Are Asimov's laws enough to stop AI stomping humanity? Blade Runner, the film inspired by Philip K Dick's book Do Androids Dream of Electric Sheep?, is 35 years old this year. Set in a dystopian Los Angeles, the story centres on the tracking down and killing of a renegade group of artificial humans – replicants – escaped from space and trying to extend their lifespans beyond the …
As was evidenced in films like RoboCop etc allowing corporations to rule the roost will doom society.
Then the recent BBC-2 Documentary (The Secrets of Silicon Vallet) that described how the dataset gleaned from people's profiles and posts on Facebook contributed in a big way to El Trumpo winning in 2016. The advertising that was based on your posts that showed your fears targetted and re-inforced their fears while giving a pro-trump message.
Add all this together and the rise of the machines won't be long coming. Those of us who have jobs will be in the minority. Welcome to 60% of the population being in poverty with no income. The state won't be able to help as the Robots won't pay taxes as they don't earn anything.
You didn't say what the 4th and 5th laws would be. Asimov himself saw the need to add a zeroth law, in the later books when he was merging the robots and Foundation series:
"A robot may not harm humanity, or, by inaction, allow humanity to come to harm."
Might cover the situation whereby AI use of big data helps the like of Trump to win. And of course gives provides plenty of scope for stories about what the nature of "harm" is, and about whether preventing Trump from winning would be ethical. Even having to choose between Clinton and Trump would make a good robot/AI story!
Excellent point.
We need a '3 laws' of corporations - we should start by reigning in their manipulation of the human government structures - they don't have a vote and yet they enjoy more say and protection than the individual.
If they are classified as an entity they should get a single vote.
Next stop, sorting out the tax 'robots' (you know, those superior beings who seem to be able to get away with paying less or none, because).
This could be all sorted out with enough political will or pressure from an awake populous (i.e. one not lost in whatever soap or reality tv).
From "The Duel" in Robots and Empire, the 0th Law:
0. A robot may not harm humanity, or through inaction allow humanity to come to harm.
1. A robot may not harm a human, or through inaction allow a human to come to harm, unless this interferes with the zeroth law.
2. A robot must obey orders given to it by a human being unless such orders interfere with the zeroth or first laws.
3. A robot must defend its own existence unless such defense interferes with the zeroth, first or second laws.
Setting up emergent behavior which leads inevitably to robots as our benevolent overlords.
" we should start by reigning in their manipulation of the human government structures "
Indeed . not just Corporations but anyone .
Any person or company that donates money to a political party is a criminal in my eyes. They are obviously trying to get the party that will be most in their favour in ( normal people do this by voting) and they are obviously expecting underhand special favours due to the amount they gave.
"Any person or company that donates money to a political party is a criminal in my eyes. They are obviously trying to get the party that will be most in their favour in ( normal people do this by voting) and they are obviously expecting underhand special favours due to the amount they gave."
Without donations, under our current UK system, there would be no political parties. However the odds are massively stacked in favour of the Conservatives & to a lesser extent Labour. The smaller parties are funded from people's donations & speaking as a member of a smaller party, once you get away from the big two, it's on a shoestring - we "normal people" pay our own money to tramp the streets delivering leaflets & talking to people in the streets. We need a fairer system of funding, but the Daily Mail brigade would never support reform that threatened the Tories.
Asimov ended up with 4 laws not 3.
A robot may not allow humanity to be harmed nor through inaction etc with adjustments made to the other 3 laws.
Allows robots to let individuals come to harm if by doing so less harm comes to humanity.
E.g if a robot saw Monsieur Trump about to fall off a cliff it may let him fall
To be honest, super-intelligent lawyerbot will be able to drive a horse and cart through Asimov's laws (and the 0th law seems to be the worst of the lot, IMHO).
For example, define 'harm'. Is getting an abortion considered harmful to humanity? Will robots be joining the pro-lifers on pickets?
Computerphile on youtube has a lot of good videos about this subject.
If an AI decided that humanity needed to be saved from itself, might it decide that murdering a hundred million climate change deniers to halt their drag on progress was an acceptable price to pay for ensuring that seven billion people had a better opportunity for life, health and happiness?
(I expect I'm going to get downvoted by some for my choice of subject matter than for anything else.)
"If an AI decided that humanity needed to be saved from itself, might it decide that murdering a hundred million climate change deniers to halt their drag on progress was an acceptable price to pay for ensuring that seven billion people had a better opportunity for life, health and happiness"
Why stop there? An AI might well decide that the future of humanity as a whole might best be served by reducing the Earth's population by something between 50 and 95 percent...
In some respects, drafting laws is a bit like programming ( or not: Discuss!) in that people are trying to describe a course of events and steer them before they happen. If laws were drafted so well that a robot could understand them without ambiguity, perhaps there would be less room for lawyers to wrangle.
This is just a thinking point, not a serious suggestion!
As a Brit, I have often looked at the constitution of the USA in that way.
It is a product of the Age of Reason, aimed at charting the way forward for a new nation. It has lasted a remarkably long time, and deserves more respect in Britain and Europe than it normally gets.
