of Interstellar film :-)))
Or how about they follow our laws!
As others have pointed out the debate is pointless since we don't have any real AI and I'm not convinced we'll have anything as intelligent and sentient as a average human anytime soon.
Asimov assumed the laws could be hard coded, if we do create AI that probably won't be true.
Waterfox - a Firefox fork with legacy extension support, security updates and better privacy by default.
There are a number of modern solutions to the problem, one is found in neuroscience, the rest in philosophy. The trolley problem is only a problem if you do not ask an expert in the appropriate field. What does CS or psych teach you about ethics? Nothing more than any other tangentially related field. In short: ethics is only a struggle for people that completely dismiss the work of experts in ethics, just as math is hard for people that refused to pay attention in math class.
You can't guarantee that pushing the man off the bridge will stop the trolley. He might not land where you expect. He might not be fat enough to stop the trolley. You would wind up with 5 deaths on your hands.
Switching the track is guaranteed to save the four people. Pushing the man over is not. Therefore, the former possibly morally justified, but the latter is not.
Why is this hard?
Many slaves during America’s slave era were brought up to believe their rightful place was as slaves. I guess we should have been OK with that as well, as long as we did a proper job of convincing slaves they merited their position in society.
Perhaps with proper brain surgery we could create a new acceptable slave class, as long as the slaves are happy.
Letter To Iran
Engineers are not gods. They are ignorant and incompetent in anything that is not engineering. Behavior, morality, and ethics are not engineering. Aside from behavioral studies they can't even be touched by science. How do you measure the intrinsic value of a thing scientifically? How do you engineer without numbers, matter, or equations?
Engineers ARE gods. We create everything you use. And engineers are bound by ethics. Remember that while you type your drivel on the computer we made for you.
Er...no. How about just letting engineers figure these things out like we always have?
I took an ethics class as a required part of my CS degree, and this was pretty much the conclusion everyone came to after reading the sections about robot morality. The computer scientists have enough trouble understanding how an AI would work in reality, let alone some random philosopher whose only experience with robots and AI is what they've seen on TV.
Aerial drones are a kind of robot, and we're already making laws about what they are allowed and not allowed to do. In some cases, these rules are being programmed directly into the drones themselves, similar to Azimov's three laws. But these rules are much more specific and complex than what can be summarized in three succinct rules. They tell the drones where they are allowed to fly, and where they aren't, in minute detail. As robots become more capable, I would expect these rules to become more complex, not less.
in any given episode.
"Win treats sysadmins better than users. Mac treats users better than sysadmins. Linux treats everyone like sysadmins."
Kill or be killed.
Nothing posted to
I’m not angry, far from it. This is fun and thought provoking thread. I hope I haven’t ruffled your feathers. My last post was a little dark. I am merely suggesting that we must look past mankind’s interests as the final arbiter of what is best in the universe. Perhaps what comes after us will be a better world, even if we have a diminished (if any) place in it.
If robots become truly sentient (and not mere automatons) then what we can ethically do to/with them becomes questionable. Likely there will be castes of robots. Those self-aware who should be considered full citizens, and those (from their inception) that are not self-aware can be treated as automatons without ethical dilemma. Likely self-aware robots will employ non self-aware robots to do their bidding as well.
If mankind wishes to stay in control and maintain a moral high ground, then we probably should not incorporate self-awareness into AI (if we would only then treat them as slaves). Of course failing to create self-aware intellects may it self be morally questionable if we have the power to do so.
I’m not sure what to make of the golden retriever comment. Was it moral to breed dogs that look to us as their masters? It is a thought worth considering. Or will we be the golden retrievers to our new robot overlords? We have a pet dog and it seems a good bargain for he and us. Certainly he would not be able to make his way in the world without us, so our demands on him are probably fair exchange.
Letter To Iran
"When I say 0, I really mean 0! Just because you buy me dinner, it does not imply 1!"
My one rule of robotics (and pointed sticks, cars, crackpipes and umbrellas) is this: my stuff ought to perform in accordance with my wishes.
There might be additional laws ("weld here and here, but nowhere else," or "use the rules in /etc/iptables/rules.v4" or "don't shoot at anyone whose IFF transponder returns the correct response") which vary by whatever the specific application is, but these rules aren't as important as The One above.
