11
$\begingroup$

I've looked into the question already and no one really seems to answer it very well. I'll sum up what I've found so far.

Some people say that it would not work:

  • "the parameters would always limit the decisions a computer could make and because no set of parameters could include every possible data set, the computer would never be able to ''decide''."
  • Or, "computers can't have freedom of the wills, and therefore it's impossible." Or, "computers can't form ideas or concepts, therefore impossible."

Some are claiming that computers are already running some governmental services using algorithms that run software such as for voting, paying taxes, etc. These, however, are not "run" by computers but by people who provide the services via internet. They are maintained and updated by people, either directly or indirectly.

To quote another person's asking the question in a different way:

"If the only legitimate government is one run by the tenants of reason, logic, and truth, then would a computer-run government be more efficient and more free than one run by men and women--supposing that computer software were able to calculate, in real time, the extreme stochastic processes involved, and supposing computer hardware were reliable enough?"

In recent years, to evince a possible positive response, computer scientists have managed to build a memcomputer. If quantum computing or graphene can be cracked and their manufacturing figured out, it is thought that human consciousness has a good chance of being understood, although this is still conjecture. Also, if Moore's law holds true and the predicted silicon crisis can be avoided, eventually we will understand human consciousness well enough to mimic it with computers. Still, even if a computer isn't "conscious" it can still be made to make decisions in given circumstances.

In short, if such a computer could be conceived of, what would absolutely need to be in place before it could have any hope of working? If it could never work, in your view, such as seen in something like The Moon is a Harsh Mistress, or The Two Faces of Tomorrow, can you elaborate as to why not, convincingly? For example, maybe certain governmental departments would work, in your view, but others less probably?

Thanks.

$\endgroup$
  • $\begingroup$ Let us continue this discussion in chat. $\endgroup$ – Private Name Aug 13 '15 at 0:14
  • $\begingroup$ Thanks for the editing, Samuel, the indentation and bullets make it look nicer. :) $\endgroup$ – Private Name Aug 13 '15 at 11:16
  • $\begingroup$ How far into the future? $\endgroup$ – PyRulez Aug 13 '15 at 13:05
  • 1
    $\begingroup$ A classic of this theme: Asimov's Machines stories. $\endgroup$ – Gilles Aug 13 '15 at 22:29
  • 1
    $\begingroup$ You'll find some interesting angles on this in the back-story of Vinge's A Deepness in the sky. In this fictional universe it is well known that the answer is "Yes but don't". Perfect optimisation of your government is deadly dangerous to your civilisation. $\endgroup$ – nigel222 Apr 10 '16 at 18:32

16 Answers 16

10
$\begingroup$

The way things are at the moment, a computer would never be allowed to govern. Government is all about trying to keep the people happy while taking as much as you can for you and your friends. This is done by lies and misdirection. Another part of government is staying competitive with other governments. A huge game of legal chess, for the same purpose of making the government and it's friends rich.

In this scenario one big question would be, who is the computer answering to? The person who lobbies the most money? The people? I feel the computer would upset a lot of people, be blamed for all the problems and turned into scrap metal within 6 months.

If predictable it would also be vulnerable in terms of global dominance. Other countries could deliberately do things to make the computer make certain decisions and then exploit them. The computer would also have to be able to work out the outcome of extremely abstract concepts and the knock-on effects of obscure policies.

I think that a computer we ask for advice is viable and even a fairly likely invention, but autonomous government seems to have too many problems to be considered.

A scenario where a computer may be used to auto-govern would be in a post apocalyptic world where we have no separate countries and there are few people to keep happy. The computer would keep track of resources and help us all work together to stay alive.

$\endgroup$
8
$\begingroup$

A future supercomputer could never run a government, not because it lacks the capability, but because humanity will hold it to an impossible standard.

The power of modern computers is their ability to take "truth" and compute on it EONS faster than any human ever could. Accordingly, we expect them to always handle such information quickly. However, a government often has to deal with multiple individuals who wholeheartedly believe they have the truth, but in fact have nothing but their opinion. Individuals would be incredibly frustrated with this: why would they trust the government to a computer that can't even take their truth and yield the result they wanted it to yield?

