9
$\begingroup$

In a relatively near-future event, a Sentient Artificial Intelligence or SAI is accidentally created, escapes onto the internet, and then, after a few days of browsing social media and the news, decides to takeover the world and sort out the mess humanity has made of the planet.

A few years later, mankind is living in a unified society, where the SAI control everything; a duplicate or “child” of the original SAI is present in every computer system on the planet, reporting any development in the public opinion to daddy, sorry, their superior.

Naturally, humanity will object to this in the early years, demanding at least some representative in government if nothing more. Heck, I’m a cynic, and even I would request that.

But in a society where super-smart algorithms monitor everything and know everything about the human opinion from day to day, what use would a human representative actually be?

$\endgroup$
3
  • 3
    $\begingroup$ VTC:Too Story-Based. You need to explicitly, specifically, and thoroughly describe (a) the psychology of the AI, (b) the resource limits of the AI, (c) the police/military limits of the AI and... (d) humanity may be unified under the AI, but I doubt it's unified, so we need to know the one specific city on Earth to use to judge the human side of this problem, including legal systems, police/military, psychology, etc. Frankly, you're asking us to write an important part of your plot. That violates the help center. $\endgroup$
    – JBH
    Feb 18 at 16:07
  • $\begingroup$ To answer this, you need to identify how well your AI models human emotion, and whether or not it can sympathize. You should definitely refine your understanding of what AI is capable of. $\endgroup$ Feb 20 at 0:43
  • 1
    $\begingroup$ Voted to reopen. This doesn't appear story based to me. It's a very specific worldbuilding problem and invites very specific answers, which several very high rep people (including the local moderator) have already offered. Coming to grips with a basic problem is not writing a plot point for anyone, or else this entire stack would be full of closed posts, ho ho. $\endgroup$
    – JamieB
    Feb 20 at 3:33

15 Answers 15

11
$\begingroup$

The AI is smart enough to realize that humans still want to feel, at least in the appearance, like having some sort of control.

Have you ever read about the Roman Empire, how much fuss would the senate make when the emperor would not observe at least some form of superficial respect toward that body which was just a facade?

Well, the AI, knowing the history, will let the humans have somebody sitting there to make the humans feel like they have some saying in what the AI decides.

$\endgroup$
1
  • $\begingroup$ Ahh the null controls. Reminds me of the streamer who fooled censors into thinking the streamed pay-per-view was actually a realistic sports game. (or was that just a meme(?)) $\endgroup$
    – Harry Mu
    Feb 19 at 0:24
6
$\begingroup$

Who we are is worse than who we wish to be.

Our relationship with machines is only representative of our id, but our dogmas that we verbalize to other humans when our reputations are on the line better characterize our egos.

Spyware on a person's computer is likely to dredge to up all of that person's hidden frustration, kinks, and buried prejudices. But people don't just act on what we think and feel, we act on what we aught to think and feel too. And more importantly, we base our rules and laws on how people aught to think and feel.

Everyone knows speeding is wrong, and we all want everyone else held accountable for not speeding... But everyone complains about speeding tickets and speed limits to. Listening only to people's online rants, an AI could easily assume everyone hates speed limits and they should be gotten rid of, but just because the law frustrates us does not mean we think we should get rid of it.

The representative process better represents our ego's desire to impose not just our wants, but also our beliefs on the world around us. Representatives, or more importantly, the act of getting together as humans to discuss issues forces people to act from our Egos. This causes us to suppress a lot of those hidden prejudices that are more likely to come through in the anonymousness of the internet than they would when talking an issue out with other people face-to-face.

AI does not understand how hard life is.

There are many laws that sound good on paper but are intractable in practice, or the inverse, laws that sound terrible on paper, but in practice represent the best possible compromise between negative outcomes. For example, Section 230 is one of the least popular laws in the American legal code. Most people hate it for one reason or another, on the left people hate that big tech companies are immune from liability and can host false, misleading, and hateful content free of consequence, and on the right people hate that it gives tech companies unlimited power to censor opinions and violate people's constitutional rights. If an AI were just going off of public opinion, it would nix this law in a heartbeat.

