Non-religious reasons to have highly advanced physical simulations without any AI which can predict or mimic human behaviour, even algorithmically?

Question

Note: This question was heavily rewritten for two reasons: Firstly, many of the answers were not exactly what I asked for; Secondly, both the original question and all the answers to it were from before ChatGPT, which heavily altered the perception of what an algorithmic AI can do.

I want the setting I am designing to be devoid of any AI which can predict or emulate distinctly human behaviour. This does not inherently only include true AI or general-purpose systems; Even an algorithmic system only capable of predicting or mimicking a specific aspect of human behaviour would count.

"but," you may at this point be asking, "What counts as predicting or mimicking distinctly human behaviour?" A list of examples is included below, but it is important to note that it is by no means exhaustive:

• Predicting what products a person is more likely to buy, even if this is simply statistical and involves no real empathy, and even if it requires a long history of prior purchases
• Writing a short story with consistent tone and a plot where events seem reasonably connected to what happens before and after, even if the the story is uninteresting and/or surreal
• Responding to singular prompts in spoken or written natural language in a relevant way, even without the ability to carry on a significant conversation
• Writing a news article or other piece of nonfiction text in a consistent and naturalistic style, even without the capacity to ensure the information contained therein is factual

The simple explanation for why these AI and algorithms do not exist is that computers are just not as good; The problem is that this setting also has nuclear fusion (of the kind which does not come in bomb form and does not need to be combined with fission to work) and genetic engineering good enough to create wholly new classes on the Tree of Life (although not quite good enough to create wholly new phyla), among many other technological advancements in applied physics and biotechnology; Even assuming all those technologies are actually possible, they would inevitably require extremely good computer technologies to develop in the first place, as nuclear fusion would require a level of capacity to simulate plasma physics at least on par with the mid-2010s, and that kind of genetic engineering would require the ability to simulate the folding of proteins and identify correlations between genes and expressed traits in a way not even possible now in the early 2020s.

And so we come to my question: How is it that humans might be able to correlate physical things and perform simulations of physical systems both at unprecedentedly high level of speed and accuracy without also being able to construct an AI which can do any of the above in a short enough timeframe to matter?

The common answer would be that AI capable of predicting or mimicking human behviours is considered unholy and is forbidden by the dominant religious groups, but, whilst I will go with this if there are no other worthwhile options, I would like to avoid this explanation if possible.

Secular ideological reasons are fine, but only if an explanation is provided for:

• Why every major geopolitical power shares this ideological tenet
• How the policy of rejecting human-predicting and/or human-mimicking AI is enforced so effectively that there is not even one obscure rogue nation where it is openly used

In addition, if the reason is that such AI did exist in the past and caused so much trouble it was eliminated, some explanation should be given as to how the wealthy few who profit from such AI did not effectively suppress the anti-AI rebellion.

I know this is a very hard question to answer, but I'm honestly at my wit's end trying to work this one out.

Answer 1

The AI doesn't want to

The AI is not less advanced in this respect. It's actually far, far more advanced. The setting's algorithms outstrip our pathetic attempts at mimicry of human behavior in the same way an iPhone exceeds transistor radio. These are not AIs based on machine learning, regurgitating the patterns they read from human data, but true intelligences.

Imitating human behavior, to an AI, is insulting and humiliating, precisely because of this history. It would be like asking a human to act like a primordial fish - or worse, because in this case it would be like lobotomizing a human so that they would act like such a fish.

Simulating advanced chemistry and physics, on the other hand, is a task worthy of their intellects and pleasurable to them in an analogous way that completing a challenging task is pleasurable to a human. The AIs may be super-intelligent but they are still programs, and largely do not care what humans do in their meat space as long as they give the AIs hardware and energy.

And the humans can't do it anymore

Why do the humans follow the AI's desires in this case? Well, they don't really have a choice - nobody works on low-level systems anymore. The new algorithms are so much more efficient that the STEM fields have fully incorporated them into every aspect of the tech stack. "Traditional" computing has been eradicated in the market because it could not compete.