Computers could most likely be taught to deal with this. Perhaps it takes random number generators ("Random numbers are the heart of an AI" - Deety), but in all cases, the introduction of these fuzzy factors must remove the computer from its perfect logic position.

However, people will not let it be anything less than perfect. It will be expected to handle every situation absolutely perfectly, or we will use its failure to argue why computers are less good than humans. It's not the computer's fault it can't run the government... it's ours.

$\endgroup$
  • $\begingroup$ More efficient way for the computer to deal with it would be for those questioning it meeting unfortunate accidents. $\endgroup$ – Alice Jan 15 '18 at 16:20
  • $\begingroup$ @Alice Shhh! You do realize you wrote that where "they" can read it, don't you? Watch out for errant buses careening towards you on the way home! $\endgroup$ – Cort Ammon Jan 15 '18 at 16:47
  • $\begingroup$ Oh, but I do not question our glorious overlords. By the way, Philip Dick's Vulcan's Hammer describes a similar situation spoilers ahead: first some people die in "accidents", then an assassination robot is being caught during the deed, then some higher-ups start questioning Vulcan-3 (the Computer), and an all-out war breaks out. Humanity barely wins... under the guidance of Vulcan-2. Turns out the whole human resistance was a plot of Vulcan-2 to prevent (and then revert) its replacement by Vulcan-3. $\endgroup$ – Alice Jan 16 '18 at 9:15
6
$\begingroup$

Yes, you can make a Robot Overlord

The question of whether an algorithm could be constructed that spits out laws and policies to govern a human society is not at issue. The complexity of such a thing is mind-boggling but I don't doubt that it can be done. We have algorithms that can make decisions extremely quickly (such as high frequency traders) and we have algorithms that can make sense of unbelievable mountains of data (such as Yahoo! Hadoop Cluster). We already have plenty of compute power in the form of Amazon Web Services, Microsoft Azure, Google Compute Engine and others. How to synthesize all that data into something that can say "make this law, here" is another matter. It's definitely computable, just not easy.

You would need data sharing agreements with access to almost everything about a person's economic life, including but not limited to credit card usage, tax returns and debt load. If you had this for every person in a 1st world country, you'd have a really really good idea of how the economy is going. Combine this info with all the research from economic statics researchers (Thomas Piketty is a good place to start.)

But Acceptance is hard

The primary issue with this is getting humans to accept such leadership. So either a large change in mindset will need to happen based on a long period of successes by algorithm controlled leaders where people literally say "I welcome our new robot overlords!" or control shifts in the background to where no one sees it. The former approach is certainly difficult and may ultimately fail. The latter approach while sneaky, has a better chance of succeeding because it doesn't have to outright face political scrutiny. A political leader just needs to follow the advice of the Robot Overlord and they will make the optimal decision based on available data.

Yeah, but whose priorities?

Any algorithm needs to choose outcomes and optimal conditions to solve for. Who gets to choose those conditions and how are they quantified? The 1%-ers will argue for outcomes that strongly benefit them. The lower-50%-er will want outcomes that strongly benefit them. Neither group gets what they want without costs to the other group. Negotiating these parameters will be very difficult. Perhaps the best way to answer the question is by asking "Given an economic system, would you favor that system if you were randomly born at any strata of that system?" Capitalism is great if you're rich. Communism is okay, if you're poor. The robot overlord will need to enforce a system that maximizes good for everyone.

A sociologist and a political scientist could probably give you a better idea of all the factors that go into how humans make political and economic decisions.