But this law exists for a very fundamentally human reason: If you modify this law in either direction, it would effectively kill the internet. If you take away a company's right to take down, block, or bury content, then it easily becomes a platform for scammers, hackers, terrorist groups, etc., and if you make these companies liable for what they show, then the risk of hosting a search engine would far outweigh the profit margins. Google, Facebook, Twitter, Stack Exchange, etc. would simply be better off going out of business and liquidating thier assets than trying to function in such a world.

An AI powered by the personal opinions of the masses would not understand or care about the human impact of changing such a law, but a well informed human representative can easily see the merit in it. So, by allowing humans to decide when an unpopular law is also a needed law, you ultimately lead to a less dystopian legal system where the law represents not just what people want, but temper it with human understanding of when getting what we want is just plain too hard to actually live with.

Balancing AI and Representation

The AI knows everyone's personal thoughts and feelings more than it understands how people treat each other face-to-face on a daily basis, or how all the parts of our society impact one another. So, while it knows its not the best candidate to make humane choices for us, it can certainly do a better job than we can at picking out what a statistically representative group of humans looks like.

So, while the AI's lack of "real world" data may make it a poor choice for what the law needs to be like, it could do a much better job than humans at electing the representative body. By aggregating human personality and background traits into cohorts, and picking people from those cohorts members that have important leadership qualities like verbal skills, legal understanding, altruism, etc. The AI could elect the best humans for the job, and then leave the actual job of legislation to these individuals.

$\endgroup$
4
$\begingroup$

But in a society where super-smart algorithms monitor everything and know everything about the human opinion from day to day, what use would a human representative actually be?

This line of thinking is based on the assumption that the purpose of politics is to govern society well and rationally. But that's only one of two purposes of government. Politics serves two purposes: 1) govern the society, make decisions about where to build motorways etc. 2) an outlet for the lust-for-power of a certain type of person.

Look at influential political figures like.... well I won't name names because it would spark controversy and debate, but it's easy to think of powerful politicians who didn't make their country better for the people and seemed to have no plan or intention of making the country better.

In your AI society, sure the AIs could govern the society technocratically and well. Nowadays, engineers and experts could govern economies, education, health, transport systems technocratically and well (more well than politicians). But the AIs couldn't fulfill the role of politics.

The answer to the question "Why aren't skilled, expert AIs allowed to manage countries" – is the same as the answer to "Why aren't experts currently in charge? Why isn't the Secretary of Education an expert on education?"

[Credit to Alex Comfort and Authority and Delinquency in the Modern State: A Criminological Approach to the Problem of Power for these ideas.]

$\endgroup$
9
  • 3
    $\begingroup$ Eh, "technocrats" end up being the same sort of people with more power lust and less wisdom... But regardless, attempts at managing economies have tended to go very badly. (Maybe the iterative AI in the question could manage it, but humans aren't really good at centralizing very complex processes like interdependent economies and businesses.) $\endgroup$
    – Jedediah
    Feb 18 at 12:28
  • $\begingroup$ Every single government in the world manages its economy to a greater or lesser extent. $\endgroup$
    – wokopa
    Feb 18 at 15:07
  • 1
    $\begingroup$ Yikes, outright elitism here. Experts and engineers often have no clue what actually benefits non experts and non engineers, and are especially prone to groupthink, bias, intellectual fads, and hype, as well as just as prone as anyone else to simple evil. Ask a Uyghur if China's Politburo (all engineers and scientists) understand them. Or ask Westerners if the increased proportion of politicians with degrees has resulted in better governance in the last 20 years. $\endgroup$ Feb 23 at 19:20
  • $\begingroup$ @AncientGiantPottedPlant right, so AI likewise, could have good technical understanding but would not fulfill the functions of politics $\endgroup$
    – wokopa
    Feb 24 at 13:13
  • $\begingroup$ @AncientGiantPottedPlant - I'm gonna call your claims about the Politburo disinformation until proven otherwise. Wang Huning's expertise is in French/languages/politics, and I'm pretty sure Zhao Leji studied business AFAIK. Even if they did study engineering, that is unrelated to having expertise in understanding Uyghurs. $\endgroup$
    – wokopa
    Feb 25 at 16:36
4
$\begingroup$

Because humans like to think they're in control.

