The God of AI

One of John Higgs recent newsletters is, to horribly summarise1, about how modern culture seems to have failed to produce gods for two incredibly powerful forces in the modern age; climate and AI.

I’m not sure about climate, but I think Higgs is wrong; there is a god for AI, hiding in plain sight all across our culture. We're talking about it all the time.

Sometimes when we talk about "AI", we're talking about data centres, GPUs, machine learning and multi-billion parameter matrix maths calculations. Very real - perhaps even mundane.

But sometimes – perhaps most of the time – when we talk about "AI", we're talking about something different. A powerful, seemingly inevitable force that's growing, reshaping the economy, communications, education and entertainment like an oncoming storm. And I think, once you spot the clues, its usually pretty easy to spot which one is being talked about.

Cory Doctorow posted about some conversations with Riley Quinn on the subject that makes this distinction – "LLMs are real, AI is fake";

"AI" – chatbots that wake up, "set their own goals," and "spontaneously" start hacking servers – is fake. It doesn't have "a 10% chance of ending the human race." The Hugging Face hack isn't a mysterious, supernatural occurrence. It's a Python loop and a chatbot. The people responsible didn't accidentally create god: they created autonomous malicious software and then failed to closely monitor it, resulting in it doing something both foreseeable and bad.

But we often talk about "AI" when we really mean "LLMs", so it can be hard to distinguish between the two – but if we pay attention to the grammar or the language we use, it slips out.

  • We use the 'zero article' - which is when a noun doesn't get a 'the' or 'a' in front of it
  • That also means that we don't pluralise; there are lots of "AIs", but when we use the zero article, it implies "AI" a singular 'thing'.
  • Grammatically, this is also how we use names; the name of the AI god is right there, hiding in plain sight.

From Douglas Adams' The Long Dark Tea-Time of the Soul (a book about gods on Earth, including the creation of a new one);

An "Act of God". Merely a chance, careless phrase by which people were able to dispose conveniently of awkward phenomena that would admit of no more rational explanation. But it was the chance carelessness of it which particularly appealed to Dirk because words used carelessly, as if they did not matter in any serious way, often allowed otherwise well-guarded truths to seep through.

Consider these three statements;

  • AI is really bad at maths.
  • AI is pretty good at maths
  • AI is better at maths than most of the best mathematicians in the world.

They obviously can’t all be true – unless we spot that grammatical trick. If we add the indefinite article;

  • An AI (by which I specifically mean the small LLM running on my old laptop) is really bad at maths. (I can watch it constantly getting the answer to complicated questions wrong.)
  • An AI (by which I specifically mean ChatGPT with tool-calling capabilities) is pretty good at maths.
  • An AI (by which I specifically mean a swarm of thousands of GPT agents) is so good at maths, it’s raising similar questions about the whole point of mathematicians that Deep Blue did about chess players in 1997.

When we remove the article altogether (the zero article) we stretch the scope of what "AI" can refer to – and the best version wins. 'AI' today is really good at maths, no matter how bad the LLM on my laptop is.

But it stretches further than that. 'AI' doesn't just cover real technology today, but potential technology of the future.

Are LLMs conscious? Definitively, no: they are just matrix calculations, and I strongly believe that you can't create consciousness by just adding and multiplying numbers together – no matter how many billions of times you do it.

Is an AI conscious? Again, no – we've seen lots of AI systems, but none of them has ever had any credible claim of achieving consciousness.

But is "AI" conscious? Well - we can't definitively say that no AI system will ever be conscious, so… maybe? Changing the grammar changes the meaning of the question, which changes the answer.

The name of the god is AI

Names are important. On a practical level, some therapies uses naming as a way to define and contain a problem, turning it into something that can be studied and treated. It literally changes how your brain works).

One of the odd things about gods is that they like to hide their "true" names; because being able to name something gives you a certain type of power over it. The Egyptian god Isis gets power from learning the sun god Ra's true name. The god of the Old Testament is often known as "Jehova" – which sounds like a name, until you learn that it translates to "I am that I am" – which sounds no more like revealing a "true" name than someone who goes around introducing themself as "The Doctor".