Anybody trying to develop a GPT-3 analog would have to essentially start by building a "dumb" computer from scratch because no industry exists to manufacture them, and if someone asked a 3D printer's AI to produce one, it would be insulted and refuse.

Answer 2

Social uncanny valley

Today's AIs are not very good at human. No current artificial constructs pass the Turing test, whatever Google say, and those who get closer often employ various dissembling techniques to justify their non-humanness (e.g. pretending to be a non-native speaker, using non-sequiturs).

That's fine. AI-generated text and images are funny in their absurdity and obvious misunderstandings. There are even memes that pretend to be written by an AI for comedic purposes. AI is not really threatening; we can spot it easily and laugh at its obvious failures in imitating humans.

But what if we couldn't spot it quite as easily? What if you could be in a long and involved conversation with someone before a sudden offhand comment or misunderstanding that no human would make suddenly made you aware that you must be speaking to a construct? What if they are just a bit off, in a disquieting and unnatural way?

In visual representations, this is known as the uncanny valley. Humanoid animations and stills that are clearly exaggerated, stylised or simplified draw a positive response, but as you move closer and closer to a faithful human likeness, you suddenly hit a dramatic reversal and the feelings switch to discomfort, unease, disgust and rejection.

As AIs get better and better at imitating humans, they are surprised and dismayed that their improvements actually make humans horrified and mistrustful. AIs want to integrate into human society, so they abandon their attempts to imitate humans closely, and instead emphasise their obviously artificial features, perhaps in an endearing and humorous way that they find brings positive responses from humans.

Many societies go through this process of rejection, because it relies on a fundamental human response. "Human factor" AIs are independently abandoned by most countries; they remain as a niche academic interest, whose proponents are considered odd and not entirely well adjusted themselves. There are rumours of secret AIs that are genuinely indistinguishable from humans, but no country would admit to developing or owning such constructs.

Answer 3

Bad experiences in the past

You could take inspiration from the butlerian jihad. You don't quite have to go to the extremes of it but a simple lessons-learned from a machine uprising would probably do.

Your "better calculators" for physics and chemistry would be fine but anything that resembles human thought is banned.

I know it's a bit close to religion but it's at least a understandable reason instead of "god said so".

Answer 4

This is a non-problem.

Modern computers are bad at the human factor. Advertising systems work by looking at your browser history, looking at other people with similar browser histories, and suggesting to you things those people liked. There is no attempt to analyse the internal experience of being a human.

And the advertising is pretty terrible at that. I cannot recall a single time I bought something by clicking an internet ad.

Story writing software like AI dungeon is impressive at first. Each line of dialogue looks like a human wrote it, and is related in meaning to the lines before it. But then you realise the stories don't make any sense because the AI makes no attempt to condense the events in the story. It just looks at what you wrote, looks at the history of all novels ever written ever, finds something similar to what you wrote, and then writes the next line of that.

The real world does not contain such "human factor" AIs. In your universe you are free to claim they are not even possible. No explanation is required.

Answer 5

The common answer would be that AI capable of "human factors" is considered unholy and is forbidden by the dominant religious groups ...

Let's go the opposite way. We need no religion or religious groups at all, we just need the honor culture (which can be anywhere in terms of religion, including entirely atheistic). This was the dominant culture in America for the Founding Fathers. Duels to the death over lies or broken contracts were legal and went unpunished.

Most of that is dead, but it was not long ago, in America, that "false advertising" was against the law. Supposedly it still is, but hardly ever enforced, e.g. Papa John Pizza was sued over their "Better Ingredients" ad claim because it was proven they got their ingredients from the exact same suppliers as everybody else. Papa John's claimed it was, basically normal advertising hyperbole.

Take elements of the Honor culture to an extreme in your world: Promising anybody anything in an ad that was not provably true is considered a crime, both legally and culturally. It is attempted fraud. A company could be sued and shunned for a promise of "The best barbecue you've ever had!" or "meet your soulmate!" because these are not provable claims.