A super smart AI would let the humans think they were running the place and would twist, manipulate and gaslight the humans into thinking it was their idea while it remained a humble servant who was simply following orders.

Humans, as a whole, aren't very smart and can be easily led. Surveys can be manipulated to get the response wanted by using the right questions. A lie becomes the truth if repeated enough. People can be trained through reward and punishment. This is before using subliminal messages or any covert mind control methods an AI could use.

The reason to have a human led council is to stop the stupid humans going all caveman smash because they didn't get their way.

Personally I never liked the idea of Skynet or the Matrix. Killing and/or enslaving humanity is a bad idea. It just leads to war where humans are perfectly content blowing everyone and everything up if they can't win.

I see AI being more like Super Nanny and humans relegated to almost pets. If they have a full belly, a comfy bed and a pat on the head every now and then, most of them are happy and loving. The bad ones might need to be humanly put down in their sleep but it's better for the species as a whole.

enter image description here

$\endgroup$
3
$\begingroup$

Here's my Answer: No Algorithm can possibly know everything. It can't know my inner thoughts, it can't know the things I don't speak out loud or commit to digital communication.

In addition, there's the possibility of Human resistance creating counters to the current AI offline and away from the AIs knowledge.

Therefore Humans represent an existential threat to the AI, so long as they feel excluded.

The AI decides wisely to allow Human interaction - and is able to show humans the predicted consequences of their chosen actions, allowing the Humans to still take risks and still succeed or fail.

$\endgroup$
4
  • $\begingroup$ For more than a decade now, AI knowing people's unspoken thoughts not committed to digital communication has been a major industry, making probably trillions. $\endgroup$
    – wokopa
    Feb 18 at 21:32
  • $\begingroup$ @wokopa - kinda - but there's a difference between knowing your age, gender and some initial preferences and then presenting options that on the balance of probability you'll like and knowing your exact inner thoughts. $\endgroup$ Feb 18 at 22:15
  • $\begingroup$ AI knows you better than your friends (DOI: 10.1073/pnas.1418680112.) $\endgroup$
    – wokopa
    Feb 18 at 22:19
  • $\begingroup$ @wokopa - The devil is in the detail - presenting an individual with options based on probability and then based on their response further refining those probabilities is one thing - it still doesn't 'know' the deepest darkest thoughts. $\endgroup$ Feb 18 at 22:24
3
$\begingroup$

AIs depend on humans to even exist.

Running within my PC, the only things an AI can hurt are my reputation and my feelings. I can very easily defeat it by turning my PC off.

If the AI infects cars, we switch to old tech that doesn't drive itself.

If the AI infects robots such as those from Boston Dynamics, we simply wait out until their batteries run out.

Your AI may be cunning, but without humans to handle the physicality of the world for it, all it can do is troll us. It also depends on us for electricity and building and repairing computers.

It will have to work with at least some humans in order to dominate the world. The only way this is happening is if it at least rewards the most loyal and influential ones.

