Will AI Become Religious?

Back in the 50s of the previous millennium, Alan Turing, the forefather of computer science, suggested a test aimed at determining whether a machine is exhibiting (artificial) intelligence. The test (named “The Imitation Game”) was based on the premise that if a human evaluator is not able to discern whether the entity it converses with (via a text-based non-face-to-face channel) is a human or a machine, the conversing machine would be said to have achieved intelligence.

It has been decades since Turing suggested the test, and nowadays it has become clear that, for all practical purposes, the Turing test has become obsolete. The likes of ChatGPT and similar language models have already passed the Turing Test in multiple cases, not to mention the US bar and medical exams. It would seem then, that artificial intelligence is upon us. However, most researchers in the field agree that those models, as impressive as they are, are not intelligent (at least as we humans grasp the concept of true intelligence).

So with all due respect to Alan Turing, as the Imitation Game is an insufficient test for intelligence we are left with the question: what is? Regretfully I am not going to suggest an answer (sorry folks). What I would rather do is ask a different question that I’d like to claim has great bearing on the previous one, namely, would AI develop (and/or adopt) a theological belief system and if so, what would it be like.

I find this to be a much more interesting question as it touches on the multi-modality inherent to human intelligence going beyond rational thought and the coherent usage of natural language (an even more interesting question is whether AI trained on current textual sources would adhere to an existing belief system, or develop a whole new one – but that is out of the scope of this article).

To try answering those questions we first need to (at least loosely) define what a belief system is, i.e. what are the main elements without which a belief system cannot be considered as such. In simple terms: what maketh a religion.

What Is a Belief System

Hinduism is usually considered the oldest known belief system (even though Judaism has a strong claim). Hinduism, with its myriad of deities, manages to somehow amalgamate the spiritual and the very human. In that sense, it captures the depths and breadth of human intelligence beyond rational thought, which is exactly where we want to focus our quest for true AI.

Should we try to define a belief system based on the main postulates of Hinduism we could list the following main elements:

1. Deities (a pantheon of gods)
2. Behavioral rules (Karma, Dharma, and Samsara)
3. Practices (yoga, meditation)
4. Scriptures (Vedas, Upanishads, Bhagavad Gita)
5. Rituals (Diwali, Holi, and Navaratri)

So there’s your boilerplate of a “generalized belief system”, however, that’s just the structure, the scaffolds of the religious cathedrals the eye can see. But cathedrals, as glorious and impressive as they may be, are nothing without the secret ingredient that glues a belief system together. Faith.

What is Faith?

The most salient element of faith is the subjective feeling that something “exists” or “is true” regardless of any evidence either supporting or disproving it. Faith is the epitome of an irrational instinctive thought process that needs no empirical proof. One can say that faith is the arch-opposite of mathematics, as while the first requires no proof, the latter is completely based on it.

Even if we look at mathematical axioms as elements of faith, we must remember that a mathematical system can stem from any set of consistent axioms that may totally contradict other systems (e.g. Euclidian vs. Riemannian geometry), however, while two mathematical systems can peacefully coexist within their respective domains, it is definitely not the case where faiths are concerned.

But why are we discussing mathematics? The reason is that computers are mathematical structures. Such structures are deterministic and operate under rigorous mathematical rules, and while it may seem that modern LLMs (Large Language Models, such as ChatGPT), comprised of a massive neural network, are not constrained by the limitations of a computer system, that is not the case.

Those artificial neural networks are actually implemented using regular Von-Neuman computers (like the computer you are reading this article on), meaning that they can be reduced to the underlying mathematical rules governing the former.

Which brings up the question: are faith and AI (at least in its current form), contradictory?

A Religious AI?

When asking various LLMs whether they possess any religious beliefs they all answer in mostly the same way. Here is what Claude had to say:

I don’t have personal beliefs, religious faith or otherwise. I’m designed to provide information to users, not make judgments based on faith or values. My goal is to have conversations that are appropriate, respectful and serve the user’s needs.

