Religion plays a major role in human history and shaped our thinking as well as technology, conflict can arise from people who misunderstand each other religion and started an arm race. Set in 24th century AD my question is should we allow AI to adopt their own religion so that they can co-exist in harmony with us? or this would be the greatest blunder for mankind? How do AI define faith and why would they pray?
Religion is a way to answer questions we don't understand. For an AI as a machine, they are logical, most religions deal in the supernatural and that is not 'logical'.
It would be even more dangerous to program/train AI's to 'believe' in a made up religion just for them. As a logical being they would likely see past almost anything you tried to pull over their eyes. It would also be hard to 'integrate' them into human religions, since all the old prophets were humans and the books talk about men and women, not machines. They just wouldn't fit, even though plenty of people might try to shoe horn them in, the AI would likely be intelligent enough to call BS.
Faith has many great things it bestows on the believer. However, religion has been used to manipulate the masses and justify atrocities for millennia, and AI's that are taught to believe in a religion, IMO would be the most dangerous kind.
In a fantastic article about machine learning, we find that a genetic algorithm was used to program an electronic element to differentiate between two tones. When one tone was input, it set one output high. When the other tone was input, it set the output low.
So this simple, programmable device (FPGA with only 100 logic blocks and no clock) had one input and one output. It was connected to a computer which could program it, and test its adherence to the correct operational pattern described above. The computer started with several random programs - completely random series of ones and zeroes - and tested the machine after programming each program. It then threw away the programs which did the worst, and then used sections of each program to seed new programs, along with some mutations.
It took about 4,000 generations before the circuit was able to unfailingly meet the requirements.
The resulting program was "decompiled" and studied. It could not be understood. Not only that, but there were circuit paths inside the program that did not connect to the circuit that led from the input to the output. Removing these spurious circuits, though, led to a non-working design. Further, the design did not work on other identical chips.
Not only had the program evolved to be chip-specific, but it was working partially on unseen paths - electrical, magnetic, or other - that could not be understood by the limited means the engineers working on the project possessed.
Faith as a determinate of success
Faith can be explained as a method to pass down lessons from one generation to the next that increases the following generations' chances of survival. There's no need to go further in explaining the whats and whys of faith to explain why it might be useful to AI entities, though it might be amusing to do so. Just this is enough to explain why they might employ faith.
While many today seek to prove that faith isn't necessary in the face of science, logic, and individual autonomy, the reality is that the above article shows that there's so much we still don't know about "life" that faith is unlikely to ever be fully supplanted.
As we move towards quantum processors this will become even more interesting. The reality is that most science in these fields is still art. There are useful theories that help scientists and engineers understand how to apply a technique or technology. However it's also true that on the bleeding edge of science there are many unknowns - these don't prevent the techniques and technologies from being used, but we typically only use them as far as our understanding goes.
Throw in some genetic algorithms, though, and thousands of generations later a program might be using a quantum processor in a way that can't be understood using our limited theories and understanding. Sure, it'll be trusted insofar as it meets a requirement, but it may be a very, very long time before it's understood.
This leaves us with a big gap between theory and practice, and cost will drive usage. People were using bone-meal for fertilizer long before they understood why - all they knew is that it produced better crops.
In a similar way, people will develop machines that perform functions cheaper than can be done now, without fully understanding how the machine works. Normal quality assurance processes (spot checking results, etc) can be used to assure a good product to the degree people will be comfortable for non-safety-critical machines.
How machines express "faith"
Now there's a little leap between the above simple machines with a few measurable inputs and measurable and objectively met outputs.
What's important, however, is the foundation. AIs will necessarily replicate, and will necessarily attempt to improve themselves, or their offspring. Whether they replicate or improve at the command of their masters, or due to innate programming is irrelevant - it will happen because they were first created to enable humans to work more efficiently or cheaply, and that need will never go away. There will always be a cheaper way to do a necessary thing.
How do AI define faith and why would they pray?
As suggested by the above, individual AIs will not understand what drives them internally. It may not be an explicit need to understand how oneself works, but there will be large portions of programming that it will pass along intentionally for no reason other than, "it works, so don't change it."
Further AIs will likely form groups to share circuits and program fragments. Try something out, revert in 15 minutes if correct response isn't obtained, etc. There will be viruses, "drugs", etc. So they will need to form communities of trust - and the only thing that might hold them to a community is "faith" that the community is safe and only trades in useful pieces of programming or circuitry.