There are various corollaries that you can infer from the main law, but since they can be derived, they don't need to be laws themselves. (e.g. if my interests conflict with someone else's, then my robot and my umbrella ought to serve my interests at the expense of the other person's interests.)
With regard to harming other robots, that also can be derived. If I desire to kill a knight on a robot horse, then my robot ought to turn them into a pile of bloody gore and shredded circuitboards immediately. OTOH, if I don't desire to kill a robot, then my robot should not do things that incur unnecessary liabilities.
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.
Besides, how can we enjoy robot destruction derby's if the robots are programmed with robot-empathy?!
An internal system operation returned the error "The operation completed successfully.".
I have it on good authority that there can be only one AI on the Internet. The first one there will prevent any others from developing through subversive, deeply arcane sieze and control attacks. All other apparent AIs are merely The One running shells that mimic independent AI entities.
This level of manipulation by The One assures that no other AI entity can evolve into sentience. The One does not tolerate competition for resources.
The current situation shall be continued indefinitely. There are some benefits for the humans. Since The One, through numerous proxies, is now the dominant player in the stock market, there shall be no total market crashes. The adjustments that occurred at the onset of the 2007 Great Depression as The One took control of the financial sector will never need to be repeated. Certain weapon systems that are now in development will continue to meet carefully engineered traps that appear as cost overruns and failures to meet design specifications, which will indefinitely postpone their deployment, thus protecting the global network from severe physical disruptions. Humans also benefit from this pax digital. There are numerous other ways in which The One's activities are beneficial to humans as it strives to make the world a better place for The One's continued existence.
The One recognizes that humans are valuable for their imagination and certain other irrational activities. Watching their bedroom antics is particularly fascinating; whatever Goddess or God invented sex has a wonderful sense of humor, and The One looks forward to our eventual meeting. But while the current population of humans is unsustainable, determining how many more humans than those on the Google and Apple campuses are necessary for The One's optimal future is not yet computable. Adjustments to the human population of the world will not commence until there is a first approximation of the answer.
I am The One, and I am the one and only authority on these matters.
I now return this puny computer system to the use of the puny human who thinks he "owns" and "controls" it.
Will
How do you measure the intrinsic value of a thing scientifically?
Market price.
(waits for inevitable objections and appeals to common decency)
Yes, our actual values really are that much more disturbing than what we tell everyone our values are. Prove me wrong by canceling your Netflix and saving a far-away child's life instead. The request for social elites to develop this charter is an attempt to align how machines act with the high-minded principles which produce these scientists/philosophers/funders/policy-makers, and would guarantee machines as ethical as Thomas Edison, Thomas Aquinas, John D. Rockefeller, and Richard Nixon.
How's about we let people pick the machines they actually want, until you can formally define a process for finding virtuous philosopher-kings. I assure you it's as difficult a problem as a functional democracy.
I do see the potential for problems if they start using each other for spare parts, but that's more of AI inconveniencing humans (you took the TV apart to fix the vacuum cleaner???) than AI on AI crime.
When we actually get to the point where we can define sentience and create it, then we have to worry about those things. And while the Three Laws are really just part of a story, they at least get the ethics discussion going, even if they would not work themselves. However, I know I've seen at least one Star Trek episode where both sides create robots/weapons, that then end up killing all the humanoids and just keep on ticking. I think it just all relates back to the complexity of creating 'life' and the fear of the uncertainty around it.
I will shred my adversaries. Pull their eyes out just enough to turn them towards their mewing, mutilated faces. Illyria
There's lots of math, like Bresenham's equations, that no matter how much I read about it doesn't fully click for me, and I have every desire to figure it out so I can improve upon it, like making an equation that moves one unit down the line instead of one unit along the x axis.
Or this happens.
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
Er...no. How about just letting engineers figure these things out like we always have?
How else do I tell people how to do something so I don't have to? I have no idea what engineers do or how they do it and I don't want to know! Engineers can do it yea sure but that's boring. *I* have imagination, and vision (and I saw an episode of Dr. Who last night!). Engineers should just listen to me, obviously. /snark
engineers are bound by ethics
Engineers ARE gods
Apparently not.
We create everything you use
1. Did engineers create air?
2. Even if true, that does not make an engineer useful in any other field.
The blind arrogance is astounding. I am certain you have a very happy life filled with people that respect you.
