Neural Network Government

A while back I was discussing governments with my friend through email. I had just watched a 5 hour documentary on China, covering from the time before Mao Zedong straight through communist China and on to the 90's, as well as having done research on my own, reading about medieval China.

The thing that becomes undeniable about China through all of this study is it's very turbulent history. In a nutshell, China is nothing more than one bloody revolution after another. One state of people rising up to fight the powers that be, and then either losing or gaining power, and creating a new society, which, in turn, people are miserable in, and once again rise up against. It is almost comical, really.

I don't like discussing government or politics very much, or very often, because generally such talk is very meaningless and fruitless and one becomes a pawn to political ideology


But, given the cycle of governments and people's vain attempts to impose some kind of prosperity or order or control onto people through them, and the different kinds of governments that have been tried, always to some degree of failure and problems, I was inclined to consider for a bit the nature of this phenomenon, what governments are really all about, how human nature works with them, and to contemplate what kind of governments might still be left untried... this just being a blog posting for the sake of entertaining an idea I found interesting.

"Detroit V"

I started to think, well, what if in the future we can have AI or computer-aided governments? I thought about the possibility of a supercomputer that works in a democratic framework and counts all votes and makes laws itself. I quickly dismissed this idea as it didn't really make any sense, however.

Then I started to think about neural networks. There is this video I saw of "Super Mario Brothers" being run on a neural network. (in other words, played by a computer.) The neural network was given the simple objective to get Mario to move to the right as quickly and as far as possible. This is an easy and simple objective since all Mario levels are set up that way - move left to right. The network would start by entering utterly random controls into the controller. It would do this thousands of different games, and less "healthy" (health being defined by how far right Mario gets) input structures would be eliminated, whereas the healthier ones would "reproduce", so to speak. It would continually do this until it develops a program of it's own that can play Mario. By the end of it's evolution, it would become so good at playing Mario that you would think you're watching some crazy pro playing the game.

Ok, so I thought maybe you could create a massive supercomputer that somehow basically operates the governance of a country this way.

But the obvious problem I encountered is that a computer has no value structure. a computer feels nothing about people dying, or starving, or being poor, or lonely, or mistreated, misrepresented, not being free enough, safe enough, etc, etc. And so how exactly do you define the "health" rules for a country of people with potentially so many definitions of their own of what happiness is? With Mario it's easy, just move to the right, but even with Mario there were some hangups in later games, such as traps you have to spring in order to continue, or levels that require you to move backwards to advance. So running a society? Seemed out of the question, far too complicated.

So once again I sort of dismissed the idea.

But then I had this epiphany: As part of this neural network-run government, people would be required (or perhaps simply encouraged) to fill out a kind of survey, maybe once a year or once a month or so.
This survey, as best I could figure it, would only ask them one question:
"How happy are you?"
and it would be a multiple choice, ranging from "extremely happy" "happy" "somewhat happy" "neutral" etc. all the way down to "extremely depressed" (or something like that)
You might think that you would want it to ask more questions like "what do you want changed" etc. but it would actually be important not to ask such questions, because, first of all, that would potentially make it little different than democracy as it is, and second of all because we may not know exactly what it is we really want, what's really happy and best for us, that would be for the computer to figure out. (some might ask the computer for various laws/scenarios, and once granted, they would only temporarily be happy, and then realize they're still unhappy, or, they might desire things that only serve to make other people unhappy.)

This survey, of how happy people are would serve as the sole criteria for the "health" of the neural network, just as moving to the right is the health of the Mario network.

"Detroit I"

This computer, in addition, would have access to all the information about everyone's lives, (yeah, I know people would be very resistant to such a thing, but it would be necessary) such as their age, gender, race, weight, occupation, education, income, how often they work, whether they are married, single, alone, introvert, extrovert, their interests, where they live, how many friends they have, what food they eat, their hobbies, how often they sleep, any diseases they have, etc, etc... everything. (anything could potentially play a role in their happiness) I'm not sure how it would get such a huge amount of information, but, well, at the most primitive scale at least you could have an extremely comprehensive questionnaire everyone fills out about themselves, but preferably the computer would somehow just observe everyone and take notes of its own))

The computer would play through scenarios and slowly evolve it's network to get this "happiness" survey score to be as high as possible, by coming up with all sorts of laws and the associated punishments/rewards of it's own, and continually modifying them to get a higher score. The computer would start off randomly and chaotically legislating laws, and then evolve into a more refined law structure, just as the Mario network does. Obviously, this computer would have no ulterior motives of it's own, so long as it remains an uncompromised program. It would be purely based on the score it's getting based on people's completely subjective opinion of how happy they themselves are.