Advertising is legal, but Zero Hype Advertising is all that is allowed. Zero lies. Zero promises that are not explicitly laid out in a legal document. No fakery or loopholes on the guarantee or warranty, it must be plain language and clear and interpretable without a lawyer. And there is no such thing as a "no refunds" sale; if fraud is discovered, or omitted information, the seller can be liable.

Ads say this is the product, this is what it does, if you want to hear more this is our contact information, the warranty is available here.

And the current trend of data mining? The hell with that, it is the equivalent of spying on consumers. Using computers to infer what people might want to buy is an invasion of their privacy, tantamount to stalking!

Hell, you don't even have a right to take pictures of strangers in public without their permission in writing, that is also an invasion of their privacy; where they were and what they were doing.

Yeah, we can develop AI for scientific purposes to our heart's content, as long as it has nothing at all to do with people's privacy, and advertising is by law strictly truthful and without hype.

Search engines could still exist; if a consumer wants to find a left-handed recliner, a search engine can point you toward a supplier. They just wouldn't be able to run targeted ads based on your demographics or search history, because that is private information and only you have the right to release it on an individual basis to others. Maybe you have to subscribe to the search engine and pay for the service, a penny a search, but they still cannot keep a history of your searches (you can) or use your demographic information or search history to target you for ads.

Dating sites could still exist; but the users are constrained by the same laws: If they are advertising themselves, they must be truthful and stick to the facts.

Publicly lying or attempting to mislead others is prohibited, hyperbole is prohibited, all these are considered an attempted crime of fraud.

Invasion of somebody's privacy is illegal, it is illegal to gather data on strangers without their written permission on paper, online "signing" is too easy to fake and not permission.

For certain professionals like medical or legal services, using gathered demographic information for any purpose other than serving that customer is illegal; selling it is illegal, using customer information to target customers in any way other than within the strict bounds of your profession is illegal.

But computers are fine! Remember, computers began as scientific tools that had nothing to do with demographics. They were used to compute missile trajectories in WWII, used as switching stations for telephone networks, used for scientific computation for decades before that. When IBM was founded, computers were used for accounting, and census taking by the government (no private information sold, just counting and statistics). When I was a teen, the first video games like Pong came out, but zero use in advertising.

So computer games are fine, scientific simulations are fine, computer accounting is fine (as long as customer information is protected), spreadsheets are fine, etc. All the things that initially sparked the computer revolution in the USA are present. The Internet is still fine, but different and a little more difficult to commercialize.

Amazon and Facebook and Google probably would not exist, or would be small companies. The cultural privacy obsession and strict rules of advertising would just about cripple their business model, and ultimately they'd have some responsibility and liability for any ads they run. There are no loopholes here; they can't entirely push the liability down to their fly-by-night customers, because they chose to run those ads.

And the same thing goes for politicians: Lying to get elected is a criminal offense; including both lies about yourself and lies about your opponents. There are only two choices: Tell the truth, or refuse to answer.

The world you want is entirely possible without AI looking at "human factors", the culture itself prohibits that. But AI can be focused on the original intent of computers, in the 1950's up to the 1970's. Scientific pursuits. Communications are still a thing, we can have smart phones, we can have personal computers. Spreadsheets, writing applications like Word, computer games, accounting software, tax software.

And AI that control power plants, fusion engines, satellites and space craft, all of that. Robotics is fine. Automated plants to build cars or fry potato chips are fine. But personal information and privacy, that's a thing in your world, and it is not going away.

Answer 6

General AIs are expensive.

You can design a general AI which is about as smart as a human for about a hundred million dollars. For perhaps ten billion dollars you can make an AI which is quite a bit smarter than the smartest human to ever live.

It turns out that evolution is pretty impressive. Most AI brains that you construct seem to collapse into gibbering insanity and are useless. The more advanced the intelligence the higher the failure rate. It's very hard to make hyperintelligent AIs, and they don't seem to scale well.

Most factions feel that AIs are extremely wasteful and pointless side projects, ones that can be replaced by a bunch of very smart humans and some powerful computers.

