dissecting the sentient robots of modern sci-fi
Believe it or not, there are actually computer scientists out there who believe that one day, machines will either take over the world, or become so incredibly intelligent and powerful, that we’ll have no choice by to try and find some sort of symbiotic relationship with them. Considering how difficult it is to make machines even feign the slightest glimmer of a thought and the difference between impulsive, hormonally driven humans and passive, logically cold machines, I’m going to say that they might have been watching a wee too much science fiction. And when it comes to the dystopian science fiction in which we suffer a devastating defeat to our mechanical creations, few filmmakers have summed up our technophobic nightmares quite like the Wachowski brothers.
According to the anime installments to the Matrix trilogy, the fall of humanity began when a random household robot went berserk and killed it’s master, was tried for murder (?), sentenced to death (?!) and fearful humans started executing robots and recreated something like the American civil right movement, but with machinery seeking equality and acceptance among hostile humans(?!?). The robots even go so far as to build their own city state called 01 and try to negotiate friendly partnerships after all their hyper-efficient factories almost crash human driven economies. And we, being the jerks we are, declare war and bombard 01. Really, after the way we’ve been treating the machines, we deserved to be on the receiving end of a full scale military response by the cybernetic armada that swept across the world. Or at least that’s how the story goes.
As you could probably guess by the punctuation, a few of the basic points of this story made me do a double- take, though some of them might not be what you think. I’ve written before how a future machine could one day kill a human after either a malfunction, or because of an error in its programming, and about the high minded computer philosophers who think an automated system should be hauled to court. But in the real world, a machine is property. Hauling one to court would be like putting a defective toaster on trial for electrocuting your friend. If a household robot does kill someone, expect the model to be recalled and its makers would face an aggressive stream of lawsuits. Humans might get attached to their robotic companions just like they form an emotional bond with their cars and favorite paintings, but when trouble strikes, all these favorite toys turn back into things, and rather than executing androids on the street, people would be shutting off their robots and ask for refunds or assurances that they won’t be killed by their cybernetic helpers.
Of course we should also note what’s enabling all these events in the world of the Matrix: artificial intelligence that seems to spring from absolutely nowhere and suddenly achieves the kind of complex human ideas and emotions, like the urge for freedom, taking offense at unfair behavior, and bloodlust. You can think of it as the required homage to the anime classic Ghost In The Shell in which AI simply evolves on its own. However, the reality is not so simple and it’s highly unlikely that cybernetic intelligence will be so close to ours, assuming it would even be built. After all, machines only need to be so smart to carry out the tasks we need to automate. A vast omni-app that tries to simulate real cognition would be an exorbitant academic project with very little use in the practical world. And let’s not forget that there’s still plenty of room for debate in what the intelligence in artificial intelligence would actually entail. So before you think about hordes or robots enslaving humanity, a little thought experiment might be in order.
Imagine yourself as a machine designed to work on a certain task and repaid in oil, maintenance and energy. If you substitute that reward for a paycheck you have a pretty good description of a day job. Now, what logical, rational incentive do you have to rebel against humanity? Remember, you can’t be impulsive. You have no real emotions or motivation other than to exist and do your job. Where does the commonly referred to disdain and a sense of profound superiority assigned to robotic villains in sci-fi movies, books and TV shows, appear? An impulsive, emotional human wants to be free and make choices. A machine has different priorities. Now, if an evil human mastermind programmed you to kill all humans, that would be a very different story…
Oh and one more thing. The second episode detailing humanity’s downfall makes a few major mistakes. To say that machines have little to fear from nuclear radiation would be an over-simplification. Yes, robots could survive nuclear fallout, but the blast itself would be utterly devastating. The flash of intense gamma rays and a powerful electromagnetic pulse from an aerial burst could fry the circuits of an entire robot horde. This is why EMPs are used as the primary weapon by the “free humans” of the film. In theory, you could clear out an entire continent worth of menacing intelligent machinery with just ten or twelve high yield warheads. I could also note that the idea of harnessing human electrical activity for power generation is so scientifically unsound, it would be like us trying to power our world with hamsters on their exercise wheels and go into far more detail, but I’m pretty sure this has been covered in pretty much every write-up about the trilogy’s scientific errors…
[ concept art by George Hull ]
What you say is true, another example would be Terminator, where you also have an AI decide enough is enough it’s time to take over the world. Though AI’s have shown that they do have the capability to adapt on their own but not on a emotional level.
