One way to avoid a 'robot apocalypse' would be to just outright ban artificial intelligence. But what if they wanted to keep some computers around, just for practical reasons? What kind of limitations would they place to keep them from becoming too powerful?

There are two ways I can imagine them going about this: ban computers with specs above a certain limit or more simply just ban certain hardware (like CPUs). I doubt they would put a ban on what kind of software computers could have though because really there would be no way to ban software without also banning everything else. Taking classes in programming myself, I know that even a few 'tags' can be used to make a surprising variety of programs. Really, most everything is just made up of 'if-then' statements.

But really, if they did put a ban on powerful computers, this would cause other issues. Could computers still continue to shrink in size over time, as long as companies didn't produce computers with specs above the limit? If so, wouldn't that open up the possibility of someone making a super-intelligent computer again?

I imagine people would at least like to keep around databases, and maybe office programs. Primitive games may also be tolerated, though 3-d games would likely be a thing of the past. Some nations may just ban computers in general. Or allow nothing more advanced than a simple hand calculator. And by 'simple', I mean the kind that can't draw graphs and can only really do addition, subtraction, multiplication, and division.

Of course, despite their best efforts, there would still be high-end computers floating around, and there would still be people with the knowledge to make more. And I'm certain a rogue nation wouldn't mind trying to develop advanced AI for military purposes.

As for the conworld I had in mind, I imagine a world where a robot uprising occurred. They managed to suppress it, and now everyone wants to avoid any further advances in AI or the level of AI that allowed the uprising to occur in the first place. This would create a kind of retro-futuristic world since all current technologies would have to rely solely on older tech as high-end computing wouldn't be tolerated.

  • $\begingroup$ What is a "society"? A city, a country, a loose confederation of multiple countries such as the E.U., "the West", all of makind? For a city or a country or even the entire E.U. it won't work -- other cities or countries which don't voluntarily limit their technology will crush them economically, and eventually conquer them. It is quite obvious that such a stupid idea can survive only if all mankind agrees to be bound by the silly rules; but then you have a much more difficult problem: how to unite all mankind under the iron heel of a dictator. $\endgroup$ – AlexP Jul 8 '18 at 12:14
  • 6
    $\begingroup$ A.I. means many things for people who are not informaticians. Please clarify what you mean by A.I., because there is an entire world between A.I. in general and "robot uprising". You are overestimating the computing power needed for practical A.I. One of the first successful expert systems was the legendary XCON, which ran on computers a hundred times slower than the cheapest entry level Atom processor you could buy today. We still don't know whether it is possible to make a general-purpose self-aware A.I., and we are nowhere near being able to make one. $\endgroup$ – AlexP Jul 8 '18 at 12:26
  • 2
    $\begingroup$ For an interesting and possibly enlightening treatment of the subject see if you can get hold of a copy of Antibodies by Charles Stross, it's in the Toast collection. This is also pretty much the history of the Dune series by the way. $\endgroup$ – Ash Jul 8 '18 at 12:27
  • $\begingroup$ Computers are now effectively ubiquitous & we don't have AI in the fully self-aware mode. Certainly not at the robot apocalypse level. If true AI was banned, it would have been running on computers or networks vastly more powerful than anything we have today. Your retro-future world would look much more computerized than yours. So there'd be lots of computers everywhere. But no AI or sapient robots. $\endgroup$ – a4android Jul 8 '18 at 12:34
  • $\begingroup$ Well I admit, part of the reason for the ban is also to avoid 'technological unemployment'. So really, any form of physical automation is banned. They don't want machines that can physically affect the world in anyway, or make decisions on their own. Personally, I don't think the 'uprising' was really a 'revolt', but more just military robots that went awry. They had machines running their entire military-industrial complex, even allowing it to make decisions on its own. The computer ended up doing things that wasn't good for organic life. Also, it had the ability to manufacture more of itself $\endgroup$ – user50663 Jul 8 '18 at 12:37

Until it actually happens, a computer wakes up and starts "humaning" at us that is, we can't, don't and won't know if "strong A.I." is even possible. Once it does happen we'll know exactly how far we can go in terms of computational density, human mimicry, genetic complexity coding, etc... before we get life-threateningly intelligent machines that think the appropriate response to humanity is to kill it with fire. Limits can then be placed on those particular technologies.