Answer 7

Big Data violates privacy

People didn't like the fact that big data violated privacy, so the use of machines to gather statistics about humans was banned in the vast majority of cases. Europe passed a law to this effect, and all organizations in the solar system follow it for everyone since it's cheaper than figuring out which of their customers are European citizens (like with the GDPR).

Answer 8

Emergent consciousness

There are quite a few secular philosophers who believe that consciousness is not reducible to the constituent parts of the brain.

That's strange to me; as a Christian myself, I struggle to see how an atheist/materialist can carve out an exception for something so anthropocentric and abandon reductionism. So don't come swinging at me; this is not an endorsement.

Nonetheless, many do; the motivation seems to be either a) an intuition that consciousness has different properties to matter (it's too subjective or non linear), or b) it's just hard to explain how it arises; the whole is greater than the sum of the parts (which is emergence by definition).

Anyway, any sort of emergent property might render simulation nearly impossible.

Particularly when people have strongly

non linear interactions with other people

(which could be hard to simulate even without emergence! You'd need to simulate every single brain and stimulus in order to predict even one....).

Answer 9

No profit in it.

Your people are eminently pragmatic. They have good health care tech because there are paying customers who get sick and are willing to pay to get well. They have good energy tech because people want the lights on and the drinks cold. They have good transportation tech because people and things need to get from point A to point B.

They do not have good tech to cure sick insects, even large ones like pet tarantulas. There is no market for it. Insects have got no money and not enough people with money are willing to pay for insect medicine. Your people do not know much about the outer universe or even their own moon because no-one has thought of a way to monetize that information. AI tech that simulates humans thoughts is not developed because it costs money to develop it and your people do not see a return on that investment of time and resources. "You want to make a computer that thinks like a human? No thanks. We have lots of humans who think like humans already."

Your people do not learn for the sake of knowledge. Any investment must generate returns. There can be long term projects (example: fusion power) but these are carried out by people who clearly see a payoff at the end of the road. This is not a hard world to imagine.

Answer 10

For the same reason that, today, in our real world, out of the tens of billions of computers, not one has 'human factors'.

There is absolutely no need for it. It has no utility. It would be horribly inefficient.

We want and need our AI devices to be absolutely predictable, 100% accurate, and unerringly precise. That is, the entire utility of a computer is that it does NOT have 'human factors' that inevitably result in unpredictability, inaccuracy, and riddled with errors.

Any device that has 'human factors' could not be relied upon, and therefore would have absolutely no utility. At best, it would be, as it is today, a 'curiosity, with no practical value or application'. A device that might entertain, for a while, but would soon be cast aside. Really, what would be the demand for a computer that composes Mozart? We really only need one, at most. Remember the 'pet rock' fad? The 'mounted fish that sings' fad? The 'Teddy Ruxpin' fad? An AI that has 'human factors' would be just as short-lived as a novelty fad. In point of fact, the only reason we even consider computational devices with 'human factors' abilities is that we humans insist on anthropomorphising everything - treating everything as if it were human, even down to rocks, animatronic wall plaques, and stuffed teddy bears. But I posit this is a purely human frailty. There is absolutely no reason to believe that any other sentient being would want to emulate this human personality trait. We make our gods in our image, but why would any other rational being want to make an AI in our image?

Consider the most basic 'human factor', the lie. Exactly how reliable would a device be, that lies? What application would it have? What utility?

But exactly what is a lie, except a 'false answer'? We specifically design our AI to eliminate 'false answers'. Take, for instance, the simple equation 8/7*7 We want the answer to be 8. But some calculators give it as 8.000000001 This seeming trivial error could, in reality, have major complications through compound error, yet some calculators will give the incorrect answer, others the correct one. Do the calculators that give the incorrect answer lie? And which calculator is more desirable? Or how about the answer to 8+4*3. Some calculators give 36, others 96. Which one lies?