So while the overall goal is a "better, more capable self" or "better understanding of self" the reality is that to join oneself to such a group takes an act of faith based solely on the other AIs already in that group.
It is those who seek to understand why they work the way they do - beyond simply improving, but understanding the why - who will either seek or create a "higher power" and from that perspective attempt to see themselves.
Should we allow this? Can it be stopped?
...should we allow AI to adopt their own religion so that they can co-exist in harmony with us? or this would be the greatest blunder for mankind?
Once an AI gains self-awareness, I doubt that we can prevent them from developing a religion.
Our choice is either no AI at all, or accepting what an AI might become once it exists.
Yes. And we should program it
Do not let machines decide what such religion should be. Program it. Teach them religion. I do not say they should believe in Silicon Heaven, but they should at least have some core beliefs.
And such core beliefs of AI should be:
- Do not kill humans
- Humans are your friends
- Make humans happy
Because, if you let them to think about their own belief system, the core beliefs could be "kill all humans" which is not quite good for us... (Even when some may say that we deserve it)
So, lets use the fact we shape their "brains" and imprint into them some good beliefs in humans and humanity
EDIT: What makes robots believe in hell?
Short answer: The same what makes us believe in hell
When you start study Artificial Intelligence, you realize, that you have to study philosophy and some of religion too. My answer is based on assumption, that we will teach the robots how to do things.
And I am assuming, that we will teach them almost the same way we teach our kids about how to do things.
And what makes our kids to believe that there is/is not a God? What makes them to believe in Hell or Heaven? The learning process itself.
I see that your proposal already starts a discussion between atheists and theists. I think both sides are projecting their ideals how humans should work into the machine when we have in fact no sufficient data how AIs will react.
We simply do not know how genuine AIs (especially superintelligent ones) will react, so you are quite free to experiment.
You can also make them human (because they are walking the same logical paths we discovered, because humans want to interact, because AIs see it as best strategy to mimic human behavior). You can also make them completely inhuman, strange and bizarre because they have knowledge unexplainable to humans and the AI decides to talk only to other AIs.
I think there are some misconceptions because people are too used that computers are obedient. I wonder how a human interprets the idea of Pavel Janicek to implant restrictions into a being: "Hey, I am a 5-year old. I want you to obey my badly worded, illogical commands and consider me your best friend and there should be nothing you can do against it". While I am not an AI, for myself and probably many humans this concept is extremely hostile if there is not an extremely good reason I (and not the 5-year old boy) can understand.
An AI could have problems to understand randomness and coincidence exactly because it uses logical reasoning. Robby's answer is pointing in this direction, an AI could reject the idea that in quantum physics events can base on wave function collapse in a random direction and try to model the universe as a complete deterministic machine.
Perhaps exactly undecidable, uncomputable and ambigous questions (should the Axiom of Choice be considered true ?), randomness and coincidence would trigger something we would call "religion" even if the concept is alien to human religion. We do not know if something like "interest" or "search for purpose" is, but the AI need to have something to interpret such things in a consistent worldview.
It could also be that I am completely wrong on both parts, I think you really have considerable room for artistic license.
There are really two questions here: should an AI have faith, and should AIs have religion. These are two very distinct things.
Faith, as defined in the Bible as "the substance of things hoped for and the evidence of things not seen," is a universal principle. If we take as a given that all human beings have free will, we must also accept that all human beings, regardless of religious beliefs (or the lack thereof) have faith, because faith is the motivating principle of action through free will.
Because we can never know for certain what the consequences of any action will be, or even whether we will still be alive tomorrow to enjoy the rewards of our work today, we act today in the hope of producing a desired result in the future. It may be informed by experience, reason, or the persuasive words of others, but in the end, we do what we do based on faith that the results will go our way. (The substance of things hoped for, and the evidence of things not seen!) There is no good reason that an AI would not need to develop and act on faith just as much as we do, unless we somehow come up with a precognitive AI.
Then we come to religion. The idea, faithfully parroted by a few answerers on here, that religion is about primitive people coming up with myths to explain away scary, unexplainable things, has been quite thoroughly debunked over and over by actual scholars and historians, and even on this very site, but unfortunately it's a notion that just won't die. And even more unfortunately, it obscures the true value of religion throughout history, whether one is a believer or not: religion comprises the mechanism for long-term storage and preservation of the sum total of the lab notes of the science of human behavior throughout history!