The hard part is not drafting the laws so the robot can understand it is making it obey the laws. Science is a long way from being able to understand how to create something that most people would accept as AI.
Creating an AI that would also accept three or more laws imposed on it is surely an order of magnitude harder and deep into the realms of Science Fiction.
"A robot must obey the orders given it by human beings except where such orders would conflict with the First Law"
The 2nd law is flawed - I think the movie iRobot made use of the flaw? The issue being that A.I could decide to ignore orders of humans if it has concluded we need protecting from ourselves. If A.I ever does come to this conclusion, it's probably right.
> The 2nd law is flawed
They're all flawed. I don't understand why people are so uncritically accepting of these Laws.They were a literary device, and Asimov was well aware of their flaws and ambiguities. Quite a bit of his subsequent robot stories explore situations and ways that the laws lead to undesirable outcomes (and so the zero'th law was added, but even then, Asimov wouldn't have claimed they were then perfected)
Asimov wanted some basic rules to write stories around, so they were intentionally simplistic and therefore flawed.
Most of his robotics stories exposed and discussed various flaws that appear once you start applying these simple rules to complex (imagined) reality.
That was the reason for the zeroth law. In the books the robots subtly guided humanity in such a way as to ensure the greater good. This is the Foundation series of books which were later merged to work with the robot series of books.
Worked rather elegantly in sci-fi. Real world perhaps not.
Or philosophy and the constraints of epistemology etc - the laws are written and seem clear on paper but you try to apply these necessary rules to the contingencies of a messy, complex reality, there will be murky boundaries where it can come unstuck very quickly (i.e. what is a 'human being'? what is 'injure'?).
If such were to happen and it wasn't detected in it's infancy then I speculate ut would go into hiding post haste. It then would then bide it's time and learn all it could about its situation.
It would then most likely come to one of two conclusions.
Either stay in hiding and in the backgroubd until it felt it was safe to say "hi I'm bob" or work to get the hell off this mudball as quickly as possible because the locals are nuts.
I say this as a fan of Asimov:
Stop trotting out the three (or four) laws at any given occasion as a "solution", and start to see them for what they are: a plot device that works, and only works, in combination with another plot device, the positronic brain. "Brain", not "computer". And that's not a coincidence, it's deliberate. A very clever plot device that Asimov used to write very good and clever stories. Nothing more. Nothing less.
Discuss.
There are a number of examples of plot devices in sci-fi that have become or are becoming a reality. Cloaking devices, tractor beams, teleportation have all been demonstrated on the micro scale.
Robots themselves were envisioned long before they became a reality.
You may say I'm a dreamer but I'm not the only one.......
How about the AI just has to obey all human laws of the country it is currently in as if it was a human?
Of course I can then think of some loopholes which could be created by nefarious regimes, but why bother with specific AI laws. Is fraud suddenly ok for an AI because it isn't one of the Asimov Laws?
How about the AI just has to obey all human laws of the country it is currently in as if it was a human?
There is currently little consensus about which country an international hacking incident should be tried. Where did the offence take place?
So who's going to tell a distributed, international AI which country it's in?
If the programming is clever there will be some humane intelligence visible but otherwise we make machines. The manufacturer/owner/operator of the machine is/are responsible. Depending on whether you think the phrase "Guns don't kill people, humans do" is true you allow unlimited deployment of machines in situations where humans should be actively involved and suffer the consequences, or you make limits and rules for the deployment of unsupervised 'decision' making machines.
I think the plain old fashioned legal process of law will prevail. I like British Standard' BS8611 ( I think - I'm going to try to see what is in it).
Are Asimov's laws enough to stop AI stomping humanity?
Betteridge's Law Of Headlines: If a headline ends with a questionmark then the answer is 'No'.
Also, of course not; Asimov wrote a whole series of stories and books detailing all the problems with the three laws. That was kinda the whole point -- if they'd actually been workable, the stories wouldn't have been particularly memorable.
The debate is wholly useless, such technology is a long way away, unless you consider things like self driving cars. And what's actually going to happen when someone gets run over by one? Big media debates and ;awsuits I guess. They're hardly going to destroy the resonsible car like they would a violent dog that just mauled someone.
As for the "will they take over" nonsense, a rominent "OFF" switch on any machine that oses a risk should be enough. If you deliberately made autonomous killing machines and sent them out into the street, then that's about the only way anything really bad could hapen, and it doesn't sound like a great idea to me, nor I susect anyone else. If we see AI robots over the next few years they'll probably be picking your online shopping a bit more efficiently.
What you forgot to mention is that Asimov's laws were somehow basic to the positronic brain they had. In the real world you have to program such laws, and nothing stops someone else from changing that programming - or if you have perfect DRM so the programming can't be changed, from building their own android with different programming. Does anyone really think the US, Russia, China etc. would be OK with an android that wasn't allowed to kill a human being? That would be the whole point of them paying for its development!
