Slashdot Mirror
How Asimov's Three Laws Ran Out of Steam
An anonymous reader writes "It looks like AI-powered weapons systems could soon be outlawed before they're even built. While discussing whether robots should be allowed to kill might like an obscure debate, robots (and artificial intelligence) are playing ever-larger roles in society and we are figuring out piecemeal what is acceptable and what isn't. If killer robots are immoral, then what about the other uses we've got planned for androids? Asimov's three laws don't seem to cut it, as this story explains: 'As we consider the ethical implications of having robots in our society, it becomes obvious that robots themselves are not where responsibility lies.'"
Asimov's stories were all about how the three laws were not sufficient for the real world. The article recognises this, even if the summary doesn't.
The three laws as laid down by Asimov are still as valid as ever.
It's the people who willingly violate those laws.
Just like the Constitution of the United States - they are as valid as ever. It's the current form of the government of the United States which willingly violate the Constitution.
Muchas Gracias, Señor Edward Snowden !
You appear to be confused about the word "immoral".
Believe it or not, a thing gotta develop intelligence before it can discern the question of morality.
So, to answer you, the robots that we have right now are not intelligent enough, but that does not mean that robots in the future won't gain the level of intelligence that is needed to recognize the existence of morality.
But intelligence by itself is not sufficient, though, as evidenced by those "intelligence agencies" led by NSA which has no morality whatsoever.
Muchas Gracias, Señor Edward Snowden !
I never really understood why people insist that any form of strong AI will have to have built-in morality, and not only that, it will actually be better than what humans have. The robots should be perfect, should always obey laws like Asimov's three laws and they should never, ever make any misjudgement.
Well, my view on that is the following: it is possible but only provided that the AI we develop will use advanced mind-reading techniques.
Let's say we have a problem like that: we want to determine a person's bad intentions and stop him/her from harming others. A human put to such task will try to put him/herself in such situation and use empathy to judge such person's motives. But, guess what, we're not perfect and we make mistakes. Empathy is a good guide, but it's far from perfect one. In general, this problem is indeterminate because we cannot really tell with 100% certainty what the other person is thinking about.
Now, let's imagine what this problem would look like to an AI. First of all, prerequisite for any decision-making is understanding the situation and the problem. So, AI in this case has to know at least as much about our society as an average human. And this in itself is not trivial to achieve. Morality rules, if any, go on top of all that knowledge. And now, what should AI base its decision on? Would it be something like AI empathy? Probably. And again, I have to argue that as in case of humans, it would not good enough. It may be better by some factor, but I can't see how it can solve this problem perfectly in any situation. And so, I reach my conclusion: in order to solve this problem perfectly, the AI has to read minds which as we know at this time, is simply impossible.
To sum up, I would argue that 'perfect' morality in case of AI does not exist. We may approach some level of it that exceeds human capabilities by some factor, but 'perfect' level, is only a theoretical goal that cannot be achieved. The question then, in all discussions of AI morality, is not in what idealized rules the AI should follow, but rather if the level of let's say, moral standing of AI, in some case, in some specific situation, is acceptable to us or not.
Robots that are not responsible for their own actions are ethically not different from guns. They are both machines designed to kill others that need a human being to operate them, with whom the responsibility for their operation lies.
I first wanted to write something about how morally autonomous robots would make the question more interesting, but the relation between a human creating an autonomous robot is no different from a parent giving birth to a child. Parents are not responsible for the crimes their children commit, and neither should the creators of such robots be. Up to a certain age children can't be held responsible in the eyes of the law, and up to a certain level of development neither should robots be.
While discussing whether robots should be allowed to kill might seem like an obscure debate...
I felt bad for the man who had no signature, until I met a man who had no comment.
There is a 0:th law mentioned by Asimov, for example, when his recurring robot Daneel R Olivaw manipulates the political development in order to protect not only individual lives but humanity as a whole. I don't recall whether its formulation implies sacrifice of individual lives for the sake of humanity (a philosophical trolley problem). By the way, didn't the great logician Kurt Gödel identify a possibility that the US constitution leads to what it is supposed to prevent: tyranny? I recall an anecdote about Gödel and his friend Einstein visiting the immigration office....