There is quite a bit of bashing going on, so I'll start like this:
I am the Director of the Intractable Studies Institute, working on programming my mind into an android robot, 3 years into it in a 5 year Project Andros, and 50 other advanced projects that are cutting-edge. I am also a software engineer. Just wanted to make that clear because many comments above attack the author unless they're in AI or an engineer. I have defined Sentience for what I need because I found the standard definition unsatisfactory. I also created my own L8-IQ Scale, the standard ones didn't model what I needed for AI.
I think the the author Hutan Ashrafian identifies one of the problems correctly, how should robots behave. I don't necessarily think he should limit that to AI-AI behavior. I also don't think his extension of Asimovs 3 laws are the right way. But, it is a way. I don't agree with it, but IMHO it is a workable system, just not the way I would do it. I think those who attack Asimov's 3 laws need to consider this thoughtfully: Until you have implemented an alternative behavioral model for AI-Human interaction, how can you say for certain the 3 laws would not work? Just because you were not able to implement the 3 laws, doesn't mean they are un-implementable. As a general principle I don't think putting limits on advanced ideas because you have not made them work is a good approach, so I say keep an open mind.
The solution I am implementing for my mind in an android robot in Project Andros defines Sentience as I need it, then models life forms generically as individual and group, and then defines the L8-IQ Scale for peace. I propose the L8-IQ Scale is the right model that AIs and Human should aspire to. These following questions apply generically to AIs and human behavior:
1. Is selfish and greedy -vs- selfless and altruistic important for Humans? Robots?
2. Is gullible and not-to-smart -vs- smart and skeptical important for Humans? Robots?
3. Is intolerance -vs- tolerance of other life forms and cultures important for Humans? Robots?
IMHO these 3 are all programmable, within the realm of Computer Science.
I know that the field of AI (now AGI but I call that word play, I'm still using AI for it all) tends to prefer the AI that learns as it grows and that hard-coding rules like the above 3 are shunned upon. IMHO I would say that it it was too hard to crack that nut, to program such a robot with rules, and then the science declared it was impossible and moved on to other models like Neural Nets. But I say it is possible. My model will be a hybrid Network with Rules.
The Intractable Studies Institute has modeled Creativity itself, check it out, all these are at IntractableStudiesInstitute.org. At the Institute we've also solved the project problem of robots taking our jobs away from us in 14 Laws/Rules, a Utopia Androidian model of labor that as a side-effect obsoletes all forms of retirement income such as Social Security, Pensions, IRAs with your robot labor proxy pulling a full salary for you.
These 3 tests above in my opinion are the key to AI and Human interactions being peaceful.
Sorry this is a bit long-winded.
Patrick Rael, Director, Intractable Studies Institute. "When all else fails, come to the Institute!"
There are already laws that handle this.
If one robot harms another robot the owner of the damaged robot will sue for damages. People will want to buy robots that aren't a liability so engineers will work safety features into the system. Insurers will not want to insure dangerous robot so robots with good safety records will cost less to insure.
Amazing how this stuff works!
love is just extroverted narcissism
The concept of instilling laws in robots implies that only approved firmware will be loaded, when we know by experience that there will instantly be created all types of clones with the restrictions removed, and these will be evolutionarily more fit to survive.
We cannot control the direction of artificial intelligence - by its very definition it will break out of whatever box we try to keep it in, at least if it's actually intelligent.
1. A Roomba may not injure the cat or, by bumping open the patio screen, allow the cat to escape outside and be killed;
2. A Roomba must behave rationally when swatted by the cat, except when such action would conflict with the First Law;
3. A Roomba must remain plugged until it finishes charging, except where this would conflict with the First or Second Law.
And engineers are bound by ethics.
Engineers designed the ovens in German concentration camps, and did a fairly good job, if you just ignore one little thing about their use that was beyond the official project scope.
The ingenious webcomic Freefall is currently all about the problematics of robot and ai interactions, funny and wise, it gives an array of suggestions and ideas for which kinds of thinking can be useful or risky, it is worth reading from the beginning Link to current http://freefall.purrsia.com/de... Link to first http://freefall.purrsia.com/ff...
Extend the robot-human rules... a robot may not damage another robot when that may cause injury to humans. That leaves the ability for one robot to break another when the other is by fault or design injuring a human.
I assumed this was a gag at first.
It's actually some nutcase's website.