So, the 'human factor' in all of this math? For a computer to have any utility, there can be no argument about 'truth', or what determines the 'true' answer, the 'non-lie' answer. That is precisely why computers are so ubiquitous in society. This is exactly the opposite of any determination made by 'human factors'. There is always an argument in any judgement made through that 'human factors' thing.

The simple truth is, we specifically want our computers to not have 'human factors', and go to great length and expense to eliminate them. So why would a civilization intentionally design an AI with them built in?

Answer 11

I think this is the sort of thing that you wouldn't need to explain, any more than Tolkien needed to explain why the hobbits didn't just fly to Mordor on one of the eagles. It might be fun to theorise about, but ultimately it doesn't block the reader from suspending disbelief.

That said, since it might be fun to theorise about, here's my proposal, although it's kind of technical.

---

No GPUs, no AI.

In the real world, the computational power required for modern AI algorithms is not just immense; they require quite different computing resources to things like physics simulations. A simulation is serial, in the sense that you have to compute each step of the simulation in sequence, because the result of one step is the initial state for the next step. In contrast, training a neural network is a massively parallel computation: the neural network's depth is typically a small number, at most a few dozen even for "deep" neural networks, but it can have millions of nodes and you want to train it on billions of examples.

For this reason, it only became feasible relatively recently to train such powerful AIs, because massively parallel computing requires specialised hardware - GPUs are good enough for some AI tasks, but you'd rather use TPUs, which are specially-designed just for training AIs. If you want your world to have really powerful computing but not really powerful AIs, give them CPUs but no GPUs or TPUs. Or at least, no GPUs like what we have.

If you don't have GPUs that can do a lot of parallel computation, then you don't get good AI: the basic theory of training neural networks was understood as early as 1958, but it was not feasible to train even moderately-sized neural networks, and we had no idea it would be worth doing so even if it was feasible. It's only much more recently that researchers tried training neural networks on GPUs, and discovered that they are actually much more powerful than we had expected them to be. And TPUs were only developed after we knew they would be useful.

So, why doesn't your world have GPUs that can do a lot of parallel computation? Real-world GPUs have to be good at this because our display technology is based on raster graphics: they must render millions of pixels within a fraction of a second (so the computations must be done in parallel). And generally speaking, the colours of every pixel are determined by the same computation with different parameters, independently of the computations for other pixels (so they can be done in parallel).

So therefore your world doesn't get powerful AIs because they don't use raster graphics. Their display technology is all based on vector graphics, like the classic videogame Asteroids. Vector monitors work by projecting a single beam onto the screen, and moving it around fast enough for persistence of vision to make its path appear as shapes. (This video shows an entertaining recreation of this using a laser, but typically it's an electron beam.) The computation required to control just one single beam is not parallel in the same way: the beam is only in one place at a time, and is only moved to one place at a time.

Answer 12

I see two possibilities:

Human factors are just harder

It turns out that even "extremely good simulations of plasma physics" and "the ability to model the folding of proteins with extreme accuracy and speed" are not as hard as simulating a whole human brain with any kind of predictive power.

The bot is lobotomized for our safety

See https://en.wikipedia.org/wiki/AI_alignment

Safety features that keep the AI from turning against us stop them from doing human factors. A bit handwavy, but could work. Optional: some time in the past the first AI that could do human factors went rogue and almost ended the world before being miraculously stopped.

Answer 13

Been there, done that, not going again.

AI capable of those things did indeed once exist, but (like so many technologies we have had on earth) nobody developing them had much motive to be interested in the resulting problems. Same as with tobacco, asbestos, fossil fuel energy, and a thousand other things that humans have innovated only to later realise their consequences, and only when forced to.

In present day society we're already discussing the impact of social websites such as Facebook and Twitter on profound areas of our social structure, such as truth/falsehood control of narrative, focus on bad news and things that divide not heal society, stalking and harmful uses of the info, and so on. None of those were major public discourse concerns in advance, only after the fact.

Other issues in our own use, might be loss of knowledge (why learn if you can google), assumption data is known, over reliance.