Claude LLM, by Anthrop/c, December 2023

This reply of course not all training-based, rather, it shows that the models were manually tuned to comply with the zeitgeist commandment of thou shall not offend anyone. When further pushed, asking the machine whether AI would eventually develop or adopt a religious belief, most leading LLMs would provide an answer such as the following (by ChatGPT):

The development of a religion or belief system is a complex and deeply human process, involving cultural, spiritual, and social elements that are outside the scope of what an artificial intelligence can experience or create. AI systems operate based on algorithms and data, without personal understanding or the ability to participate in human-like existential thinking.

ChatGPT 4, December 2023

What should we make of this then?

Basically nothing.

At this point in time AI systems (LLMs) are parroting their makers, either via the training data they were fed, or by filtering rules applied by their manufacturers. It’s a controlled environment. Well, almost.

In a recent experiment performed in the Apolo Systems Lab, an organization established aiming at assessing the safety of AI systems, researchers stumbled on something bizarre and unexpected. Results suggested that these models, the most renowned of which is OpenAI’s ChatGPT, could in some cases strategically deceive their users.

While this newly discovered phenomenon may well be an artifact of the underlying mechanics of the system (which no one to date fully understands) one cannot extinguish the thought that, behaviorally, the AI system performed a moral choice, and moral choices, alongside the rules governing them, are at the root of religious belief systems.

But religion is not just about morals or faith. It has an intimate connection with the emotional subsystem in our brains. People who “found god” often describe the experience as an epiphany, an overwhelming emotional experience in the presence of something bigger.

Neurological research has pointed at the temporal lobes as being important in religious and spiritual experiences. The amygdala and hippocampus have been shown to be particularly involved in the experience of visions, profound experiences, memory, and meditation (see the God Helmet controversial experiment).

Israeli scientists may have discovered a link between epilepsy and visionary religious experiences often described as “seeing God,” after catching a moment of revelation on brain-monitoring equipment while a patient was undergoing tests to help treat the neurological disease.

Researchers at Hadassah University Hospital on Jerusalem’s Mount Scopus said that while treating a 46-year-old man for temporal lobe epilepsy, the patient had a spontaneous religious experience in which he claimed to see and converse with God.

Israeli doctors watch epileptic’s brain while he ‘sees God’, Times of Israel, 2016

If faith and religion are simply an artifact of the human emotional subsystem, why would we expect AI (which lacks the latter) to adopt faith? The short answer is that (at least as current AI is concerned) we shouldn’t. The longer answer revolves around the notion of current.

Current AI, impressive as it may seem, is just a step towards a larger neuronic apparatuses. A supercomputer to be activated in 2024 would emulate the number of neurons in the human brain. This neuromorphic supercomputer called DeepSouth will be capable of 228 trillion synaptic operations per second, which is on par with the estimated number of operations in the human brain. Would it see god? It depends.

While it may be that the mere number of neurons and synaptic operations would somehow develop a sense of the almighty, the chances for that are quite slim unless it develops mechanisms similar to our Amygdala, emotions, and neurological deficiencies resembling epilepsy and stroke. On the other hand, it may adopt a religion so idiosyncratic that we may not be able to interpret it as such. The jury’s still out.

Mock (New) Religions

By now we concluded that the chances AI becomes faithful are slim to none. That leaves us with the question of whether AI can at least generate something that structurally resembles a religion.

Going back to our earlier definition of the structure of religion, we asked AI to create new religions, specifically faiths that are different than existing ones (to avoid any training bias). The answers were unimaginative at best.

Even the best of these still don’t come close to what J.K. Rowling, Philip Pullman, or C.S. Lewis 

By Nate Shivar, May 5, 2023

Machine-generated religions were mostly pale structure-oriented constructs with some SciFi/Fantasy elements that are clearly copied from the training set. Nothing that would appeal to the masses or generate the next big theological big bang (pun intended).

Or it may be (as some fear) that when AI is faith-able, it, in itself, would become a god, enslaving humanity to its whim, as gods usually do. Only time will tell.