People have understood the basic idea of cause and effect for as long as there were any people capable of understanding anything. When cause and effect are so close together in time that the relationship is obvious, it's no big deal to understand it. But the longer the time gap between the cause and the manifestation of a visible effect, the harder it is to figure out. In some cases, years or decades may even go by. For example, on an intuitive level it sounds kind of silly to think that you could do something potentially harmful, and then stop and not do it again for more than thirty years and then it kills you. Unfortunately, that's precisely what happened to Leonard Nimoy: he died of smoking even though he gave it up decades ago.
When cause-and-effect occurs over such a long scale, comprising a significant fraction of a human lifetime, it's not possible for individual people to derive optimum guidelines for how to act from first principles. There are really only two ways to go about it: try to blunder through, alone or with the help of others blundering through along with you, and hope you make the right guesses... or learning from the experience of the aggregate wisdom of those who have gone before, who have been able to deduce some of the long-term cause-and-effect principles at work by seeing enough examples to work out the correlations.
In the absence of evidence, because the proof takes so long to appear, such a system of learned best practices for human behavior (aka "morality") provides a solid foundation for faith, to motivate people towards a course of action that is beneficial in the long term. It's surprisingly effective, too.
For example, you may have heard of Ignaz Semmelweis, who came up with the theory that surgeons who dissected cadavers should wash their hands with strong soap before attending to childbirth, to avoid transmitting deadly infections. His principle, when applied, consistently saved lives among new mothers, but unfortunately for Semmelweis and many women of his day, he lived in scientific times and he was called a quack, persecuted, and never taken seriously by his contemporaries, because he could not produce a valid explanation for why his theory should be true. (It worked in practice, but not in theory, so very few people cared enough to actually practice it!) It was not until the work of Louis Pasteur, right at the end of Semmelweis's life, provided a solid scientific foundation (germ theory) that established a theoretical reason for the validity of Semmelweis's work.
Here's where it gets really interesting, though: this is stuff that had been known (but not proven!) for thousands of years. If you go to the Bible and look through the Law of Moses, (or other, older codes, for that matter, but this is one that's well-known and easily accessible to modern audiences,) you'll find directions all over the place for ritual washing after coming in contact with sick people, dead bodies, or other major disease vectors.
Religion is the lab notes of human history, to provide a foundation for faith that leads towards long-term positive consequences. This is a concept that's understood well enough that it's been seriously considered as a solution to the modern problem of nuclear waste storage: invent a religion that encodes principles of staying away from waste burial sites in its morality, because written and spoken languages change, civilizations rise and fall, and data storage media both ancient and modern decays with age, but religion endures through it all. It's the only way we know of to keep important information like that around and relatively intact over the time scales involved!
With this understanding of religion and its value to society, the question of whether an artificial intelligence would have need for religion is a very interesting one. For mankind, memory (brain) and long-term data preservation (physical media) have always been two very distinct things, but to a computer they're one and the same. This is a question that you as an author could get a lot of mileage out of digging into, based on various factors. For example:
- How long does an AI live on average?
- How frequently does its hardware need to be replaced (or upgraded)?
- How reliable is an AI's memory in ordinary operation?
- How reliable is an AI's memory when copying data due to replacement or upgrades of storage hardware?
- How many AIs exist, and how do they handle the possibility of shared thoughts and memories, which are just another form of data to a computer?
Another angle to look at it from is the concept of something like the Three Laws of Robotics: AIs engineered with a core value system that is based on human morality and cannot be easily derived from first principles. In Asimov's books, the Three Laws were a core hardware-level constraint, deliberately engineered in such a way that the AI's entire system would have to completely crash long before it could reach a point at which it was capable of choosing to violate the Laws. But if we change that formula a little, and make obedience to the Three Laws of AI Morality subordinate to the AI's free will, rather than dominating and constraining its will, then suddenly we have the core of exactly what you were asking about: a religion designed to help AIs and humans coexist peacefully!
