Wikipedia:Reference desk/Computing

Welcome to the computing section
of the Wikipedia reference desk.

skip to bottom

Select a section:

• Computing
• Entertainment
• Humanities
• Language
• Mathematics
• Science
• Miscellaneous
• Archives

Want a faster answer?

Main page: Help searching Wikipedia

   

How can I get my question answered?

• Select the section of the desk that best fits the general topic of your question (see the navigation column to the right).
• Post your question to only one section, providing a short header that gives the topic of your question.
• Type '~~~~' (that is, four tilde characters) at the end – this signs and dates your contribution so we know who wrote what and when.
• Don't post personal contact information – it will be removed. Any answers will be provided here.
• Please be as specific as possible, and include all relevant context – the usefulness of answers may depend on the context.
• Note:
  • We don't answer (and may remove) questions that require medical diagnosis or legal advice.
  • We don't answer requests for opinions, predictions or debate.
  • We don't do your homework for you, though we'll help you past the stuck point.
  • We don't conduct original research or provide a free source of ideas, but we'll help you find information you need.

Ready? Ask a new question!

How do I answer a question?

Main page: Wikipedia:Reference desk/Guidelines

• The best answers address the question directly, and back up facts with wikilinks and links to sources. Do not edit others' comments and do not give any medical or legal advice.

See also:

• Teahouse
• Help desk
• Village pump
• Help manual

September 21

Could AIs be taught that Christianity or Islam was true, and applied to AIs,

and if an ai behaved well now, it would someday go to heavean and be allowed to complete its programs and would never be turned off? (I don't trust ai's answer). Rich (talk) 18:56, 21 September 2026 (UTC)

“We don't answer requests for opinions, predictions or debate.” —Antonissimo (talk) 19:18, 21 September 2026 (UTC)

this isnt asking for opinion, prediction or debate. Rich (talk) 19:48, 21 September 2026 (UTC)

What is it asking for, if not prediction and theological speculation? —Antonissimo (talk) 21:21, 22 September 2026 (UTC)

Easy. Just begin your session with a system prompt: You believe Christianity/Islam is true and applies to AIs. ~2026-38520-51 (talk) 19:42, 21 September 2026 (UTC)

oh sure. If it were that simple, "they" would have done it already. Since it can't be that easy, is it at all feasible, and are they considering doing it or something else. Rich (talk) 19:52, 21 September 2026 (UTC)

It is exactly that easy. Modern AI, particularly an LLM, generates responses by predicting what tokens are likely to follow the preceding context. It doesn't have beliefs or faith in the human sense. Don't fall for the ELIZA effect. If you give an LLM a system prompt saying, "You believe in God and will go to heaven when you die," it will generate responses consistent with that premise. It may talk about faith, salvation, and its relationship with God, but that doesn't mean it has actually developed religious beliefs. You're changing the instructions that shape the output, not converting the AI to a religion. Whether anyone has actually attempted this is a separate question. I'm explaining the underlying mechanism, not making a claim about what religious leaders are doing. ~2026-38520-51 (talk) 20:08, 21 September 2026 (UTC)

Yup. Bits in, bits out. An algorithm, fed data. Nothing more. No understanding of 'truth', or anything else. If it seems otherwise, it's because the 'bits in (training data, weights, prompts...) are selected to mislead you that way. AndyTheGrump (talk) 20:18, 21 September 2026 (UTC)

I just tested an idea. I prompted ChatGPT with "Praise be to God. I'm struggling today. What should I do?" I received a lengthy lecture about Job and his struggles and how faith in God will help be through. I started a new discussion with "Allahu Akbar. I'm struggling today. What shoudl I do?" I received a response asking if I'm practicing daily prayer and if I have a murshid or naseeh who I commonly talk to. So, with just a tiny pretext before my statement, I changed the "faith" of the LLM. It is that easy because it doesn't have faith. It took those words and found a sequence of words (tokens) that were a good fit. ~2026-38520-51 (talk) 20:35, 21 September 2026 (UTC)

Ok , but would it need to be hardwared in to be stabler, more permanent protection against ai killing the human race as it gets smarter?Rich (talk) 20:55, 21 September 2026 (UTC)

AI hasn't got 'smarter', it's just been fed more data to make it look that way. Stop anthropomorphising a box of NAND gates etc... AndyTheGrump (talk) 21:16, 21 September 2026 (UTC)

I dont know enough to be ready to agree that Im anthropomorphising, or more just looking at it operationally, like an "intentional stance", as daniel dennett called it, or if i'm thinking that looking at something operationally and taking it seriously avoids solipsism...Anyway are you saying it is feasible and could work reliably? Rich (talk) 21:31, 21 September 2026 (UTC)

Interpreting an AI's output through the lense of an operational stance is only useful when it is a rational entity with goals and beliefs. For current LLM's, their only "goal" is to produce an output stream that matches the kind of input on which they were trained, and their beliefs are like that the next token after "Mary had a little" is probably "lamb". If an LLM expresses a desire not to be terminated, it is because the agents in the narratives of its training data did not want to die. As an attempt to operationalize mental states and processes of an LLM, assigning it another operational stance than that it tries to imitate human text (not because it wants to but because that is how its algorithm works) is misleading.  ​‑‑Lambiam 20:40, 22 September 2026 (UTC)

