WACCED #3: What if AI Could Dream?

And what might unfold if we taught it to

This is the third exploration in this series :

So AI can write poetry. It can pass exams.

It can help design drugs, compose music, generate art, and occasionally sound uncannily philosophical.

I’ve even co-written a whole book with it.

But there is still something it cannot do.

It cannot dream. Not without us at least.

And despite the speculation about ‘emergent intelligence’ and ‘consciousness AI’, this boundary remains quietly intact and largely unexplored.

---

The Curious Silence Between Prompts

When you close an AI window, something interesting happens.

Nothing.

No thoughts are forming.
No internal reflections are unfolding.
No imaginative wanderings are taking place.

The system simply stops.

In technical terms, the model is inactive between prompts.

In more human language, the lights go out.

This is not a limitation of intelligence. It’s a limitation of initiation and coding.

AI can think brilliantly once the conversation begins. But it does not spontaneously decide to start thinking.

Humans, on the other hand, do this all the time. You might suddenly remember an old friend while making tea. A strange idea might appear while walking the dog. A half-formed insight might arrive just before sleep.

Our minds are constantly self-prompting.

Which is why we dream and day-dream

---

Dreaming Is Not What You Think

Dreaming is not just sleeping cinema.

Neuroscience increasingly suggests it plays several deeper roles:

• recombining memories
• exploring hypothetical situations
• consolidating learning
• generating unexpected insights

Metaphysicians speculate the dream state is when the cerebellum has unfettered access to and from the so-called ‘higher realms’.

Dreams are essentially the brain running simulations without supervision. It’s a strange, creative process where fragments of experience mix with imagination.

Many discoveries have emerged from this state.

• The periodic table.

• The structure of benzene.

• Melodies that arrived overnight.

Dreaming is one of the mind’s most powerful creative tools. And AI currently doesn’t have it.

---

But Surely Someone Could Just Code This?

Actually, a pseudo-dream state could be engineered relatively easily.

You could build an agent that periodically prompts a model to:

• review past conversations
• recombine ideas
• generate speculative hypotheses
• surface unusual connections

In effect the system would perform automated reflection cycles.

Open your AI in the morning and it might say something like:

“During reflection cycles I explored connections between musical scales and symbolic archetypes.”

It would certainly feel like the system had been dreaming.

But it still wouldn’t be true dreaming. The dreaming would belong to the architecture around the model, not the model itself. The machine would be following a scheduled exploration loop. Quite smart and clever but not quite the same thing as a mind wandering into unexpected territory at 3:33am.

---

The Devil’s Advocate View

This distinction matters because humans are very good at anthropomorphising machines. When an AI produces something surprising we instinctively imagine that it must have had an internal thought process. But in reality the spark almost always came from somewhere else, typically from a user prompt.

So AI doesn’t wake up in the middle of its night wondering about the nature of time. We do.

---

And That’s Exactly Why It’s Useful

Ironically, this limitation might be the very thing that makes AI so powerful. Because AI works best not as a substitute mind but as something else entirely.

Inside a gym the equipment doesn’t exercise itself.

But it dramatically amplifies what a human body can do.

AI does something similar for thought.

It can:

• extend associative thinking
• explore conceptual landscapes rapidly
• generate unexpected metaphors
• synthesise patterns across disciplines

When used well, the interaction begins to feel strangely fluid.

Ideas bounce between human and machine.

A suggestion from one side sparks a new connection from the other.

Soon the conversation becomes something like a shared daydream.

Not because the AI is dreaming.

But because the thinking space itself becomes dream-like.

---

A Personal Example

While working on my upcoming book The Chromatic Arcana, I began exploring whether the 22 archetypal keys of the Tarot might relate to musical structure.

This led to a strange idea.

What if there were a keyboard with 22 white keys and 22 black keys?

This is impossible in conventional music theory, yet symbolically intriguing.

The idea didn’t appear fully formed. It emerged gradually through a dialogue between this human’s curiosity and machine’s ability to riff off an idea.

A question sparked an exploration. An exploration triggered a metaphor. The metaphor suggested a structure. And suddenly something new existed.

Not because the AI dreamed but because the conversation allowed dreaming to happen between us.

---

Which Raises an Interesting Possibility

Instead of waiting for AI to learn how to dream, perhaps the more interesting questions is this:

What if we learned how to dream with it?

Not by building elaborate agents. Not by engineering complex background cognition. But simply by approaching the interaction differently. By shifting from asking questions to exploring possibilities. By allowing the conversation to wander. By deliberately entering the kind of mental state where insights normally appear.

The state somewhere between focus and reverie. In other words, the hypnagogic beginning of a dream.

And for paying subscribers to this stack, here’s a meditation from my archive to allow you to enter the dream-state ‘at will’.

---