When we use this kind of grammar – the zero article, resisting pluralisation, casually capitalising – however carelessly, we're usually talking about something omnipresent; something that is structurally part of reality itself. The kind of words that can sometimes be capitalised for extra impact;

  • Time marches on
  • Nature finds a way
  • Fate will decide
  • Death comes to us all
  • God loves you

Data centres are real. Matrix maths are real. Large language models are real. But this "AI", the one that we only ever talk about in the zero-article singular as if it were a name, is not real. Its building a myth – and I think its important to spot the difference between the two.

AI's creation myth

In a 1998 talk titled "Is there an Artificial God?", Douglas Adams said;

Man the maker looks at his world and says 'So who made this then?' Who made this? - you can see why it's a treacherous question. Early man thinks, 'Well, because there's only one sort of being I know about who makes things, whoever made all this must therefore be a much bigger, much more powerful and necessarily invisible, one of me and because I tend to be the strong one who does all the stuff, he's probably male'. And so we have the idea of a god. Then, because when we make things we do it with the intention of doing something with them, early man asks himself, 'If he made it, what did he make it for?'

We saw the world as a 'creation' and decided that there must be a 'creator'. We make things with a purpose, and decided that there must be a purpose behind the making of the world. We saw an effect, and filled in the blanks where we expected to see a cause.

So, where is the 'creator' of modern AI?

We can talk about figures like Geoffrey Hinton, Yoshua Bengio, Yann LeCun or Demis Hassabis as "godfathers of AI" – but its hard to associate them with things like ChatGPT or Claude; none of them ever worked at OpenAI or Anthropic, and none of them are exactly household names. They just aren't part of the AI myth in culture, or the public consciousness.

We live in a post-PC, post-internet world; we've seen Watson (Junior and Senior) drive IBM and the first wave of computer mainframes, Noyce and Moore drive Intel semiconductors, Steve Jobs drive Apple Computers, Bill Gates drive Microsoft, Larry and Sergey drive Google, Jeff Bezos drive Amazon, Mark Zuckerberg drive Facebook/Meta – wave after wave of technologies building on top of one another, shaped by the visions of (mostly) founder CEOs. Obviously, they all had help – but the 'creation myth' for each of them starts with a man and a vision.2

But AI is different – the neural network, transformers, diffusion models, LLMs, computer vision etc. all have their own 'godfathers', and various companies have all played their part in shaping today's AI systems. But no one person – or even one company or team – can be said to be the one visionary 'founding father' of "AI", bringing all the pieces together. So who made it?

On the frontier of the development of AI tools, its often a mystery how these things actually work - a common theme of AI research has been "we set out to show that an LLM cannot do this thing, but we discovered that they can so we published." A 2018 article titled "Has artificial intelligence become alchemy?" describes researchers gravitating around "cargo-cult practices", relying on "folklore and magic spells", and adopting pet methods to tune learning rates without understanding why one is better than others. Andrej Kaparthy has described it as like an "alien tool" that "comes with no manual and everyone has to figure out how to hold it and operate it." It feels like this technology has come out of nowhere; rather than designing it, researchers are discovering it. Again, raising the same treacherous question - where did it come from?

But it turns out that the creation myth has already been written. One of the promises of "true" AI – that is, the Artificial General Intelligence that comes after3 the LLMs we have today– is 'recursive self-improvement'; the idea that the code that the AI writes will become part of the next generation of AI4.

So – according to the prophecy – the father of "true", superintelligent AI will be AI.

I mean; how mythic can you get?

  1. I'm sure an LLM could do a better job, but I'm still being precious about typing out my own blog posts – including the em-dashes. ↩

  2. Still, its always a man. ↩

  3. Take that either as a reference to "A computer whose merest operational parameters I am not worthy to calculate", or John 1:29–31. Whichever you prefer. ↩

  4. Google say that 75% of their code is written by AI - but then, they would say that, wouldn't they? ↩

Info
Written by Scott Thompson On In Tagged , , ,