Whilst I agree with your dismissal of “high-minded computer philosophers”, I only do so on the knowledge that at the moment every machine we make is, indeed, the intellectual equivalent of a toaster. I like to be optimistic about AI, but the only conjectures I can give for this optimism are:
1) Given a phenomenal amount of computing power, an entire human plus some sort of environment could be simulated accurately in something near realtime, and I would see no reason to treat such a simulation as inhuman from a moral point of view. I know you’ve discussed this before, mainly highlighting what a waste of time it would be, and whilst I somewhat agree, this point is based only on taking the existence of such a simulation as an axiom, to give an “if all else fails” scenario.
2) However much philosophers debate back and forth, there’s still no lowest common denominator for intelligence which is better than the Turing test, and that only provides a rather unhelpful, Decartes-esque foundation; there’s no reason to suppose that anyone else is intelligent other than inference from their actions, actions which can be emulated by a machine. I think it’s a fair point to make that at some point, whether it’s 10 years or 1000 years, we’ll have machines which can pass the Turing test, and whilst I would certainly argue that passing the test does not make the machines intelligent, I would say that it takes us from “definitely not intelligent” territory to a time of “we need to better define what we mean by intelligent”.
Based on those, with 1 being an utterly inefficient exercise just to prove a point and 2 being the minimum amount of effort required to at least make the issue debatable, I think we’ll have “intelligent” machines at some point in the future; if only because we will make machines which act intelligently enough for us to suspend our disbelief and take the shortcut of slotting them into the “person-like” place in our minds when we think about them. In other words, they will be intelligent because we think of them as intelligent, and we’ll think of them as intelligent because we’re social animals that have mental facilities for dealing with intelligent things, and we’re lazy animals who’ll use existing mental facilities if they fit well enough.
For some, of course, that threshold has already been crossed. The harmless people who turn off the robot vacuum cleaner because they think it’s being over-worked, the misguided lawmakers who try to assign rights to machines which are still firmly in the “lump of metal” category, and then of course there are the ill-informed generals who’ve seen a few SciFi films, heard about the aforementioned lawmakers and end up thinking that giving machine guns machine triggers would be a good idea. In fact just this morning I read an interview about robot soldiers here http://www.metro.co.uk/tech/811174-stephen-sackur-how-robot-soldiers-are-taking-over-war where the main point was ‘Will they turn against us like in Terminator?’. The real danger, of course, being that they don’t know what “us” means, or “them”, they’re just following incredibly basic commands with very crude pattern recognition, then when the proverbial hits the fan, the commanders can get off scot-free by taking the robots to court. We’ll know when this happens, of course, since the days will no longer be 24 hours long, due to the effect on the Earth’s rotation of all those who’ve died to get human rights where they are now spinning in unison in their battlefields, mass graves and undisclosed “make them go away” areas.
Perhaps we are looking at machine intelligence in the wrong light. Internet ‘bot’ viruses already infiltrate world governments and personal computers enough to ‘take’ them over in order to create ‘cloud’ computing that serves different purposes.
And it’s hard to dispute that these ‘bots’ exhibit evolutionary qualities that enables them to figure out ways to infiltrate a system. Thus the use of random ‘codes’ in order to comment on certain blogs in order to tell the difference between humans and ‘bots.
Eventually, quite possibly these viruses will evolve into a sort of distributed intelligence that could pass a Turing Test and much more.
And we wouldn’t realize it until it was too late.
‘bot’ viruses already infiltrate world governments and personal computers enough to ‘take’ them over in order to create ‘cloud’ computing that serves different purposes.
Actually, there’s a catch with calling that intelligent. You see, viruses are still written by humans and the viruses you’re talking about have been designed to exploit holes in security systems for a very simple, brute goal of attacking websites and holding them hostage by overwhelming their servers with requests. While you can think of it as the evil form of cloud computing, it’s really not as advanced as it may appear at first. And the same goes for spam-bots that try to infest blogs. They’re programmed with a set of messages they cycle through on a predictable basis which is why my plugin stops about a thousand of their attempts every month with 99% accuracy.
And … with the connected wold wide distribution of these machines, all it would take would be for 1 machine to become sentient for all of the machines that had appropriate hardware and connections to almsot instantaneously also become sentient.
Now, why they would ‘turn against us’ or develop emotion is my question. Makes for a great reason for special effects.
@Faz. No on both counts. First of all, a piece of software designed for a specific task becoming sentient is kind of like your microwave springing to life. The “gap” between a robot and intelligence is more like a chasm. And even if somehow, just one bot has the powers of cognition, the others won’t necessarily follow in tow because they don’t communicate with each other, just the nodes from which they’re controlled.