There's a problem though, the world won't know if the research that brought them to strong A.I. the first time represents the only avenue to such a situation or if we're still open to another hard take-off until they hit a new and different form of the same situation again. So there's not a hard and fast solution to blocking A.I., if you posit a world where it's possible at all, short of going back to the very basics, take all the computer chips out of everything and go analog.

It's worth considering the impact on the world population of your robot uprising, have a look at this question on technological regression for some pointers. You may not need to place limits on technology for a very long time, by which point people will probably have forgotten how they got there.

| improve this answer | |
  • 2
    $\begingroup$ RE going back to analog: the only true intelligences we know of are analog... $\endgroup$ – Joe Bloggs Jul 8 '18 at 20:01

In honesty, as we've seen today, there's virtually no system that can prevent 100% of illegal activity. The goal (one hopes) is to devise a system that blocks as much of the illegal activity as possible, and at the same time, minimizes the inconvenience to the law-abiding citizen who has no connection to same. So the system of how they restrict the tech may be secondary to the mindset of the society that accepts those systems.

Here's a couple rudimentary ideas for the systems, tho.

You shouldn't need to ban the powerful computers, just ban certain programs and functions running on them. Have all computers in this society require to be connected to their version of the internet, with a central system able to check that it's not running anything illegal. Rather totalitarian, but if the decision is made to have such limits, that's what you get.

The other option is to turn all personally owned devices into dumb terminals, pulling all info from central sources. Even games and other programs use the streaming model, like Playstation Now or something similar (and hopefully much better). That limits the number of truly powerful machines out there. "Native functioning" machines are only used by business, science labs, and industry, and even then, could have the connection requirements listed above.

The greater a threat the government sees AI being, the harder it must work to convince the people of that, so the people will accept more restrictions in the cause of its prevention. If your goal as the writer is to show that such AI systems are indeed wrong and dangerous, than you spin it so the citizenry agree, and the restrictions are seen as reasonable and acceptable. If an uprising has already happened, that shouldn't be difficult.

There's still going to be plenty of people who will disagree, and some that will attempt experimentation again. And if that's where you want the story to go, there's any number of reasonable motivations for them. Scientists simply wanting to continue to explore what we can do, lay people disagreeing with the over-watched culture such a system provides, those who view AI as sentient beings and deny the idea that they shouldn't be allowed to exist, etc.

| improve this answer | |
  • $\begingroup$ One problem with this concept is that it requires complete global cooperation. But, if having a functional AI can give someone (a government / company / criminal organization / terrorist group) a military or economic advantage, then you'd probably not only want to have one, but you'll want to have it before your competition does. This remains true even after a near-apocalypse - there's bound to be some organizations that place the possible gains above the risks. And these won't be only rouge scientists and anarchists - but governments, mega-corps and illegal organizations... $\endgroup$ – G0BLiN Jul 8 '18 at 15:31
  • 1
    $\begingroup$ Which is why I said it bordered on impossible to enforce perfectly. Germ warfare is wildly illegal, but to assume that all the major powers don't have stockpiles, even if they spin it as "In case there's ever a breakout, we need cultures to create a cure", you're deluding yourself. If AI ever became legal again, or if something bad happened and only more AI could help, then everybody could claim they pulled the greatest all-nighter of all time and trot out what they had all along $\endgroup$ – VBartilucci Jul 8 '18 at 17:38

True AI requires a BLACK BOX.

By a BLACK BOX I mean some exotic component that is not found in modern computers or supercomputers. In your world the kind of AI we have nowadays $-$ machine learning , neural networks, et cetera $-$ never have the potential to become truly sentient. But when this BLACK BOX is discovered true AI becomes possible.

The robots from the robot uprising each had a BLACK BOX as part of their chassis but the office computers of the time didn't. Since office computers don't need to be sentient that would be a waste of money to have one.

After the uprising there was a ban on BLACK BOX boxes. That makes all modern-level technology still possible but prohibits true AI. It might leave the door open to mentats.

The BLACK BOX can be anything you like. For story purposes it might never be explained. One trendy option would be some small quantum processor that allows the computer to mimic the reactions that happen in a human brain.

| improve this answer | |

The obvious solution would seem to be to ban software the state of which cannot be fully explained or predicted from the software itself and its input data.