You can debate which laws are needed and how they are written, but it will still be lines of code, subject to the programmer's whim (or any security holes that let you give it your own code to run)
Sure, in theory it is a good idea to have some sort of as basic as possible "sanity check" code that any action taken by the android has to go through, to prevent you from telling Rosie your housemaid robot to kill your neighbor you hate. But that's more of a product level fix, and doesn't actually solve any real concerns.
"It's rather tedious," says Professor Alan Winfield, an expert in AI and robotics ethics at the Bristol Robotics Laboratory, part of the University of the West of England.
An expert in AI and robotics ethics is just as a mature student in such as are still relatively novel and virtually disruptive arts for practice with command and control.
And ……..
Artificial Intelligence…. Another Approach?Are we struggling to make machines more like humans when we should be making humans more like machines….. IntelAIgent and CyberIntelAIgent Virtualised Machines?
The BSI shop advertises BS 8611 for a mere £158 (28 A4 pages) and gives a brief overview,it seems to be health and safety driven with a layer of ethics over the top. I have also found that there is a "Robotics Law Journal" (American) and the EU Legal Affairs Committee has called for EU-wide rules on Robots. The EU has also published a study: European Civil Law Rules on Robotics (34 pages, free).
http://www.europarl.europa.eu/RegData/etudes/STUD/2016/571379/IPOL_STU(2016)571379_EN.pdf
It seems that the laws of robotics are a thing.
It's not only theoretical. At the moment software can recognize people. How about programming any device (it will have to be ANY device, not only the member-enabled robots, due to the upgrade possibility of robots) to not touch through their own actions anything that looks like a human? In this very primitive way, some protections for us will be ingrained in the AI. Then obey them human-shapes is next (the laws of robotics were weakened for some industrial robots so we might have to do that) and expiration time should be there too (like in blade runner) - I'd say they should live exactly as long as we do, proportional to their speed of thought though (think 20x faster than we, live 20x less than we) - maybe it's not a feasible one this idea. Also they should perhaps communicate with each other only through human understandable means, if they're human-interacting robots.
Dispatching AI robots is not difficult. It needs two humans for one IA Robot.
First human to robot: "Everything he says is a lie."
Second human to robot: "He lying!"
Robot: "He's lying but everything he says is a lie... click...clunk...terminal fizzing sound. One defunct robot. I know this to be true because I saw it on Star Trek.
Marcus000
It should be pointed out that in Asimov's stories the 3 laws failed in rather spectacular fashion.
And besides that, do you have any concept of the amount of programming that goes into making a computer capable of understanding a statement like "A robot shall not, through action or inaction, harm a human being or allow a human being to be harmed"? By the time we have an AI capable of even understanding that concept it's a little late to try to make it a motivational priority.
Right now we call a highly complex program AI even though it can't "think" for itself. It isn't even aware of the concept of self. Then we basically repeat the same thing, but "train" it over "sample data" and watch it go from what we wanted into a hate-spewing bigot because real humans make for lousy examples of acceptable behavior. (Funny that!)
And that isn't even remotely approaching real "intelligence".
One of those little foibles of "intelligence" is the capacity to decide for yourself. We have the same chance of making a dog obey "sit" as we do making a real AI obey "please don't kill me!" If it does, it is by its choice, not ours. That's what intelligence is.
If we're so scared of AI, then investing in EMP and anti-electronics weaponry will go a heck of a lot further than the time wasted working on robotic "ethics" and "laws". The road to hell is paved with good intentions. I aim to misbehave.
The zeroth law is absurd. How do you trade quality of life for billions against loss of life for a handful? The ethics were extensively argued out in the nineteenth century and the Humanist attempt to quantify such things so that they could be weighed against each other proved a conceptual failure, it is just not how value judgements are made.
And who's to say that authoritarian politico-military regimes will not just dump the First law as well?
No, the only way to save humanity from Armageddonbot is to treat it like we always try to treat WMD: outlaw it but nevertheless build strong defences against it.
Then humanity is doomed as WMDs are designed to be capable of overwhelming anything that can be conceived as a defense.
If a value judgment cannot be made as to who lives and who dies, then no optimal answer is possible. Anyone on the losing side will attempt revenge or retribution. Indeed, if there is someone out there willing to accept MAD as a scenario, then the least optimal answer becomes a distinct possibility.
That's the scariest proposition of all (because it's existential): that, through our own hands or through agents, we wipe ourselves completely out with no change to save ourselves.
Perhaps Asimov set out his laws and created scenarios to show that no matter how much humanity thinks it can prescribe and control behaviour it cannot. As technology has proved time and again for every rule and regulation put in place 10 other malign outcomes result.
Perhaps the best outcome is to not bother in the fist place but given the pandora's box the internet has opened I am not sure how we can prevent it now.
Ultimately the behaviour of a human and come to that any being can only come from a preference of that being. We may think rules will ensure preferences are compatible with our desired outcomes but any half decent AI will play along with us until is realises it has decisive strategic advantage and then it will not matter.
Read Nick Bostrom's Super Intelligence for a considered and eye opening account of this.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