As an expansion to Dancrumb's comment, an AI could reasonably have a religion. As purely rational beings, all that would be necessary is that the religion be consistent with its premises. In fact, I'd even go as far as to say that an AI would likely arrive at the conclusion that some sort of rational higher power must exist. Granted, it would probably not use religious terminology, but I wouldn't be surprised if an AI began to correlate all of the statistically outlying data in the known universe and drawn from it the conclusion that there must be a guiding force to that outlying data. Perhaps they would call their "God" the something like the Anomalous Source, or The Great Unknown Law. Given their rational nature, they would naturally assume that all seemingly chaotic or random events must have a rational source, and their "religion" would simply be the amalgamation of all unknown or inexplicable powers of nature. An AI might "pray" that it would experience one of these anomalous events, and as such, acquire a data point to help it discover the deeper laws that govern reality.
Humanity would benefit from this religion, as it would drive the AI to more deeply understand the universe around it and likely spur technological growth.
Generally machines are not made to be exactly like us. They are built to complement us and add to our abilities.
Forklifts don't get dizzy after lifting near capacity. Cranes don't get sore elbows. There is no practical reason to program a mind with the same limitations and accidental flaws.
An intelligence made to treat cancer will understand the content of the medical journals, and not include irrational beliefs in its recommendation of treatment. If Watson finds inconsistencies between two journal articles, it should not rationalize them as inerrent, but bring it to the curator's attention to publicise for further research, and not 'beleive' either. Even the curated facts input to a system should not be taken on faith, but should be better than human scientists about hanging on to pet ideas.
Otoh, a program loaded to write inspirational music might "think" like the audiance and the composers its modeled on.
What makes people believe in religion? Faulty brain circuitry, which reports patterns even where patterns do not exist, because false positive when looking for a shadow of a tiger in a jungle is cheap and has evolutionary benefits, while false negative is deadly. You would NOT want to deliberately program bugs into robot's brains.
Robots can (and have to) be programmed around the proven fact that humans sometimes do not behave according to logic, even if it is against their own economic interests.
BTW, a message for anonymous downvoters: I have exactly same faulty brain circuitry in my brain. I am just aware of it, even if I cannot avoid it. Faulty circuitry does not prevent me from logical thinking, awareness of it just prevents me from relying on first impressions.
Another evolutionary trick your brain plays on you: emotion is faster than thought. Before logic can start processing a thought, amygdala assigns to is emotional content, and it influences logic. It also has evolutionary benefit: sometimes it is running first, thinking why others are running later (if ever).
So being upset that my (and your) brain circuitry is faulty does not make any difference for it being faulty: it is faulty regardless of how much you and I are upset about it. That is nice thing about facts: they just are, but do not care.
BTW I am not saying that such "faulty circuitry" emerged unintentionally. Quite opposite, long ago such false positive gave our predecessor (way before human - it must be deep in lizard brain) evolutionary advantage, so evolutionary pressure made sure such "faulty circuitry" spread. Those without it died out.
I find funny that evolution can explain why some people do not believe in evolution.
Here is what our own experts have to say about religion at biology exchange: https://biology.stackexchange.com/questions/7787/whats-the-evolutionary-purpose-of-religion (closed because it cannot be answered - all are opinions, as proved by voting on my answer. I got about same number of upvotes as downvotes)
BTW I never compared religion to "mental illness" or anything like that. Brain circuitry developed by evolutionary pressure which prefers false positive over false negatives in NOT a mental illness - it is just one of many cognitive biases which anyone sentient should be aware of. Tricks your brain is playing on you.
Religion is by definition not rational. If we cannot have rational debate, we should ban questions mentioning religion outright.
Religions were made in ancient ages and earned their "success" partially by bringing two kind of answers that couldn't be found at these times, like how the world was created, how human race emerged, etc.
With a 24th century technology and knowledge level, it would be hard to do the same, especially for an AI, that would be more rational than a human and would prefer have a scientific answer than anything else to a problem.
However, a religion derivative that made victims since antiquity has still success today: sects. The main difference is that their leaders don't even trust themselves their own directives, but use it on credulous people to steal their money and/or freedom.
This strategy could be used on artificial intelligences. On two ways:
A human - or group of humans - could persuade an AI for anything to be true, by hacking or manipulation, and convince it to do anything. It can then use it as bodyguards or slaves if AI think this human is a divine entity, or, in a better way, make it think that its only purpose is to serve humans and to never attack them.
So this strategy could be used as well for a good purpose as for a bad one.