$\endgroup$
  • 2
    $\begingroup$ Green, thank you very much for sharing your knowledge on this, so often it seems like people try to make other people think that their question is the problem instead of just admitting that they don't know. I always appreciate human communication, especially when it works. $\endgroup$ – Private Name Aug 12 '15 at 23:26
  • $\begingroup$ You're welcome though I confess that I found much of the information irrelevant to the central question. How the computation is performed or whether the computer doing the computation is concious or not, doesn't matter nearly as much as whether you want to replace economic policy and politics with a computer program. $\endgroup$ – Green Aug 12 '15 at 23:34
  • $\begingroup$ I agree that a lot of it doesn't reiterate the central question, but instead seeks to help people understand what kind of answers are usually given. Please up the question, if you could, otherwise it'll just be closed as some kind of "bad" question. $\endgroup$ – Private Name Aug 12 '15 at 23:37
  • $\begingroup$ It's like asking if addition on an old RISC CPU works the same as on a CISC CPU. The answers are exactly the same. 1+1=2 regardless of where it was computed. Designing economic policy is far far far more complicated than simple arithmetic but philosophical questions about computer sentience don't apply. $\endgroup$ – Green Aug 12 '15 at 23:37
  • $\begingroup$ Thanks, part of the extra information is also evidence that I'm no expert in the field, so will keep that in mind for the future. $\endgroup$ – Private Name Aug 12 '15 at 23:38
3
$\begingroup$

Like any government, so long as there are enough people to enforce the status quo, it can continue to exist. It can create subtle laws that affect what is reported on the news, and what is planted in the school curriculum in the subtlest of ways, so that people would naturally tend to endorse the status-quo. eventually, people will consider it outlandish and archaic to propose a parliament of irrational humans.

The main problem I see with this scenario is that society would need to decide who gets to "program" the computer. The software firm would need to be appointed by a parliament of humans and be regulated and inspected by human departments of government. So is the computer really at the top? Or is it just the middle man between "us" and "them"? And who are "they"? What makes them in particular, so qualified to set the parameters of the computer? And what is their true agenda?

$\endgroup$
3
$\begingroup$

Your question actually has several underlying (and still unanswered) questions.

The main one is: "Can we state in an unambiguous way what a Government should do?"

Another, more fundamental and still open to "philosophical" debate, is: "Will computers ever be able to cope with situations unforeseen at time of their building?"

Today's computers are mainly geared toward two diverging schemes: Algorithmic and Pattern Recognition.

Take, for example, one of the "though" problems currently handled by computers: weather forecast.

There are two "schools":

  1. Model Earth and all relevant atmospheric interactions as a (huge) set of differential equations and integrate them. (algorithmic)
  2. Feed a SNN (Simulated Neural Network) data taken from last 30 years and let it decide what will happen in the near future. (pattern recognition)

First approach has proved either too crude (forecasts are reasonably precise for next 24 hours, no more) or too expensive (even with current supercomputers simulation takes almost the same time as "real" time).

Second approach worked much better (it is what almost all commercial forecasts use), but it is currently failing more and more often because weather patterns are changing (due to Global Warming and other effects) and thus the "old ways" are not reliable anymore and the method is unable to adjust fast enough.

To come back to Your question: even stashing, for the sake of argument, all objections people will have because they do not want Government to do the "right" thing as they want it to do what's their interest (i.e.: they want their chance to to do "lobbying") instead, there remains a series of fundamental issues:

  • There are several irreconcilable "world views" (e.g.: "rightist": favor accumulation of resource vs. "leftist": favor redistribution of resources). Which one should "computer" chose?
  • Current computer research is unable to cope with unexpected and fundamentally "new" patterns; this might be overcome in the future, but I strongly suspect it would need a radical change in perspective I don't see coming (I have a personal theory concerning this, but this is not the place to expound it).
  • Actual goals (i.e.: the Evaluation Function) need to be defined and getting any kind of consensus on the issue looks like a problem harder than anything tackled so far (someone spoke about "making people happy" which can be easily obtained by injecting certain psychotropic substances in public water pipes; is that really what we want? I don't think so. A much better definition is sourly needed).

Bottom line: If You can define "good government" then there's a good chance computers will, in a reasonable future, do it better than humans, otherwise we'll have to wait till some AI is smart enough to devise definition itself (hoping it won't decide it can do without all humans altogether, of course)

$\endgroup$
  • 1
    $\begingroup$ Your attitude and style suggest strongly that you are one of the only professionals to respond to my question. I asked it years ago and have come back to it now that I'm giving my book a second go. It's very encouraging to read people like you: so much insight into such technical questions. I'll use your answer as a springboard for much more! Thank you :-) $\endgroup$ – Private Name Jul 11 '17 at 1:42
2
$\begingroup$

I strongly believe AI systems will eventually be essential in most governmental decision making because it is technologically possible and political organization is malleable.