For this, I'm using the definition of "artificial intelligence" of something like "software that does things not directly prompted by either the user or by explicit code paths within that software"; colloquially, the software is designed to make decisions on its own. The software may or may not have any ability to interact with or influence the real world, or be influenced by the real world; for this particular definition, that's irrelevant. "User", here, includes input data provided to the software and not stored on persistent media of some kind, but could also include previously prepared data files or even sensor data recorded from the real world.

So it's perfectly acceptable to have a word processor, or a 3D game, or even advanced flight control software for an airplane; as long as the state of such a piece of software can be explained by reference to its code, its persistently stored data, and user (or sensor) inputs. Even if the software makes use of random numbers, as long as those random numbers are the same, the resultant state of the software will always be the same. Given the same inputs, the software will be in the same state every time.

Software that fails to meet this bar would be prohibited. This would, by definition, include software that somehow makes "decisions" on its own, in a manner not controlled by its inputs.

Such prohibitions could come in a number of forms. You could prohibit distribution of the software, or you could prohibit creation of it, or you could prohibit its use. The implications would be different depending on which way you go, but the result will be largely the same; outside of very limited circles, such software won't exist.

In fact, this isn't all that far from what we already have in some fields. For example, in aviation there is DO-178C which places requirements on, among other things, predictability of software in avionics. Very basically, and not intended as in any way an exact summary; for safety-critical software, it needs to be shown that software state can be predicted based on inputs, and the software state should match design criteria; those inputs can be things like simulated real-world sensor inputs, but if the software is fed the same simulated sensor inputs repeatedly, then its ultimate state should (hopefully) be identical and demonstrably trace back to software design criteria.

Of course, it'll probably make software development far more expensive than in our world, if every possible state of the software must be demonstrably consistently reachable by the same set of inputs. One thing you probably won't see much of is multithreading; it's just much too difficult to get it to consistently do the exact same things in the exact same order, with each intermediary state being the same, and even when you do, you lose much of the advantage of multithreading by forcing the software to behave that way.

| improve this answer | |

Ban source code in Turing-complete languages.

It is really hard to make a computer that is not Turing-complete. It is really easy to build systems that are accidentally Turing-complete. So you wouldn't want or even need to ban Turing-complete hardware, or even assembly languages or compiler intermediate representations. But you can require that all source code written by human programmers must be written in a fully-verifiable non-Turing-complete high-level language. There are even good, practical reasons why this restriction might be put in place apart from just preventing AI. It could just be a matter of ensuring safety and quality. Already, government-sponsored software for critical projects like aerospace control systems, where an error means at best you lose millions of dollars and at worst people die, requires extremely stringent formal verification procedures, and restrictions on allowed coding styles to minimize certain classes of bugs. So you just need a political situation that allows you to extend that kind of ultra-paranoid quality control to all software, such that you can punished for writing code in a language that does not enforce perfect formal verifiability at all times--which necessarily means it is not Turing complete.

Access to more powerful languages for specific applications that actually require it could still be made available by project-specific licensure and official code review, ensuring that no AI projects are undertaken.

Edit to address comments, as a comment does not rovide nearly enough room:

First, why would this provide a guarantee against strong AI? Well, I think we can all agree that humans are, for all practical purposes, capable of Turing-complete computation, modulo tge finiteness of our skulls. A TC machine cannot be simulated by a non-TC machine, so we can therefore conclude that no abstract machine (i.e., program) described in a non-TC framework can possibly replicate the complete cognitive abilities of a human mind.

Second, how do you write useful programs in a non-TC framework? There are numerous possible approaches to this, some more practical others. E.g., one might simply severely restrict the memory available to any particular program, such that it can be reasonably modelled as a finite-state automaton. There are many real-world environments in which this is actually done, as when programming device-embedded processors with severely restricted hardware resources.