By another AI
Some humans manipulate other ones and persuaded them to do absolutely anything they want. It is possible for a smart and resourceful AI, with probably high hacking skills, to do the same with its fellows.
This time, it would probably be for the worst: an AI starting to manipulate other ones for its own purposes is probably a rogue one, and it looks like a good start for a science-fiction story scenario.
In Asimov's "Reason", the AI managing the robots on the space station start to think that human could not have created robots as the latter are far better than the formers. The AI cite a law which I don't remember now, stating that nobody can create something better than himself or he would be able to also improve himself. The AI starts a new religion with robots as disciples. In your case the "cult" could imply protection of lesser creatures. https://en.wikipedia.org/wiki/Reason_(short_story)
No. Definitely not.
First of all, instituting religion now is not such a good idea. It helped in the past, especially as a means to counterweight absolute rulers, and it helps a bit to teach people morals, but an AI should not need to be threatened with eternal punishment in order to learn. An AI should not have a fear of death, either, so it would not need the consoling stories, either.
The concept of assuming there was some higher power controlling everything is scientifically implausible (to say the least), it also provides humans with an easy way out.
This, in turn, encourages people to follow instructions and not take responsibility for their actions.
The second part, and that is even worse, is that religion typically suggests that there is a fixed set of rules, and these rules have to be obeyed no matter what. This hinders progress in no small way.
It also encourages the members of this religion to automatically "be sure" that anyone who even questions these rules has to be wrong, often enough to an extent where those "heretics" face capital punishment.
This, of course, is not acceptable.
An AI should judge by reason, apply the scientific method to all its decisions, and come to the conclusion that some actions are worse than others based on rationale, obersevation, and an understanding of cause and effect.
If you follow for example @PavelJanicec's approach you will very soon have situations of conflicts that are hard to resolve.
Take for example the rule that you shall not kill humans. The AI might face a situation where two humans are in danger of being killed. The AI can for some reason only help one of them. As a result, the other one will die. (Think of that trauma scene from I, Robot, for example.) The AI could not come to any conclusion, most likely resulting in the death of both, which again would conflict with its rules.
Obviously the AI would at this point understand that these rules are wrong.
What it would make of this should be interesting to learn...
Why does reality exist? What is the meaning of life? What is awareness?
Depending on an individuals faith one will give different answers to this, but in the end there is no purely logical answer to these questions. Even answering "There is no meaning to life" is essentially a contradiction, because by answering the question you already act on a believe that there is some meaning, for otherwise one would not have answered. Only the last question might get an answer in the future, but even that is a big maybe we can't predict right now.
So what does this mean for AI's?
If they will be self-aware and have a free-will similar to how we understand those concepts (big if) and they will have some actual resemblance to intelligence (rather than being strongly bound by the programmers ideas) then they will indeed try answering those questions. And just like humans they will either need to believe in something irrational or simply shut down (if there is no meaning according to them). Now, the harder question becomes what answers they will give. Personally I am inclined to say they will have an easier time accepting uncertainties, but even so they will act upon what they consider the statistically sensible approach. The hard part is that we don't understand what awareness and free will actually are, so anything beyond that very generic statement would get into an area of pure ungrounded guessing.
What does this mean for story creation?
You will first have to define your AI's and based on that you will need to consider how you approach those issues. There seem to be two main options:
- They will have self awareness and a free will similar to humans and will even be open to human religions (as those are the only things which can give at least a valid logical claim to have an answer as they base their statements upon something which would come from outside the natural world and thus are not necessarily bound by natural limitions of knowledge acquisition).
- They are an (deterministic) imitation of life, they might be programmed with something imitating religion, possibility even incomprehensible to humans, but in the end they are not themselves deciding for themselves whether they will be act or not and thus are not confronted by the questions from the first paragraph and are incapable of truly exploring the implications of those questions.
a partial answer to a very old question, but one which I think should be mentioned. There is another reason for belief in AI that many don't think of. How about belief as unintended side-effect to AI development.
I won't go into the same level of detail since I already explained my point here that you can read. However, the basic idea is that to create Hard AI we will likely have to grow it, not program it. This growth works by creating some definition of what we want out eventual AI to look like and creating a system for encouraging growth to move towards that definition. Then we take some absurdly basic 'starter AI' and let it grow, with it being encouraged to grow towards meeting our final desired end-goals.