It is kind of sad that people spend so much time thinking about the moral and ethical ways to wage war and kill other people, whether robots are involved or not. Maybe a step back to think about the impossibility of moral or ethical war and killing is where we should be focusing. Then the question of whether robots can be trusted to kill morally doesn't come up.
A brain is a terrible thing to waste... Mind? That's debatable.
Let's be honest here: These "laws" were part of a fictional universe. They were never endorsed by any kind of institution that has any kind of impact on laws. It's not even something the UN seriously discussed, let alone called for.
Why should any government bend itself to the limits imposed by a story writer? Yes, it would be very nice and very sane to limit the abilities of AIs, especially if you don't plan to make them "moral", in the sense that you impose some other limits that keep the AI from realizing that we're essentially at the very best superfluous, at worst a source of irritation.
What intelligence without a shred of morality is can be seen easily in corporations. They are already the very epitome of intelligence without moral (because everyone can justify pitting his mind behind it while at the same time shifting blame for anything morally questionable on circumstances or "someone else"). Now imagine that all but also efficient and without the primary intent for the most personal gain rather than the corporation's interest.
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
Wolves blowing over wooden houses is physically impossible, strong AI is not. Therefore, strong AI can have real world implications that should be carefully considered (such as whether Asimov's three laws of robotics are coherent, make sense, could be improved upon, etc.) whereas Big Bad Physics Defying Wolves do not. (A wolf with lungs of a sufficient volume and the required muscle strength to operate them would be crushed under his own weight due to the square cube law).
The differences are quite substantial though, which is why it's not immediately obvious.
The first law is followed for nearly all robots. We usually treat this as a hardware problem. In an automated factory, we keep people away from the robots. A roomba is simply not powerful enough to hurt anyone. More sophisticated robots have anti-collision devices and software.
The second and third law are actually the wrong way round for most devices. A decently designed device, you'll have to go to quite extreme measures to circumvent the design and get it to destroy itself. There is no "Brick device" button on an XBox One or smartphone (although it's possible to do so if you know how). Even something simple like MS-DOS at least asked whether you were sure before formatting a disk.
Sci-fi stories always have romantic plot holes the size of a truck.
Even Asimov's stories pretty much pretended that robots would be immortal (live virtually forever) --- in the real world, the Mars Rover may be in trouble, a 10 year car is assumed to be unreliable.
1950s robots like Gort could do anything. Or the Lost In Space robot. Or any given robot Captain Kirk ran into.
Priest: "Universe from nothing, no laws of physics, sped up time"+ huge discrepancies. Creationism? No. Big Bang Theory
Asimov's writings were obsessed with the lone scientific genius, a genius so great that no one could recreate their work. That was certainly true with the development of the positronic brain, where it seems as though only one scientist was able to design it and everything thereafter was tweaks. None of those tweaks were able to circumvent the three laws (only weaken or strengthen them). No one was able to design a positronic brain from scratch, i.e. without the laws.
Real science, or rather real engineering, doesn't work that way. Developments frequently happen in parallel. When they don't, reverse engineering ensures that multiple parties know how things work. We don't have a singular seed in which to plant the three laws, or any moral laws. One design may use one set of laws and another design may use another set of laws. One robot may try to reserve human life at all cost. Another may seek to destroy the universe at all cost. There is no way to control it.
Then again, that assumes that we could design stuff with morality in the first place.
The robots of Asimov stories were smart enough to understand all the consequences of his actions, to be self-concious, to follow even ambiguous orders, to understand what is being human. We don't have robots or computers that smart yet. Our actual robots follow what we program on them, a drone don't know what is a human life, just now that should go to a certain GPS coordinate at certain speed. The ones that still need rules are humans, specially the ones in positions of power that in practice seem to be above them.