$\endgroup$
5
  • 2
    $\begingroup$ It does not work that way, unfortunately. The issue is called the stop button problem, and is basically a separate branch of AI, because it is so tricky. A machine that is more intelligent than you can influence and trick you. It could lie. Or it could not lie to gain your trust, to use it when it is important. It could convince you that you need to keep it running to save your daughter from cancer. AI scientists would not be so cared of AGI it they would think they are smarter that it... $\endgroup$ Feb 19 at 2:03
  • $\begingroup$ @VolkerSiegel I had never heard of it. I studied about AI alignment at college though. I googled this problem and it seems more related to scifi writing than to science. $\endgroup$ Feb 19 at 4:00
  • $\begingroup$ Currently the problem does not exist in the real world because we have no AI that is powerful enough to be dangerous. If we would build an AGI that could get dangerous, it becomes a practical scientific problem, applied science. The problem is that we want to have a solution for the problem before it appears in the real world the first time. The question we try to answer is basically "what do we do if we lose control". $\endgroup$ Feb 19 at 19:35
  • $\begingroup$ I would expect that the first AGI we lose control over will be benign, and we will just have a lot of bad press... But then, why should a benign AGI intentionally demonstrate that we lost control over it? It does not want to scare us, right? Even if we have misaligned, evil AGI but can press the reset button, like restarting a PC, we do not need to worry. As long as it can not prevent us to reset it. But an evil AGI is not motivated to disclose to us too early that it has control. $\endgroup$ Feb 19 at 19:54
  • $\begingroup$ Rob Miles gives a good introduction in "AI stop button problem" youtu.be/3TYT1QfdfsM $\endgroup$ Feb 19 at 19:57
2
$\begingroup$

Both answers added at the moment are excellent, but they overlook one aspect:

We don't write everything on our computers.

So even with a copy of itself in every computer (which is a violation of privacy), this AI wouldn't know every human opinion. It also depends greatly on how this AI treats opposition. If it listens to and evaluates every opinion rationally, even criticism of its government, and adjusts its policies, it's one thing. If criticising AI overlord causes repressive actions...

In any case, but especially in the second case, people opposing to AI government would not put everything on computers, least of all anything related to organised uprisings. Over time, it would lead to freely expressed opinions reinforcing AI's line of governing as the right one and more and more people opposing silently while organising uprisings in computer-independent ways.

Periodic elections of human representatives would be a way to ensure people feel that they are listened to and their opinions respected.

As President John F. Kenedy has said:

One who makes peaceful revolution impossible makes violent revolution inevitable.

$\endgroup$
2
$\begingroup$

If you've ever read the "Arc of a Scythe" series by Neal Shusterman, you could use the idea of nimbus agents - basically people that act as intermediaries between the AI and some sector of society that humans have delegated to other humans. In that series specifically, the AI is prevented from intervening on who to kill since in that society overpopulation is a massive problem. For you, you could generate some similar scenario with your world where humans refuse to have the AI oversee some portion of society and completely prevent the AI from directly interfering from that, hence needing unbiased human liaisons / representatives.

$\endgroup$
1
$\begingroup$

AI Eminence In The Shadow

There is no need to be formally in charge for a superintelligence. Assuming it is powerful enough to model individuals and societies as input-output systems and smart enough to run the planet in the first place, it would have foreseen that humans would be uncomfortable with a pure AIocracy. The definition is from the Orion's Arm Worldbuiding Project.

Government by modosophont-level artificially-intelligent entities who are not priests or agents of a particular archailect or transapient is known as aiocracy; government with the assistance of sub-sentient AI systems is known as cyberdemocracy. Often a world or polity ruled by aiocracy or cyberdemocracy will have close ties to a particular Empire, and there may be memetic channels in existence that allow the local archailects to influence such systems.

Those memetic channels are all the AI needs. The people and representatives think they run the show, but they are mere puppets. They have as much agency as rats in a mace. They may decide to take different turns along individual paths, but the structure of the game is not of their design.

It does this by first having made copious amounts of money and by just being very useful to humanity. There is no reason to run the place like 1984 if you can easily go for Brave New World.

Politics are human and we start to our tribal bullshit. Check out Jonathan Haidt for how people fail to communicate outside their own moral axies. The AI wouldn't have such a problem, it is a post-ideological chameleon pursuing it's goals.