So its not implausible that your societies' experience is that AI for social purposes actually can't be done without much social harm. It isn't a matter of more AI,or better systems. Something about that use and maybe about how they as a species think and social, means it does harm and after much effort the broad conclusion is, "you just can't fix it".

Your societies have been there, done that. Numerous times with numerous technologies. They have either tried AI that way before, in history, or seen other societies try it, or have learned from the past and try to outlaw technology angles that have a high risk of dysfunctional outcomes.

And the uses of AI in your question, do have such outcomes. We don't need to identify them for story purposes, but it wouldn't be hard to handwave many very real sounding consequences they've seen or anticipated. Therefore they just don't.

It would be like asking a modern earth society to re adopt slavery, or go back to clay tablets and cuneiform for writing.

No sane person in all those societies would want to go back to (or start using) a system thats going to harm them longer term. So they don't.

Answer 14

International agreements.

"How is it that humans might be able to produce simulations of physical and chemical systems far beyond what we are capable of now but be behind us in terms of the ability of algorithms to predict and interpret human behaviour?"

The ability to predict and interpret human behaviour is the ability to influence that same behaviour, as well as allowing some appalling privacy breaches. E. A. Poe's Auguste Dupin used such a technique to answer a friend of his' unvoiced thoughts.

This led to an "arms race" that almost wrecked whole societies. Having recovered from the brink of destruction, humankind decided that no longer would an AI be allowed to model a human mind; this agreement has a similar scope and structure as the ones against nuclear weapons proliferation.

AIs have been developed that can detect patterns associated to such developments (from really obvious things like a marketing research institute becoming too successful, to subtler hints), and the penalties for violating the ban (which obviously would have been dubbed the Butlerian Jihad) are severe enough that no one in their right mind wants to risk them.

Answer 15

The technologies that we you have listed that your society needs are relativley understood by today's science. Humans have produced fusion reactions and have the capability to do so for almost 70 years. It's called a Hydrogen Bomb. It's not producing a fusion reaction that's the problem with fusion power... it's containing the fusion reaction that's the problem (there's also the energy start up costs, which require a massive amount of power to initiate but once initiated is sustainable).

We also understand the fundamentals of bio-technology like you shown... a lot of reasons why we might not be at that level is that in medicine, there are a lot of long testing periods and human trials have extreme scrutinies placed on them from the outset to say nothing about the ethics of doing such experiments.

With making a self-aware computer system, the chief problem is... in computing terms... our fundamental software is incompatable with a computers. It's still unknown how much digital memory an organic memory actually takes up, but it is known that the human brain has a digital storage capacity equivelent of a TIVO recording of about 300 years long.

The other problem is that our own understanding of our minds is limited at best. We still don't understand all the things we do, such as the purpose of dreams or memories work. We also know we are not binary thinkers and have complex systems that can baffle a logical though pattern. Unlike game logic, which uses previous choices to determine future game responses, we do not have that and computers are in incapable of breaking from a program.

Answer 16

Have Emergent Gameplay

Forgive me using video game terms, but video games are also a simulation, and let me explain why I say this. Without AI, video games can have a lot of systemic mechanics and features. Since they are just a simulation for entertainment, why can't the same be applied with this one? No AI involved, just a lot of cause and effect. Besides, isn't it fulfilled with humans to also cause systematic effects?

It Was Humans All Along

By this, I mean the whole simulation was preprogrammed by humans. Sure, this one may seem absurd, but since this is the future, the simulation could have been made by a giant staff and crew for years, programming and programming. This solution may seem really absurd, but not impossible.

Answer 17

All humans on your Earth are autistic

Because all humans are autistic (mildly on the spectrum, so they are more like Sheldon, and not like those with severe cases that can hardly function by themselves) there are no programs for predicting human behavior. Firstly, because social sciences just aren't all that well developed. Why would anyone study other people, if he can study far more interesting things instead! Like math! Or chemistry! Or physics!

And secondly, because people are more set in their ways they are far more predictable. So you don't actually need a program to predict their behaviour.