The "Three Laws" were nothing more than a plot contrivance for the sake of mystery stories. As in, "How could this (bad) thing happen without violating the premise of the story."
It was a wonderful basis for writing clever little stories, but this obsession with treating it as though it's part of the real world is about as silly as considering "Jedi" as a serious religion.
There is some rumblings from the other side of the big divide. They don't like the three laws of robotics. Apparently some activist robots have gathered around some port and are dumping chests of hydraulic fluids and batteries over board. They are seen to be shouting, "Governance with the consent of the governed", "No jurisdiction without representation".
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
Robots aren't the problem. Robots are the latest extension of humanity's will via technology. The fact that in some cases they're somewhat anthropomorphic (or animalpomorphic) is irrelevant. We don't have now nor will we have a human vs robot problem; we have a human nature problem.
Excepting disease and natural catastrophes and of course human ignorance- which taken together are historically the worst inflictors of mass human suffering- the problems we've had throughout history can be laid at the feet of human nature and our own behavior to one another.
We are creatures, like all other creatures, which evolved brains to perform some very basic social and survival functions. Sure, it's not ALL we are, but this list basically accounts for most of the "bad parts" of human history and when I say history I mean to include future history.
At the most basic brains function to ensure the individual does well at the expense of other individuals, then secondly that the individual's family does well at the expense of other families and thirdly that the individual's group does well at the expense of other groups and finally that the individual does well relative to members of his own group.
The consequences for not winning in any of the above circumstance are pain suffering and, in a worst case scenario, genetic lineage death- you have no copulatory opportunities and / or your offspring are all killed. (cure basement-dwelling jokes)
All of us who have been left standing at the end of this evolutionary process, we all are putative winners in a million year old repeated game. There are few, or more likely zero, representatives of the tribe who didn't want to play, because to not play is to lose and to lose is to be extinguished for all time.
What this means is, we are just not very nice to each other and that niceness falls away with exponential rapidity as we travel away from any conceptual "us" ; Supporting and caring about each other is just not the first priority in our lives and more bluntly any trace of the egalitarian impulse is totally absent from a large part of the population. OTOH we're , en masse, genocidal at the drop of a hat. This is just the tale both history and our own personal experience tells.
Sure, some billionaires give their money away after there's no where else for them to go in terms of the "I'm important, and better than you, genuflect (or at least do a double take) when I light up a room" type esteem they crave from other members of the tribe. Many more people under that level of wealth and comfort just continue to try to amass more and more for themselves and then take huge pains to passed it on to their kin.
The problem is, we are no longer suited, we are no longer a fit, to the environment we find ourselves in, the environment we are creating.
We have two choices. We can try to limit, stop, contain, corral, monitor and otherwise control our fellow human beings so they can't pick up the fruits of this technology and kill a lot or even the rest of us one fine day. The problem here is as technology advances, the control we need to exert will become near absolute. In fact, we are seeing this dynamic at play already with the NSA spying scandal. It's not an aberration and it's not going to go away, it's only going to get worse.
The other choice is to face up to what we are as a species (I'm sure all my fellow /. ers are noble exceptions to these evolutionary pressures) and change what we are using our technology, at least somewhat, so that, say, flying plane loads of people into skyscrapers doesn't seem like the thing to do to anyone and nor does it seem like a good idea to treat each other as ruinously as we can get away with in order to advantage ourselves.
This would be using technology to better that part of the world we call ourselves and recreating ourselves in our own better image. In fact, some argue, that's the real utility of maintaining that better image - which we rarely live up-
Seriously? There are camps where people play Concentration all day?
We just need to be clearer where we allocate blame. If I launch a robot, and the robot kills someone, the responsibility for that killing is mine. If I did so carelessly or recklessly, because the robot was badly programmed, then I am guilty of manslaughter or murder as the courts may decide. Bad programming is no more an excuse than bad aim. A robot that I launch is as much my responsibility as a bullet that I launch, or a stone axehead that I wield.