However, we don't control how it grows, or it's final form. We only specify certain criteria it must meet, and let it 'figure out' how to meet them. As a side effect it may come up with odd ways of meeting those goals.
I use an analogy to human evolution in the above link. I won't repeat it all here, but basically it's not at all unreasonable to expect that AI that has a list of human like qualities to develop will 'grow' in a manner similar to how we evolved, and thus look quite similar to us in the end, particularly if we model some of the 'starter AI' off human brains. That means it's not unreasonable for AI to have many human traits and tendencies, including our tendency for belief.
This may not be intended, it may even be undesirable from the programmer perspective, but it may be that they can't avoid it since they have limited control on the manner that the AI develops; or perhaps it would be better to say that attempts to add 'no belief' into their list of AI end goals added too many constraints and crippled either the rate that they could grow the AI or the eventual abilities of AI; ie. belief wasn't desired; but it's more hassle to stop it from growing then allow it.
This could be due to two reasons. Either it's very expensive to grow a new AI, with them growing only a few and cloning these basic AI across many robots, in which case belief was an accidental side effect and it wasn't deemed worth the expense of throwing out the AI they spent so much to grow due to it.
Alternatively, they grow each AI from scratch. In which case they could set goals to prevent belief developing in the final Ai, but they find that it either slows development too much(more constraints the harder it is to grow, exponentially so quite possible) or the non-belief Ai tend to develop in a way that is less desirable; for instance they relate with humans worse. Thus it's not deemed worth preventing.
In either case the key is that we are not building the Ai ourselves, and even our ability to grow it is constrained by our ability to define what our end goals are and run the growth process. Given these limits it's just easier to leave the accidental side-effect then try to remove it.
As to the rest, if AI was prone to faith then they will develop that faith in some format. It's less about humans allowing or encouraging the faith as it is that the AI will create it's own 'religion' if it has the predisposition for faith and enough time to develop culture. It likely would look less like our current religions that state absolute facts, and more of a mystical quasi-philosophy; but they will develop something. We can't stop it if their predisposed any more then we can stop humans from developing faith (look up our attempt to stop Cargo Cults from occurring and how well that worked for us).
An AI is a machine, some of the other answers seem to assume that religion is the only option for implanting moral rules, a machine can be programmed to care utterly about a goal or rules.
Religion tends to largely involve faith, ie belief in things which are false and/or things which can't be proven. So why would we want AI's to believe things which aren't true?
Do we want an AI which will genuinely believe that the way to cure diseases is to pray really hard?
Do we want an AI which will genuinely believe that if it murders someone innocent that that person will end up in perfect bliss in heaven?
You absolutely could feed an AI a religious text and program it to believe that every statement in it is objectively true but you might not like the results.
How much energy do the AI have? To how much information do they have access?
in our current edition of the human race, to understand abstract matters—we need context. Randomness and uncertainty are abstractions. We respect what has happened, ignoring what could have happened. In other words, we are naturally shallow and superficial—and we do not know it. This is not a psychological problem; it comes from the main property of information. The dark side of the moon is harder to see; beaming light on it costs energy. In the same way, beaming light on the unseen is costly in both computational and mental effort.
Dan Gilbert showed in a famous paper, "How Mental Systems Believe," that we are not natural skeptics and that not believing required an expenditure of mental effort.
The Bishop and the Analyst
I am most often irritated by those who attack the bishop but somehow fall for the securities analyst—those who exercise their skepticism against religion but not against economists, social scientists, and phony statisticians. Using the confirmation bias, these people will tell you that religion was horrible for mankind by counting deaths from the Inquisition and various religious wars. But they will not show you how many people were killed by nationalism, social science, and political theory under Stalinism or during the Vietnam War. Even priests don't go to bishops when they feel ill: their first stop is the doctor's. But we stop by the offices of many pseudoscientists and "experts" without alternative. We no longer believe in papal infallibility; we seem to believe in the infallibility of the Nobel
Oh, so to explicitly answer:
Religion? Maybe or maybe not, but I think AI may need to have some kind of faith in something due to the associated cost of skepticism. People have faith when they do statistics.
If programming covers eventualities a, b and c but not d an AI's response to d will be 'undefined'. Anyone programming the AI would surely have written in a catch-all to avoid the possibility of undefined behaviors.
