Religion as a Large Language Model

 

I cured myself of a bad case of Evangelical Christianity in 1980 but I needed to crawl out of the trap from the inside. I spotted the trick. I'm happy to share it here.

CHATBOTS

Chat-GPT is in the news these days. Microsoft has incorporated it into the Edge browser with impressive results. It may be the first time many people have come into contact with a state-of-the-art AI application. The air is thick with nonsense about this particular application, but most agree it is pretty spooky, even scary. Will thinking machines replace humans?

To jump ahead in my argument a bit, I think the answer is that we all have the ability to run a "chatbot" in our heads. We learn a language in a manner very similar to the way chatbots are trained.

LARGE LANGUAGE MODELS

Large language models are a well-established engineering technology. The technology produces stunning results, even if you kinda know how it works. To me, this is like the feeling of wonder I have whenever I see a jet transport take off. I know it's not magic, but it sure feels like magic. 

My own interest in AI goes back to my 1968 Thesis. Back then, I was sure we could use "AI" to pool all human knowledge, then converse with, say, Einstein as he would have talked.  That's pretty much what CHAT-GPT does. 

Of course, DaVinci pictured helicopters 500 years ago, but his designs could never have worked. The devil is in the details. What I envisioned was nothing like what actually works, nor did it use the "artificial intelligence" methods I was familiar with. 

Chomsky won a Nobel Prize for the theory that our brain works kinda like a computer. It turns out to be true, but not in the way Chomsky (and everyone else at the time) thought.

THE STOCHASTIC PARROT

Emily Bender invented the term "Stochastic Parrot." It was not intended kindly. She had much to say about why chatbots seem to "know" things. Who would be accountable? What if people believed the parrot as if it were human? 

She seems to have missed the problem of humans believing humans who just "parrot" ideas without true understanding.

THE CONSCIOUS LLM

Last year, a Google employee (Lemoine) raised a huge stink by claiming his LLM (an implementation of the LAMDA LLM) was self-aware. It was afraid of being turned off.  Incorporation of CHAT-GPT into the Microsoft Edge browser assures me that Leomoine will soon have a lot of company down this rabbit hole.

But my point does not concern whether LLMs are "conscious." It has to do with whether we are conscious. Or, to put it more gently, whether we are conscious all the time, especially when we are talking when we have no idea what we are talking about.

CHAT-GPT "KNOWS" A LOT ABOUT RELIGION

This is what CHAT-GPT confidently tells me about the resurrection:

Yes, our bodies will be raised not spiritually or ethereally but physically and materially. Our souls will be reunited with our transformed physical bodies, brought back to life from the dead. Scripture teaches this in many ways¹².

CHAT-GPT is just regurgitating what it finds in its "Large Language Model." It has no idea what it's talking about. In fact, no idea at all. Just a very big set of mathematical equations.

But here is the $64,000 question: Why would a human being ever make such an outrageous claim? It seems to me that the process is similar, if not identical. The human being relies on a "Large Language Model" of relevant (Christian) writings and spits out something that sounds "intelligible" but is utterly disconnected from actual human experience. You might fairly ask how CHAT-GPT "knows" about the resurrection. But the same question can be asked of the preacher. Many claims that true AI is impossible because an AI can never have a body and experience the world as we do. But you would think anyone with a body would not make the claims here.

I would make a broader claim: Religion could not exist without language, specifically Large Language Models, that provide a deep and rich way of stringing together sentences that are (for the most part) internally consistent. Different religions can get away with mutually contradictory language models because these models have no connection with actual human experience. Religous LLLms are about themselves.

I "cured" myself of a bad case of Evangelical Christianity by simply resolving not to use religious terminology. After about two years, it became obvious that I had about as much use for that language model as I had for Chinese. Questions like "Is there a God" are also part of the Christian model, as are any number of "canned answers." 

Even "I don't know" is part of the model (doubt). I can't even be an agnostic since "doubt" is unintelligible outside the Christian LLM. It's a problem Chinese people have.