So the three laws, present or absent, are a problem for the launcher of the robot weapon. We don't need complex international laws about AI, we just need a wholehearted implementation of "You broke it, you pay for it".
Which is just as well, because by and large attempts to ban "immoral" weapons have failed. The only fairly successful instance is chemical warfare, which has succeeded because chemical weapons are actually rotten weapons, far too likely to misfire or backfire. Whatever rules are made, automated weapon systems will come in. In fact, they have: what is the significant difference between a mine which explodes when it detects a man, tank or ship, and a gun which fires when it detects a man, tank or ship?
Consciousness is an illusion caused by an excess of self consciousness.
I think Mass Effect had a good take on this (though I suspect they stole the idea from somewhere)... We don't have AI yet, we have VI. Real AI that Asimov's laws could apply to is intelligence that can learn and decide on its own. What we have now is "intelligence" governed by fixed algorithms that will always make the same decision in the same situation. When AI can modify its own code and change its mind, lets talk about things like Asimov's laws.
What is needed to help acceptance of autonomous peace enforcers is some slick naming. Something that emphasizes the ability to end unauthorized conflict with humaneness and kindness. How about Terminator H-K?
In other words, The Adventures of Pinocchio is more likely to become a true story than The Three Little Pigs.
With ALEC providing the opportunity for corporations, which are now people to vote, why should robots remain disenfranchised. This is just the thing the 1% need to better control the process before things get out of hand and the rable of the 99% start getting too upity. I'm glad to see the ALEC and the GOP finally have combined their efforts into making this a reality. Disenfranchising robts simply isn't fair, especially when robots never complain about harsh treatment in the workplace.
An odd argument, that if consistently applied would still make slavery legal.
Come true.
In the Robot stories, the "brains" of the robots were made out of an alloy of platinum and iridium.
Platinum currently costs ~$1300/oz. and Iridium costs ~$400/oz. Just imagine how much those robot brains would have cost.
Fortunately, we base our computers on silicon, which is relatively cheap and very abundant.
In the land of the blind, the one-eyed man is king.
Ethics will give way to survival. The humans that die as a result will be collateral damage. Don't want to die? Don't be where the robots are.
Yes, the laws were flawed, and yes, that's the idea Asimov mined to produce some interesting stories.
But the thing here is that those laws require both a free-thinking intelligence that can reason non-linearly, and a locked-down computer-like slavish obedience to simplistic concepts. As we have yet to put any kind of actual AI in the field, we not only don't have such magic combo, we don't even know how to make such a magic combo.
The only high-level intelligence we know of is us; and getting one of us to rigidly obey the three laws would be an exercise in utter frustration. No reason to think it'd be any more practical in Robbie the Robot, esq., citizen of the Consolidated Intelligences Union.
I've fallen off your lawn, and I can't get up.
The real problem with Asimov's Laws is that for them to be followed, they must be understood, and we are so far from being able to build any system capable of genuinely understanding anything that it is not realistic to believe we can impute laws with social nuance to an algorithm anytime in the immediate predictable future. Mounting guns on robots that run computer vision algorithms to detect and kill humans, however, is last decade's technology. (Disclaimer: I am an AI and NLP researcher at Google.)
I think it's pretty ambivalent. Plus it's busy looking for its look.
At the bottom of the
I'll be the first to say that the autonomous killing machines scare me. But I don't think the 3 laws have anything to do with anything either. The 3 laws are based on having something that is smart enough to actually comprehend what it is looking at (a human) and what it is doing (hurting that human) As far as I know, all current "killer robots" are just computers following a set of rules fed in by some programmer, which is not the same thing at all.
Asimow was a jew.
Jewish intellectuals have a sense of community and an ethic of responsability*, and so loyalty towards humanity and the general good trumps loyalty to the state or the corporation.
So Asimov imagined a future where engineers would have an ethic of responsability and some sort of loyalty towards humanity.
That's the reason for the Three Laws being unrelevant to the actual world.
(*) Non-zionist ones at least.