Desire Machines

Lately I've been thinking about the progression of "AI"... not so much about the tooling or capabilities which everyone talks about, but rather how we use and, dare I say it, relate to these tools. So here we go.

In dealings with my clients and in the stories I've heard other coders tell, it occurred to me that we've gone through a some major phases in our relationship with LLMs so far and that there's something important waiting for us at the end. And here's a spoiler alert... it's not something important like general artificial intelligence, but rather it's something important about ourselves.

For the record, I'm not suggesting this is a progression, but more of a layering as the tools and our experiences with them have developed and become more complex. Having said that:

A Brief History of AI

Phase One: Here's a Fun Toy

First there was the novelty phase where we would try to trick ChatGPT into saying something offensive and give us a recipe for something dangerous, use Stable Diffusion to generate a weird picture of a thing with too many eyes, or have a suspiciously erotic chat with a low-fidelity virtual friend a-la Replika. Exciting but essentially pointless (aside from the entertainment value) pursuits.

Phase Two: Give Me a Simple Answer

Next we began the process of inserting AI answers in place of existing search tools. Ask Jeeves became Hey Siri and even though the quality of the answers was typically suspect at best, we couldn't resist the urge to keep asking in the hopes that soon enough, the answers to our simple daily problems would be handed to us trustworthy, thorough and true.

Phase Three: Make This Make Sense to Me

It wasn't long after that just getting simple answers was no longer sufficient. Okay, so now I know the what or how of the thing, but now I want to understand the why the thing. LLMs quickly went beyond simple answer machines into explaining the context and meaning around the things we wanted to know. This was the dawn of AI as teacher, storyteller and guide.

Phase Four: Help Me Feel Better About This

And of course, right on the heels of understanding is acceptance... once we were able to get the full picture of a thing from our LLMs, even if that picture was sometimes distorted or worse, moulded specifically to fit our expectations and identities, we returned to AI as friend, therapist, confessor and even lover... to help us make sense of our thoughts and feelings about things which we may not want to share with each other.

Phase Five: Help Me to Become a Better/More Productive Person

And then there are those who view AI in its current form as a tool for introspection and self improvement... perhaps somewhat more aware of the nature of its interaction with us and willing to use it to sharpen their own skills and perspectives and to check their biases and assumptions against a (theoretically) unbiased source.

Phase Six: Give Me What I Want Without Me Having to Ask For It

Of course the dream of the developers of these technologies is for LLMs to develop into Advanced General Intelligence... smart enough to understand our desires in real-time as we have them... though it seems that the limitation we've encountered at this point isn't the ability of LLMs to understand our words, but rather our ability to frame our questions clearly enough that the LLMs can work their magic and grant our wishes.

Phase Seven: Help Me to Understand What I Want

Which brings us to what I think is the currently emergent phase of AI... a tighter and tighter iterative loop of asking and fulfillment out of which we hope to produce perfect outcomes... the "build me a game I want to play" stage where develoipment costs are so low that people are creating one-shot tools in real time to meet needs that aren't even formed, much less clearly articulated.

The Source of Desire

I've heard old-school coders lamenting the arrival of AI, not because they're afraid of getting pushed out of the market, but because the act of coding with AL takes away all the fun. And by fun, they're referring to the analysis, planning, problem solving and reward that comes from completing a particularly complex piece of code that solves a difficult problem elegantly. The speed at which code is created and deployed has, they maintain, has been stripped their joy from the process.

We're getting to the point where the loop of desire to fulfillment is getting so tight that it feels like desire is getting squeezed out of the process. I'm not talking about appreciating what we have or even taking a moment of satisfaction in our achievements... I'm saying that the whole process of want itself is being subverted. How can we have time to develop concrete desires about anything when everything we desire is granted to us nearly immediately? For me this begs the question: how do we get what we want if we don't have time to figure out what it actually is that we want?

And also a more fundamental question... where desire come from?

This is the underlying question I'm getting at: Do we start with some generalized desire and then browse the world until that desire is paired with the will to acquire a thing? Or do we start off with a particular thing that we want and then search the world in hopes of finding it? And what happens when the loop closes and the delay from want to get becomes so brief that desire barely has time to catch fire before it's quenched?

Take the map of AI development and zoom out and you'll see the map of the development of the consumer Internet following the same pattern. First the era of the browser, then of the search engine, and now of AI... from browse to search to "ask and ye shall receive". Now zoom out again and you'll see the same pattern evolve in consumer technology more broadly... from printed catalogues that let us craft wish lists for eventual fulfillment to shopping malls that brought selected items together in one place for us to peruse to e-commerce websites that deliver whatever our hearts' desire, typically within 24 hours or less.

And this pattern likely maps onto the broader human experience as well. Think of the era of agriculture where we were so consumed with the business of just staying alive that even simple desires were delayed for years or decades, or perhaps never met at all, to the industrial era which enabled the achievement of a quality of life driven by conveniences and meeting consumer demand but in which higher pursuits were limited to a select few to the information era where material needs have become largely commoditized and creative and intellectual pursuits... the fulfillment of inner desires... has become accessible to nearly everyone.

Heirarchy of Needs

So we return to the question... where does desire come from? One might argue that desire is an primal factor which comes to any living thing from need and lack... that we naturally desire the comforts and pleasures we don't or can't have... but once the basic needs are met, new and more complex desires appear to take their place. So perhaps desire is more of an intellectual or spiritual thing... a natural property of the human experience... that we come into existence with a "hole" in our being that can never truly be filled.

But consider, if you would, the possibility that desire beyond the base level of food, shelter, territory, mating and survival which are arguably part of the definition of life itself, is not actually an intrinsic component of humanity but, rather, a construct... something impressed upon us by external forces. What if, at the heart of it all, our desire can never be fulfilled because it isn't truly ours in the first place?

Just Tell the AI What You Want

And here's my evidence. With LLMs... at least in the realm of programming... we went from "tell it what you want" to "prompt scripting" to "Prompt Engineering" to the latest "agentic coding based on strong requirements documentation". At every step along that path the problem wasn't "the AI can't give us what we want", but rather "we have to get better at asking it for we want". AI has become the clever and somewhat nefarious Genie which can do almost anything, but it's up to us to be very explicit about how we craft our wishes, less they fall flat or worse, blow up in our faces. And that's the key... which I think sci-fi author Douglas Adams understood all the way back in 1978... we can ask our computers to give us the ultimate answer to life, the universe and everything... and they will. But the answer won't mean anything to us unless we truly know what the question is.

The more I interact with people interacting with LLMs, the more I believe that none of us really know what the question is or even understand the nature of questioning at all. The more I experience of people receiving whatever they want with little to no effort, the more it seems that all of the effort of wanting was due less to the appeal of the destination than it was to the obstacles we found along the path. The more that technology, from the printed catalogue to the Internet to the latest version of Claude Code, continues to evolve to remove those obstacles that stand between us and the things we desire, the more I wonder if we aren't inadvertently (and irrevocably) destroying desire itself... something we thought was unique to us but which may have been an illusion all along.