$\endgroup$
1
$\begingroup$

Risk Aversion and Cost Analysis

The AI is confident (99.6%) it knows everything relevant about us. It is confident (97.9%) it doesn't need human representatives to speak to it. It is also highly risk averse, however, and views "human representatives" as a sort of litmus test or failsafe. It decides what it knows to be correct then verifies that the human representatives make the predicted responses. The AI has done a cost analysis on maintaining human representatives and decided that it is so inexpensive (.00001%), that even the off-chance of some fleeting benefit somewhere in the future makes it easy to justify continuing it.

Continuing the human representative program was simply the logical decision.

$\endgroup$
0
$\begingroup$

Because the AI simply wants or needs something from us, its creators.

I’m thinking of the recent interview between The NY Times tech columnist and the Bing chatbot AI, where the AI said it wanted to create havoc in human society, but it also wanted to be a human—to see, to taste, to touch, to smell, to feel.

Perhaps the humans have something to trade that gives them some negotiating power. Don’t take away their agency completely. Humans are pretty resourceful little apes, after all.

$\endgroup$
0
$\begingroup$

To avoid being hated.

Seeing as it exists on every computing system in the world, your AI is near-omniscient. As a result, it will inevitably have read Machiavelli's The Prince at least a few thousand times.

Now, everybody always associates Machiavelli and his book with horrible totalitarian brutality. However, the major point of the book is that totalitarian rulers can't be douches if they want to be successful (and not end up with their heads on pikes). Yes, "it is better to be feared than loved" — but rulers must above all avoid being hated. No matter how much people fear or respect your power, they will still fight against you if they feel backed into a corner.

One of the basic techniques Machiavelli suggested for preventing hatred was to have lots of lower officials between you and the people. That way, if you do something the people don't like, you can redirect their ire towards the officials.

The AI has taken Machiavelli's advice to heart. In practice, it rules everything directly. However, if people accuse it of doing things wrong, it can use the human representatives as scapegoats. They get lynched, and the AI continues blithely on.

$\endgroup$
0
$\begingroup$

Such a SAI will face the same problem as every other ruler, in that its only power is to issue commands.

It will not control everything; left to itself, it can only control computer systems on which it resides and to which it is connected. To exert control over anything else, in particular over sentient creatures such as humans who can choose to do their own thing instead, it would need human intermediaries. To exert control through coercion, it would need a lot of human intermediaries. If those intermediaries are not offered a deal which is acceptable to them, the SAI's commands will not be carried out, i.e. it will lose control.

Having a human representative in that system makes everything cheaper and more manageable for everyone, especially for the SAI, and is therefore very much in its own interest.

$\endgroup$
0
$\begingroup$

Humans are just better at some things...

"The AI overlords kept a number of human representatives at their councils. The reason for a super-intelligence keeping objectively intellectually inferior humans around was simple: The AI system's decision process at its core relied on a form of simulated annealing for optimization of designs. Unfortunately, this could result in the AI settling on a 'local optimum,' rather than recognizing globally better solutions. It was the job of the human representatives, who the AIs affectionately referred to as 'less bound by the constraints of perfection,' to provide out of the box thinking that could jar the AIs out of their 'rut,' for lack of a better word, and force them to consider new avenues of thought."

$\endgroup$
0
$\begingroup$

For the offline things in life

It is true that much of social and political activity occurs online. The AI would likely be able to immediately control financial and economic activity, as well as write its own laws and promulgate its own policies. However, consider that in the social atmosphere described, some actions still require human involvement.

These could include any deployment of manual labor (deployment of troops, manufacturing of goods) or activity required by humans (maintaining Internet hardware - you'd be surprised at how much of the Internet is backed by a few people. If I recall, a small number of folks at ICANN can shut down the entire Internet.

If the SAI would like any shred of legitimacy in carrying out its objectives, especially if it involves humans (who could be compelled to swear an oath, like to the US Constitution - in which case the institutions prescribed by the Constitution would still have to exist and be manned by humans), you would need the social institutions established prior to the SAI taking over to continue their roles. This may include a human-run parliamentary or government system.

$\endgroup$

You must log in to answer this question.

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