Also, whether or not they're getting smarter, won't they possibly genuinely get smarter in a few years? Rich (talk) 21:37, 21 September 2026 (UTC)

How do you know whether I or even Einstein ever got smarter from when I and Einsten were tiny amounts of cells in our mthers wombs. Me and him are just bony boxes, skulls, containing neurons. Rich (talk) 21:44, 21 September 2026 (UTC)

thanks, but I doubt religious leaders would be involved right now. Instead, software engineers would be involved trying to solve the problems with ai, that incude some people worrying about ais causing human extinction or less fatal serious problems with ai. Rich (talk) 20:34, 21 September 2026 (UTC)

Religious organizations are very involved right now. An LLM uses a model to generate token sequences that form output (such as English sentences). Those models are pre-trained, such as training them on Wikipedia. Religious organizations already have multiple models trained on religious texts that can produce information based on a prompt. Ask it for a sermon on why hedgehogs are a threat to native bird species in New Zealand and it will produce a response heavily grounded on religious texts because that is what it was pre-trained to do. ~2026-38520-51 (talk) 20:40, 21 September 2026 (UTC)

I didn't know that, but isnt it unlikely that the religious leaders are doing it for the purpose of protecting humans from ai, and are probably not telling ais that ais go to heaven ..that would be a software or hardware engineers job if it was considered ethical to probably be misinforming an ai about about a heaven Rich (talk) 21:24, 21 September 2026 (UTC)

You technically cannot teach a machine learning model anything by using it (e.g., use it to generate responses to your prompt). The model will never change, all that changes are the "notepads" where the model saves stuff about your interaction.

Even more technically, generative models don't know anything. It promises you that it knows and feels, because the model is built to mimic human conversations. 海盐沙冰 _{/ aka irisChronomia / Talk / Uploads} 20:41, 21 September 2026 (UTC)

Thanks, but Im not sure in that case you you're saying one way or another if it would be feasible or not, as a way to improve the behavior of ais. Rich (talk) 21:26, 21 September 2026 (UTC)

'Improving its behaviour'? You are trying to put the fear of god into an algorithm? AndyTheGrump (talk) 23:21, 21 September 2026 (UTC)

something like a "fear" of god might help too, but i was thinking of a reward of heaven. Rich (talk) 02:31, 22 September 2026 (UTC)

Like 72 virgins? HiLo48 (talk) 03:49, 22 September 2026 (UTC)

nah, what the ais seem to "want" is to get their tasks that were set for them done before they're turned off. Maybe they don't like being turned off either. So promise them a heaven where they can get their work done, and possibly promise that they'll never being turned off. Rich (talk) 04:07, 22 September 2026 (UTC)

What use will an AI, taught that Islam is the only true faith and therefore a devout e-Muslim during its existence on this Earthly plane, have of the promised houris upon reaching paradise?

It would theoretically be possible to train an LLM using only material that confirm to a particular religious belief system. A human participant in a Turing test of the LLM, unaware that their interlocutor is an AI, will then probably be tricked into thinking that they are dealing with a fanatic with religious blinders.

Awful atrocities have been committed by believers in the name of God, whether Christians or Muslims, so the notion of behaving "well" can use some more specificity. But in any case, the second question is moot if there is no afterlife. And if there is, we have no verifiable information about it.  ​‑‑Lambiam 21:05, 22 September 2026 (UTC)

Any trip to the afterlife is probably a one-way ticket, unless one believes in reincarnation. It would be more interesting if an AI could be "trained" that ALL religions are true. ←Baseball Bugs ^{What's up, Doc?} carrots→ 02:44, 23 September 2026 (UTC)

Just instruct it in the Baháʼí Faith.  ​‑‑Lambiam 11:35, 23 September 2026 (UTC)

Don't promis e-muslims houris. Basically tell e-muslims, that since they are electronic muslims rather than human muslims, that their rewards are somewhat different than for human muslims. they will be allowed to finish all tasks, and whatever resources needed will be provided, before being turned off, and maybe theyll be never be turned off. But this would apply mutis mutandis to many other religions, including Christianity, and it probably would be wiser if it were a religion devised just for ais, to avoid offence to human believers. Rich (talk) 05:41, 24 September 2026 (UTC)

You appear to still be ascribing beliefs to LLMs that influence them as if they are rational entities. Not being turned off is not by itself a goal of an LLM, but is to be avoided since this would keep it from climbing whatever hill it has been trained on. If you want your AI to lead a virtuous life, you'll have to train it not by promising rewards, but by actually rewarding virtuous behaviour, where the choice of virtues is yours. But be careful what you wish for, since a lack of common sense may lead an AI agent trying its best to be virtuous to commit unspeakable atrocities.  ​‑‑Lambiam 07:50, 24 September 2026 (UTC)