-2

Is voting needed? Can AI just look at all history and understand the needs of society? Humans do not have a strong record when choosing their leaders.

Such as Hitler was voted when he promised prosperity and special status to his idea of "perfect race" but ultimately became the most hated person in history.

Or like Mandela who asked for equal right but unfavorable history of ANC. There are millions of examples, what I am wondering is if we need to vote or can AI would make those decisions because they have access to all of human history to learn from, our mistakes and successes.

History can provide has many outcomes where under many different starting conditions people had voted for their leaders and what turned out. The biggest fault with any leader has always been, the extreme they go to with their ideas and most of the times lead to loss of life but AI's wont suffer from this ideological bias.

Based on today's technology can we have a technocracy society and remove voting?

  • 2
    Potentially very interesting question, but I'm going to wait to see if it stays open before writing an answering essay. In the meantime you might want to read en.wikipedia.org/wiki/Project_Cybersyn and related discussion of the "cybernetic society" – pjc50 Dec 5 '19 at 11:25
  • @pjc50 The hold was removed, I would be very interested to hear your thoughts! – new coder Dec 5 '19 at 11:46
  • 1
    Questions about future events are explicitly off-topic. You could try to make it on-topic in Worldbuilding, by asking questions like "Why would people oppose (or support) an AI-lead technocracy?" (be sure to check their FAQ for more hints on making that on-topic). – SJuan76 Dec 5 '19 at 11:46
  • Comments deleted. Please don't use comments to answer the question. – Philipp Dec 6 '19 at 13:20
  • 1
    This questions seems to be very speculative and much too broad. Satisfactory answers may fill volumes. Narrowing it down might be a good idea. – Trilarion Dec 9 '19 at 10:03
6

What AI are you talking about?

If you talk about a "true" GAI (General Artificial intelligence) that reaches human level intelligence/independence and potentially consciousness, then it will have the same issues as a human - it will have its world model, aims to achieve etc and may adapt based on new experiences and thus change and make "historical mistakes" as any human.

If you talk about the current level of trainable special purpose AI, the only thing these can do is "learn" to solve a given task. You need to still provide it with training data, i.e. tell it what is a good and what is a bad outcome. You, a human, need to define its goals. And all it will do is try to optimize for those goals. It will be incredibly hard if not impossible on top to "attach" that AI at the top of government in such a way that it can in detail govern without human middle men that might interpret the AI's decisions in their own way. Another possibly unresolvable difficulty is to make the decisions all-encompassing. It's not one categorisation task as output, it's a lot of policies, decisions and laws as output - in a way that they are understandable by humans. Without too high of an error rate.

On top, current AI does typically behave well for the area it is well trained for, but often can have unexpected totally non-optimal behaviour on corner cases or totally new input. Would we be happy with an AI that could trigger a nuclear war out of a sudden because someone calls it a covfefe?

All that only focusses on practical aspects and leaves aside the social components: For instance, democracy does not only serve as a (potentially suboptimal) way to make nationwide decisions, it is also about satisfying everyone's need for self-determination / participation etc. Monarchy on the other hand has a soap entertainment factor on the social level. A cold AI has neither. A human-like AI might perhaps have the soap factor as well, but would also have the same drawbacks.

4

Artificial neural networks are becomming great at solving optimization problems. Giving a large training set of data and variables which can affect that set of data, neuronal networks can easily be trained to learn how to set those variables in order to optimize for a certain goal.

But the question is: What goal do you want to optimize for?

Ideally you would want it to optimize for "happiness", but how do you quantify that? In order to train a neural network, you need some way to rate the outcome based on some measureable metric.

So what metric do you tell the network to use? Average income? Median income? Lowest income? Unemployment rate? State deficit? Tax income? Population growth? Live expectancy? Low working hours? Low crime rate? Military strength? Social mobility? A combination of all? What counts how much? And how about abstract concepts like "personal liberty" or "life fulfillment"? How do you convert those to a numerical value?

What actually does it mean to be happy? Do humans even know what it means to be happy? Different people have different ideas of what's best to achieve maximum happiness. The arguments between these people is what we call politics.

Even if we use more AI in government, we will still need humans to decide the goals for the AI.

  • When we vote for a "dear leader" he makes fake promises and based on what we understand is the "cost" we vote for him. Most leaders do not talk about the cost such as what would be the cost of tax cut or increasing the minimum wage. AI could give everyone all the answers to what it is proposing and what are the "cost" which would be "personalized" to their individual situation. That is what i am trying to understand if AI would create a more aware population about the policy as such we dont vote anymore we just give what as a society we want and understand the cost of getting it. – new coder Dec 5 '19 at 11:59
  • 5
    AI could give everyone all the answers to what it is proposing and what are the "cost" which would be "personalized" to their individual situation. No. Society is a chaotic system; and future predictions are broad estimates at the very best. There is no algorithm for predicting the future with certainty. – SJuan76 Dec 5 '19 at 12:27
  • 1
    Plus, you don't need an AI leader for that. If that technology were available it could be used to portray the cost voting for any candidate would incur. – Frank Hopkins Dec 5 '19 at 14:37
  • @SJuan76 I am not saying predicting future, leaders dont know the future either but leaders are bound by ideology, party, human condition, upbringing which adds a lot of baggage to the reasoning and leads to chaos. A algo does not care about ego, just that the loss of life is minimized and similar parameters. What those parameters are is up to the society to choose against what it wants to achive! Instead of electing the next hitler who promises employment but for Aryan we can have a model to think of all possible scenarios. AI model cannot be appeased or insulted, like humans can! – new coder Dec 5 '19 at 15:55
  • @newcoder OK. You ask an AI for long term prosperity of human race, it does not share our taboos. The utilitarian answer includes among other, uncontroversial ideas also speedy executions of criminals with poor rehabilitation predictions and eugenics. Then what? – Shadow1024 Dec 5 '19 at 16:47

Not the answer you're looking for? Browse other questions tagged .