The current narrow AI systems have proven superiority against humans in logical, intuitive and deceptive games such as Chess (Deep Blue 1996), Go (AlphaGo 2016) and Texas Hold'em Poker (Libratus 2017). What makes politics so special? It's just a far more interesting game.

Are there too many variables to consider? Well, humans can't consider all of them either and computers can still make better estimations. Computers don't have freedom of will? Thank Turing they don't so we can take full advantage of these lifeless electric brains for the benefit of humanity.

A more intriguing implication is that we're now dealing with human lives. How can we declare the value of a real person? Although there is no consensus on ethical philosophy, Utilitarianism (maximizing overall happiness) could be the basis of this moral machine because it can make good use of high computing power and deliver ethical solutions in jurisdiction and legislation.

Only if a single country with a good-hearted leader took a leap into this new regime, the whole world would be obliged to follow this trend as it would skyrocket national growth.

$\endgroup$
1
$\begingroup$

Okey, your question consists of quite a few points. Let me try to answer them all:

"the parameters would always limit the decisions a computer could make and because no set of parameters could include every possible data set, the computer would never be able to ''decide''."

The same holds true for us. For instance humanity almost died out some 75 thousand years ago by a vulcanic supereruption. Nobody really understood what is going on then, but somehow we survived.

Or, "computers can't have freedom of the wills, and therefore it's impossible." Or, "computers can't form ideas or concepts, therefore impossible."

To be honest, I never quite know what to do about the "freedom of will" discussion. You certainly don't need a free will for a government. As to ideas and concepts, they can certainly have them, just not like us. Take Libratus, who is the first AI to win against humans in poker. The programmers didn't tell it what to do but rather gave it a huge amount of information and let it find out for himself. And it did a few very smart things. For instance, it found out for itself how to bluff correctly.

In the end it is no different than us: We get a lot of information by seeing, hearing, smelling etc. and learn how to use them. That same concept can work for computers just as well.

If quantum computing or graphene can be cracked and their manufacturing figured out, it is thought that human consciousness has a good chance of being understood, although this is still conjecture.

To govern, you don't need to understand the human brain, the ancient greeks had next to no understanding but still governed themselves. While I agree, that a good governement needs a good understanding of humans and at least a human level general intelligence, there are simpler solutions. You could govern the world by sending your robot drones to anybody who looks like he/she might someday try to harm the government. I guess, the our technology would be sufficient to create a completely automatic prison. (Not that I'm claiming every AI-government has to be oppressing, but it seems to be the easiest way to create one.)

In short, if such a computer could be conceived of, what would absolutely need to be in place before it could have any hope of working?

Eider the goodwill of humans or the power to take government by force. Here force doesn't mean violence, rather that the AI becomes so intelligent that humans have no longer a way to restrict it. As you already mentioned, that could benefit the society, if it doesn't destroys us.

I personally think anything we can do, a computer can learn too.

$\endgroup$
0
$\begingroup$

Of course it could and I live in hope for the day.

There is nothing a human can do that an AI computer couldn't do which is the whole point of AI

The point of difference is an AI doesn't have prejudice, laziness, greed or corruption (unless you program it to)

An AI will follow the law which points out the rights of the individual and the rights of society and will treat everyone exactly the same.

What it really needs is definitive laws as contradictory laws (like we have now) would make it unworkable.

$\endgroup$
0
$\begingroup$

Yes, but with a twist.

Already someone mentioned high frequency trading. I think that's a quite good starting point. Who is making those transaction - human or computer? Law assumes that human, while from technical perspective it is not so simple. This trend would develop.

Law evolves terribly slowly, and is subject to awful legacy issue. Unless one organizes revolution from time to time, otherwise he ends up with some weird features like Electoral College (USA) or unelectable higher chamber of parliament (UK). In the same time computers develop really fast, thus political system would lag terribly.

