It isn't magic, it's math

Computers are calculators, they are not people or 'beings' of any kind.  They compute - they do not discern.  They are machines that crunch numbers, they do not have insight. They can be programmed to convert numbers to text.  We can use them to calculate the statistical associations between text fragments contained within vast accumulations of documents.

What underlies so-called "artificial intelligence" (AI) like ChatGPT is nothing more than a large collection of computed probabilities gleaned from word associations and sequences. These probabilities are often based on billions of human-generated textual documents. Statistical probabilities, derived from large collections of human documents, offer valuable insights. But the mystical overtones being used to describe advances in artificial intelligence are starting to get downright weird. I'm afraid much of the shock and awe in the popular press, and on social media, is beginning to mislead many people into adopting a quasi-spiritual understanding of what is happening in technology.

When you type in a question on ChatGPT , or using any other language model, what you are doing can be thought of as something akin to pressing the keys on a piano. ChatGPT converts your question into a numerical representation and maps that numerical representation against its sea of probabilities - the ones previously derived from a sprawling universe of documents. Then, it generates a sequence of English text based on where your question fell within its sea of textual possibilities. The collection of probabilities maintained by these AI models is so large that they can calculate resulting text which is properly formed English and has a coherent, and usually relevant, meaning.

AI models do not have their own ideas.  They regurgitate strings of text cobbled together from statistics.  That means that the quality of their answers is entirely dependent on the quality of the documents they derived their statistics from. If the document corpus is full of errors or lies, an AI will regurgitate an approximation of those errors and lies. If the document corpus is full of political bias, it will generate answers with a corresponding political bias. AI models do not "know" what they are doing.  They only parrot reconstituted approximations of what they have been "told".  In a sense, ChatGPT and all of its siblings can be thought of as massive juke boxes which can "replay" variations of the conventional wisdom on any particular subject they have been loaded with.

The mysticism currently in vogue for describing AI, especially in the popular press and on social media, is rooted in part (I suspect) in the widespread acceptance of the computational theory of mind.

In philosophy of mind, the computational theory of mind (CTM), also known as computationalism, is a family of views that hold that the human mind is an information processing system and that cognition and consciousness together are a form of computation.

For Christians (and really any thoughtful theist I imagine), the computational theory of mind reflects an abiding materialist superstition – that the material world is all there is.  It is a philosophy of the human mind that conceives of people primarily as mathematical and computational -- we are all just biological machines.

Soulless. Spiritless. Mere watery bags of matter. Having the illusion of meaning and love, but really only wandering through life doing a lot of math.

The spiritual, moral, and social implications of this conception of the human mind are vast and dire, but they are beyond the scope of what I wanted to write about here. My larger point is just that this perspective regarding the human mind motivates some amount of the mystical perspective surrounding AI.  Because, if human beings are nothing more than machines, then something that is no more than a machine can be a human being.

There is yet another motivation which I suspect is behind the high-falutin mumbo jumbo surrounding AI: the propensity of human beings to worship the works of our own hands. God himself called out this tendency in the second of his ten commandments. There is an elitist conceit that often attaches to inventiveness and seems to be further compounded by any piddling mastery of technological complexity. AI is hard for people to understand, and so the people who do understand it are, I suspect, tempted toward pride, and toward deifying the work that they themselves are able to do.  

Thus, leading minds in AI believe that they are on a path toward creating digital beings that may actually have human rights. They are openly speculating about the ethical implications of unplugging an AI - that doing so might actually have similar moral implications as aborting an unborn human being.  (My own lack of squeamishness on the question of unplugging an AI can be found here.)

But, alas, these are the inevitable questions one is faced with when you believe that human beings are ethically indistinguishable from calculators.

For me to suggest that AI is nothing more than deep statistics is not the same thing as saying that I think AI is no big deal, or that there is nothing to be concerned about.  AI is going to offer very valuable advances in many fields.  It is also going to be economically disruptive and that disruption will likely lead to an extended period of adjustment and displacement in the marketplace, as the full economic impact of AI is being absorbed.

I have written elsewhere that my primary concerns about AI are in its utility for human deception.  The uncanny statistical results that are made possible by a combination of computing power and vast document collections can be easily exploited to deceive the unwary.  Coupling AI with visual content will only multiply the deceptive quotient, since human beings are even more susceptible to visual deception than when we are merely written.

Below is an interesting experiment done just this week with ChatGPT.  It leverages AI to create video game characters who can interact with gamers in quasi-human ways.  Of course, everything the monk is saying in the video is just statistically chosen words from a model of word probabilities.

What I find especially interesting about this video is the choice made by the developer regarding which character to use - a monk - to embody the underlying AI. Out of all the possible characters that could have been chosen, the developer chose one that carries a cultural and historical aura of spiritual authority.  I doubt that was an accident, even if the developer was not entirely conscious of what he was doing.

Notice also that an English accent was chosen for the character. Some psychologists believe that an English accent creates an impression of greater believability on the part of the speaker.

We're living through a Wizard of Oz moment in which awe and reverence toward AI is being actively cultivated within the popular culture.  But just as in The Wizard of Oz itself, the reality behind AI wizardry is rather less awe inspiring and magical than the hype would have us all believe. AI is just a really powerful calculator. (And maybe also an opportunistic vehicle for math nerds to be rock stars for the first time in their entire lives. LOL!)

At the end of the day, though,  behind the curtain there is no wizard, only math.