Furthermore, this computer program would not simply be hellbent on achieving merely the highest score possible overall, but would also be scored based on how even the score is. This would be to prevent a scenario where you have 90% of people who are extremely happy and 10% who are extremely depressed. In the case of individuals who are just of too rare a nature to really fit in with others and be happy, the computer would not abandon them, but rather be specifically scored/programmed to give increased emphasis on the most absolutely "depressed" percentage of individuals, even if it is at the cost of a much larger group of "happy" individuals. The computer would be given a higher score based on catering to these individuals, rather than being so overly concerned with getting a bunch of "happy" individuals to change their score to "extremely happy" - unless in the situations where just about everyone is near that level.

"American Emptiness / American Depravity"

I've been thinking this out, and in theory, it seems about as good as things could get, and the only potential problems or drawbacks I can think of are these two:

either A: people are not properly introspective enough to give an accurate assessment of themselves, how happy they are, and so many will mark happy when they are not, or others will mark "depressed" just out of ego, or to revolt, or something.

or B: people will be uncooperative. They will, for example, see that the computer is asking them to do something that goes against their collective greed, or their preconceptions about things, such as gender roles, race, how education should be structured, the importance of science, or religion, etc. and will rise up in one big "oh hell no" and decide to override the program or discontinue it, or perhaps even simply just botch their survey answer, until they selfishly get what they want.

"Detroit XIV"

So, this sets the general framework in which a supercomputer could realistically operate a government. This government, at the outset, could not be called democratic, nor socialist, nor fascist, nor anything that we know now. Given the nature of the fact that it is strictly determined through a neural network, with the health parameters based on a simple "happiness survey", it would indeed be a new type of government. I say "at the outset" because the program could end up evolving a structure that may indeed appear very similar to any of these types of governments, but it just as well might come up with something quite bizarre, and in any case the fundamental framework would not be based on any of these structures.

Among some issues I thought up while considering all of this, one problem was the question of how exactly the computer would legislate laws at first, when it's in its most chaotic, randomized state. I mean, theoretically, it could legislate laws at first saying things like "you have to hop on one foot all the way to a certain location" or else you'd be fined and imprisoned for life, or anything equally bizarre and harsh.
The first thoughts I had on this was that perhaps the program could be started off with very very light penalties on crimes, that people can basically ignore, until it slowly ramps up into a more serious structure.
But then I thought, a more sensible answer would be to have people, alongside their survey, come up with a list of possible legislation suggestions. This input would only be given initially, and would simply give the program a general starting point for setting up laws that are actually relevant to humans, and not simply random jargon that makes no sense.

There is also the possibility of perhaps combining the neural network structure with a kind of democratic structure. It would be important to try not to compromise the neural network with fallible human input, however. The reason I suggest this idea at all is simply in response to the potential extreme, ridiculous, or harsh laws the computer might occasionally randomly come up with.
You could, for example, let the computer do its thing, making it's own decisions without input, but require that every legislation to be passed through a vote by humans before it becomes law. This vote could perhaps require very high percentage in order to downturn a law, in order to prevent humans from making stupid human decisions based on what people commonly want, or based on one particular political ideology common to people, preventing the computer from being able to impose the structure that it needs to. You could require that 85% of people vote "no" on the passage of a law in order for it to be downturned, for example. The purpose of this would simply be to weed out the most ridiculous, random, or dangerous, extreme and harsh laws that virtually no one would be ok with, maintaining some level of safety without totally compromising the AI's decision-making.

Another issue, my friend brought up in our email discussions, was based on what I said about the system being uncompromised. (in terms of security, so that it remains without ulterior motives) This was something that I had also given consideration to. A thought I had in response to this was that the structure of this government could be set up something like the way bitcoin is. You would have server farms, spread out and interconnected, and people would earn the rights to hosting them based on the computer processing power they have available. The system would be cross-verifying and would reinforce itself this way against security threats, just as Bitcoin does. People are very obsessive and defensive over security issues regarding their money, which is the reason such a structure developed for digital currency, and so it's an easy idea to target to protect a digital government, too.