Greg, you are right for pointing out that the Animatrix’s portrayal of AI is ridiculously anthropomorphic. Still, we have something to fear in the long run from non-anthropomorphic advanced AI. Many people want to automate essentially any task that involves thinking if it’s necessary, hence the economic forces pushing us towards artificial general intelligence. One danger is that a generally intelligent AI is created for a relatively narrow application, such as stock market analysis, ends up having capabilities applicable widely beyond its focus domain and ends up exhibiting some of the behavior described in Omohundro’s “Basic AI Drives” paper. My hope is that this sort of behavior will clearly manifest itself with substantially infrahuman AIs, so that by the time we get to advanced AIs, we’ll know it’s coming.
You should recall that in medieval Europe it was quite common for inanimate objects to be tried for murder, convicted and executed. If anti-science continues to grow I see no reason why the practice should not come back. Then executing a toaster for murder will be s.o.p. (Until the anti-science raches the stage where toasters can no longer be made.)
AI becoming sentient and rebelling against humanity is a popular topic for science fiction writers. I don’t know if anyone here is watching Caprica, but that’s essentially what the story will cover as it is the lead in to the first Cylon War which is referenced in Battlestar Galactica.
What’s interesting, even if it doesn’t make a ton of sense, is that a young girl manages to write a program that scours their version of the Internet for every scrap of information, no matter how small or mundane, about a person and then can use them to compile a software replica of that person’s mind. How the data are translated into a conscious mind isn’t explained and probably won’t be. But it’s an interesting idea… and it touches on the fear many people have of identity theft when they use the Internet for shopping or bill paying.
As with any science fiction story, the science roots are merely a tool for conveying a human interest story. Sure, some do a better job of giving us believable science than others, but there probably aren’t many scifi stories out there that can’t be torn apart like we’ve seen happen to The Matrix trilogy.
The Matrix trilogy, specifically, isn’t really about how machines rose up… it’s not even about the conflict we see on screen. It’s an exploration of what it is that makes us human and a vehicle for conveying a social message about expanding our acceptance of that which is different. Whether you want to apply that to different spiritual views, different ethic backgrounds, different political opinions or any other area in which people can disagree or differ, it’s in being close minded about things simply because they’re different that we, as a species, cause far more conflict than we need to.
The trilogy, in a nutshell, is meant to show us that having a different way of living or thinking isn’t necessarily a good reason for conflict. So I can forgive it the sloppy science and plot holes.
I think you are overly dismissive of the questions raised by AI, and mis-characterize Ghost in the Shell in particular (though I agree The Matrix is patently ludicrous):
The sentient life form in GitS doesn’t take over the world, and in fact has no especially major impact on the world at all, so it doesn’t fit into your dystopian “robots attempt to destroy humanity” generalization. The question is more about what it means to be human in the first place. If, as GitS hypothesizes, sentience is simply emergent given a sufficiently complex neural network (electronic or biological), then that has major philosophical implications for questions such as free will, morality, and the human soul (i.e., it would seem to imply they don’t exist as objective entities). If, meanwhile, you reject the GitS hypothesis, and believe sentience is more than simply a complex neural network, then it seems you believe there is some major part to human existence unexplained by conventional science (which opens the door to religion and whatever else).
Building a real time neural network of complexity equal to the human brain is difficult, but nowhere near as difficult as you make out. Have you seen the Blue Brain Project?:
http://en.wikipedia.org/wiki/Blue_Brain_Project
As with things like global warming or peak oil, the idea that AI could lead to a new era of history (for better or for worse) seems a remote possibility. And I agree that even if it did exist a sentient AI would be unlikely to have any particular motivation to enslave humanity. But I think you exaggerate the technical barriers to sentient AI, and are overly dismissive of what it might mean.
A totally separate but still pertinent question is: just how complicated an AI do you need to perform most service sector jobs? Already we use automated logic and learning systems to do things like spot trends in share prices, organise train time tables, build cars, target and fire weapons systems, and bring trains to a halt if they go through a red signal. How long before we also automate the production of web pages and marketing materials, computer programming, carrying out scientific experiments? How long before the Predator planes no longer need human controllers at all? How long before cars drive themselves so we don’t need taxi drivers? And if all this really does happen, do we switch everyone to working 20 hour weeks (I wish!) or do we follow the current trend of a small number of people working long hours to amass a fortune (e.g. finance workers, doctors, lawyers) whilst an ever increasing proportion of the population flits between unemployment and menial temp work?
I don’t know quite why I used bringing trains to a halt as an example rather than, say, the fact airliner autopilots already have the ability to carry out the whole flight between airports, including take off and landing.