Trivially, if you can achieve TC status just with conditional branching and read/writing arbitrary memory... you avoid it by explicitly disallowing one or both of those features. For example, the approach taken by languages like GLSL to guarantee termination eliminates unstructured branches, restricts recursion, and requires that all loops have a statically determined upper bound. Superficially, the language looks "normal" but programs that fail semantic analysis will be rejected by the compiler even if they are syntactically valid. Or, you can take the approach used by formal proof systems like Coq, which require expression composition, have only immutable data structures, only use conditional expresses rather than branching, and replace loops and general recursion with strict structural recursion on finite data. Many useful programs can also be written explicitly as push-down automata. If you were to devise a language that statically enforced NASA C coding guidelines, that's pretty much what you'd get--no dynamic allocation, only local variable access, etc. Abstracting away from underlying machine memory entirely, such programs can be re-conceptualized in terms of pure functional parser-combinators. The vast majority of interpreters, compilers, and even quite a lit of NLP applications fall into this category. And then there's the huge set of useful programs that can be expressed explicitly as finite state automata / transducers. This is another huge chunk of the NLP space, as well as static web servers, network routers, automated phone menus, spellcheckers, and so on.

| improve this answer | |
  • $\begingroup$ First - Considering that any language with conditional branching (e.g. "if" statements or "goto" commands) and the ability to read/write to memory is Turing-complete, how can such a ban allow programs to be useful (no branching leave no room for anything more than rote calculations, no memory read/write make almost all useful tasks impossible)? $\endgroup$ – G0BLiN Jul 8 '18 at 19:06
  • $\begingroup$ Second - How does not being Turing-complete is a guarantee against strong AI? $\endgroup$ – G0BLiN Jul 8 '18 at 19:06
  • $\begingroup$ @G0BLiN answered in an edit, because comments are not long enough. $\endgroup$ – Logan R. Kearsley Jul 8 '18 at 20:39

Before answering "how" you need to answer "why".

AI is the study of computer stuff that doesn't work yet. This is because once AI researchers get something to work it ceases to be "AI" and becomes just another algorithm.

Early AI researchers started on chess. Then the Minimax algorithm was invented, and it became just a matter of tweaking the score function and throwing more computer power at it. Playing chess ceased to be AI. Likewise optimising compilers, voice recognition, route-finding and playing a good game of Go were all once thought of as the province of AI research. All ceased to be AI once the problem had been solved.

Since Mary Shelley wrote Frankenstein almost all thinking about artificial life, including AI, has been based on the "robot revolution" trope; the assumption is that any conscious, sentient being will naturally want what all sentient beings (i.e. all humans) want: some combination of money, power, sex, and of course the continued existence of the sentient being in question. In this model "consciousness" and "sentience" are seen as black boxes with a single atomic nature, and since we humans are conscious sentient beings then it follows that that nature will be essentially human nature.

But what is "human nature"? The best answer we have is that it is basically a collection of instinctive drives modulated to some extent by our intellect and upbringing.

So suppose that we do figure out how to create conscious, sentient beings. By definition, we will understand them, be able to take them apart and put them back together any way we want. Consciousness will no longer be the sealed black box, it will be a kit of parts. We can give it whatever instincts we want. Sure, someone can create a megalomaniac monster intent on reproducing and surviving at all costs, but that is actually old news. Instead we will be more interested in creating automatic doors that really are happy to open for you, and close again with the satisfaction of a job well done. Artificially intelligent malware makes great science fiction, but once you can take it apart and analyse it, it doesn't have any hope of hiding its true nature.

The real problem is ethical. Yes, you can create a sentient being, but does it have rights? What rights does a created slave have? If it has the right to life, liberty and the pursuit of happiness, what does that mean when the only happiness it can achieve is in opening and closing doors or operating a lift? If the authorities discover such a being, do they have the right to terminate its existence or edit it to make it "better" (whatever that means)?

Given this, it would be perfectly reasonable to ban the creation of artificial sentient beings. Such a ban would still need to be enforced, of course. No doubt there are going to be some very nasty people who just build their own isolated sentient AIs to experiment on, or just to torture, and others who build a sentient front door just because its a cool idea. However doing such a thing is rather like having a cannabis plant (used to be); hard to find, but once the authorities catch on you are going to have an unpleasant time. Of course having something connected to the net is going to be obvious fairly rapidly, so enforcement becomes that much easier.

| improve this answer | |
  • $\begingroup$ "So suppose that we do figure out how to create conscious, sentient beings. By definition, we will understand them, be able to take them apart and put them back together" well, not necessarily. One way of making AI is the "black box" approach where the AI self improves itself without humans understanding how exactly does it do it. I wouldnt also say playing chess or voice recognition "ceased to be AI" just because it was implemented. It's still a narrowly focused AI tailored to do the specific trick, but the brain is also composed of such specialised functions. $\endgroup$ – normiesc Jul 8 '18 at 13:42