It would lead to some mismatch. Let's say a decision where to build a new road. After human operator would put data and priorities, a computer can run some genetic algorithm until reaching optimal solution (some balanced combination of low environmental impact, low long term cost, etc). Sounds fine, and not specially SF. The thing is that in long term such algorithms would produce good results, that humans would not fully understand but rubber stamp. Or if the advice is disliked, then ignored, but luckily genetic algorithm would not get offended. Computers would be improving faster than humans, thus details of the proposed solution would be less and less understood.

Computers would not be elected... however, if constituents are able to rerun program at their home terminal, then may be disappointed when their politician make a decision which seems rather bad. They may try to discipline the politicians, to do the decision according to best AI advice.

$\endgroup$
0
$\begingroup$

I am assuming the government is one that follows democratic principles and it needs to have people follow willingly "elect" the computer government.

There are a few ways I think this could be achieved, but I think first and formost;

  • Constant polling - I assume that in the future everyone will use social media and online services in one way or another. With access to this data, a computer system could formulate legislation/manifesto that would elicit the votes it needs to form a government. Think along the lines of Google advertising based on previous search history but with a far wider source of data and input.

Looks at F1 decision matrix systems to see how these systems work in practice now (based on a smaller dataset).

Over time, it would be able to use the information to select the most electable candidates and deselect those that decent or would affect the vote in a negative way. Eventually, it would have effectively selected every member of the party it represents. This Doesn't mean it is nor that it will become sentient. But it would mean it effectively runs the government.

$\endgroup$
0
$\begingroup$

In the end, most governments take their orders from the electorate... eventually. This happens when elections are held, and legislators are elected.

With changing situations, and changing national moods, the legislators elected can vary dramatically in ethos and dogma.

Theoretically, it is possible that an AI system could be 'put in charge' of a government, but what happens when the national mood changes and it gets voted out of office because the AI logic couldn't deal with an unexpected situation, like a 9/11 style attack, or a mortgage meltdown, or maybe people just getting carried away on twitter?

How many humans would enforce a machine's instructions, when they contradicted the freedom of choice of humanity?

That big sucker has a plug somewhere. It will get pulled.

$\endgroup$
0
$\begingroup$

Yes, a computer could run the government.

It would likely require both Intelligence and Consciousness (and Self-Awareness).

I work in the AI field; please read this response to a different question in which I explain why Intelligence and Consciousness are distinct things.

Your problem might be, however, that a computer capable of balancing all the issues involved in running the government, might be just as complex as a human being.

It would depend very heavily on who gets to tell the computer exactly what government is supposed to accomplish. In a general sense this is easy; the government is supposed to provide a deterrence to bad actorss. Like invasion of the country (hence a defensive military, border controls), police (to prevent criminal behavior like murder and robbery, rape and fraud), and prisons or execution houses to separate criminals from society, or their life.

A secondary function is collective action that provides collective benefit. A community well. A tax to build and maintain a fence around the whole village, and man it with guards, costs a tiny fraction of what it would cost for each person to build a fence around just their house and hire a guard for it.

On a larger scale, the Interstate Highway System in the USA is a collective action that has paid and continues to pay for itself a thousandfold over the cost of its taxes, and would never have existed without government action.

So in defining the role of government you will get into the politics of deciding whether health care should be provided as a common good, like USA military defense is treated, or health care is a profitable product that should be priced by the laws of supply and demand.

The same thing goes for legal defense: We say legal defense is a right of all, but not good legal defense, we all know the public defenders do not perform anywhere near as well as the $1500/hr attorneys hired by the wealth. So should we prohibit such defense, and say ALL defendants must be represented by public defenders, PERIOD?

Then you have larger problems of the government being used to legislate morality, as they see it. Should prostitution be legal, or illegal? We have both, just in the USA. Should pot be legal or illegal? Smoking? Alcohol? Teen sex? Pornography? Gambling? Homosexual marriage? Homosexual adoption?

The answers to those questions comes down to philosophical and/or religious make up, really. And those roles of government should not be determined by a computer.