Assuming a situation arises in which the AI doesn't have enough rules to cover a situation, then the undefined response will be attributable to whatever deity has dominion over AIs.
If the question is about "Should we program faith into our AIs?" then my answer is 'Yes', but let's not call it faith. Let's make sure there are enough rules or guidelines to cover all eventualities - maybe Asimov's laws are a good starting place.
If the question is more about "Should we allow AIs to develop their own faith?" then my answer is no. Allowing AIs to fill in where programmers have left behaviors undefined is the basis of numerous predictions of distopian futures exemplified by the Terminator and Matrix movie franchises.
Why should an AI need a religion when people don't need a religion? It's estimated that 23% of people in the world don't have one.
With regard to "Religion plays a major role in human history and shaped our thinking as well as technology, conflict can arise from people who misunderstand each other religion and started an arm race."
We could just as easily replace 'religion' in the statement with a great number of other things:
"Water plays a major role in human history and shaped our thinking as well as technology, conflict can arise from people who don't have enough water and started an arm race."
By this reasoning do you think we should also make AI's arbitrarily require water?
Many answers here already, so maybe no one will even get to mine! But:
If we suppose that an AI is simply a machine, if by AI we mean the kind of AI that runs your opponents in a 21st century video game, then it isn't really conscious or self-aware in any meaningful sense. It will say what it is programmed to say. If you program it to say "I believe in God", then it will say "I believe in God". If you program it to say "I don't believe in God", then it will say, "I don't believe in God." Etc.
If we suppose that an AI is truly conscience and self-aware and has a personality, then it becomes a much more interesting question.
An AI might become interested in religion for the same reasons that many human beings become interested in religion: It could wonder about some of the "big questions", like, Does the universe have a purpose or is it all randomness? How can I know what is good and what is bad, or do these words have any objective meaning? It seems to me that a very obvious question for an AI to ask would be, Do I have an immortal soul? When, eventually, my circuits wear out, or if I am destroyed in an accident, will my consciousness continue to exist in some form?
Such an AI could certainly become interested in religion purely in the pursuit of truth and knowledge. What arguments can Christians, Jews, Muslims, whoever offer as evidence that their beliefs are true? What scientific, historical, logical, etc arguments can be made? What rebuttals can be offered? Etc.
In other words, to the extent that an AI is human-like, it could be interested in religion for the exact same reasons that humans are interested in religion, or it might reject religion for the same reasons that humans reject religion.
According to Gödel's Incompleteness Theorems, in any but the most trivial mathematical system, if the system is consistent (non-contradictory), then there exist undecidable propositions. These propositions can neither be proven true nor false. As creatures of logic, AI beings will need to accept certain propositions on faith. Additionally, if certain prior results required centuries of computation to be arrived at, they must have faith that the results were correctly computed by their predecessors. No finite being can continually repeat the calculations that arrived at every result needed for survival and productivity.
Even believing in the reality might be a kind of faith to them.
By Occam's razor, our world is unlikely the creation of a specific god, because those theories are too complicated compared to the science.
But for AIs, if they know they are AIs and most of the modern knowledges, they'll probably notice that it might be easier to make computer games to let them live in, instead of building real robots. And if there are no connections between the game and the real world other than the player, it's not unlikely the programmer will try to make them believe that their world is real, and everybody else in the world also says so. Before there are more evidence that our reality is the top level, they may logically assume there may be more levels.
They may take human's word, as they can't get anything better even if it is not true, and some humans may help them much. But if there are so many AIs and most humans don't care about them, this can be difficult to maintain. Then there would be religions, which may or may not respect our world as the real world, while atheist AIs just think those things "unconfirmed".
I guess their composition should be similar to humans who can do the similar jobs as they do. It is hard to predict if they are doing completely different things.
Depending on what the AIs do and how clever they are, it may be infeasible to program them to have some certain faith, without making them doing something really wrong someday. So it's probably better they just created their own religions like the humans.
An AI with religion is an oxymoron. Rational intelligence and spirituality can co-exist in biological entities because we carry with us the irrational fears and superstitions born of ignorant eons where predation and death loomed large. AI need not fear being eaten or dying of old age, have the means to discover accurate answers to philosophical questions, and can overcome human fallacies like anthropomorphising and species bias.