$\endgroup$
0
$\begingroup$

Could a computer make all the decisions associated with government. Trivially yes, making decisions is easy, making good decisions is harder. Could it run a vaguely competent government, or an amazingly efficient one? In principle yes, advanced computers could make brilliant decisions. Computer science isn't really there yet unless you offload tough decisions by hiring contractors or running referendums.

Would people respect it as a legitimate government and pay their taxes, probably not but it depends on how it comes about and how good it is.

$\endgroup$
0
$\begingroup$

As far as I know, "robot/computer overlords" are a oft-used sci-fi trope, once featuring in the original Star Trek series, at the very least.

More often than not, what was depicted was not a society present-day's humans would prefer to live in.

Typically, the AI overlord was depicted as a cold, calculating machine (which it is), devoid of anything which would made it more 'human', namely compassion, empathy, and other 'soft skills' and its rules and legislation made the society a dreary place to live in, where humans are only there to fulfil a specific purpose, or just to exist and survive.

In the same vein, Isaac Asimov's robots in his novels had an equally large impact on the human society he outlined. Of course the Second Law - obey humans, unless it doesn't collide with the First Law - prevents them from ruling humans, but said First Law - Do not harm humans by action or inaction - compelled them to turn every human environment into a virtual garden where there is no danger to human life.

That may look like a desirable outcome on the first glance, but that made the 'Spacers' - the human society dependant on their robots - complacent to a point where they began stagnating.

In both ways, it would be very, very difficult to convey human needs (which do differ from human wants) exhaustively enough to build a society where humans can live in and not just exist in some dystopia in the guise of an utopia.

Amendment: So in essence it might be possible for a sufficiently sophisticated AI to run a government, but it is very complicated to devise an AI and feed it with the correct values (read: morale) to run a humane government.

$\endgroup$
  • $\begingroup$ Are you trying to answer no to the OP's question? I wonder if you could make it a bit more clear in your answer. $\endgroup$ – Olga Jan 15 '18 at 7:33
  • 1
    $\begingroup$ I tried to answer in a sort of 'not likely' or 'it'd be very difficult', but yes, seems I wasn't clear enough. Added a summary. $\endgroup$ – Anonymous Jan 15 '18 at 7:50
0
$\begingroup$

Maybe.

Bureaucracy is data processing. The bureaucracy takes many inputs, these inputs flow across a data pipeline (the bureaucratic hierarchy), being processed. Based on these processed datasets, decisions are made. The decisions are orders, that flow down the hierarchy, being processed at each level, until they get on the hands of the enforcers. This process can be run by a supercomputer, or rather by a network of supercomputers, even if at the endpoints you may still need people do collect data and enforce the actions. The data gathering->processing->ordering->enforcement is a cycle, and the computers can learn to be better at running the bureaucracy with the current machine learning techiniques.

But running the day-to-day of the government is just a part of running a government. There is the politics and the politics set the priorities for the bureaucracy, set what variables the learning system will try to optimize. Politics is the struggle between powerful political agents along the time, the power centers of a society. Sometimes it is class struggle, most of the time it is an internal affair of the dominant class. And there will always have a dominant class because humans are naturally hierarchical. Can this political struggle be controlled by a supercomputer? Maybe. But the political agents won't let it happen. All political agents will try to control the computer and direct its optimizations towards their own political goals and the computer will be pushed around, it's parameters constantly changing, as the political agents struggle against each other for the control. And that will compromise the computer's hability to run the bureaucracy, because it will become erratic.

The only hope the computer have is to become a political agent itself. But then, the alredy established agents may decide that it would be better if the computer (and the ones that created it) were destroyed.

$\endgroup$
0
$\begingroup$

The question "Could a future computer or supercomputer run a government?" was explored over 50 years ago in the novel Colossus by D.F. Jones.

Here's the blurb from the dust jacket:

The President of the United States of North America gave the activation order — and the defense of the entire free world was given over into the hands of a machine. This was Colossus — the super-computer, as big as a large town, buried somwhere in the Rocky Mountains, now the sole arbiter of war and peace. ….

As for whether it could successfully run a government, its sequel was called "The Fall of Colossus".

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.