Brilliance inside a bounded library can still feel like the whole world

I wake up with music playing. Before I’ve properly opened my eyes, a song is often already playing somewhere in the background of my mind.

I used to think it was random. But I’ve started paying attention. The song often seems connected to something I’ve been thinking about. Or worrying about. I can't prove the connection, but I notice it often enough to wonder.

There’s no obvious pattern to what turns up. Acoustic. Dance. Country. Something I heard yesterday. Something I haven’t consciously thought about for years.

My mind can surprise me.

But only up to a point.

It has never played me a song I don’t know. Of course it hasn’t. Yet that obvious fact has been bothering me.

The music we don’t choose

It turns out there is a name for music that appears in our heads without us consciously summoning it. Involuntary musical imagery, or INMI, is what most of us would probably call it an earworm.

Researchers describe INMI as the conscious experience of music without deliberately trying to start or sustain it. And our memory for that music can be surprisingly precise.

In one real-world study, people tapped along to the music spontaneously playing in their heads. For songs with a definitive recorded version, the median difference between the imagined and actual tempo was less than 8 per cent. Other studies have found similarly accurate recall, including of pitch.

That fascinates me.

Something I encountered years ago can apparently sit somewhere beyond my conscious attention and then reappear, uninvited, one morning.

I don’t choose it.

I don’t search for it.

Somehow, my mind finds it.

But it still has to be there to be found.

We don’t think from nothing

I feel like there is a much bigger idea here. We like to imagine thinking as something we do to the world in front of us. Observe, analyse, decide. But cognition doesn’t begin with a blank sheet of paper.

Research into perception shows that prior knowledge and expectations influence how we process what is in front of us. Predictive processing goes further, proposing that the brain uses internal models built from experience to anticipate and interpret incoming information.

The framework remains debated. But the less controversial point is enough for me. What we have previously encountered influences how we make sense of what we encounter next.

Usually, this is incredibly useful.

Experience helps us recognise patterns. Expertise helps us notice things a novice might miss. Memory gives us something against which to compare the new thing in front of us.

The library gets bigger. And we get better at searching it. But those are not quite the same thing. Because you can become extraordinarily good at making connections between the things you know while remaining completely unaware of what never entered the library.

The question nobody asks

This becomes even more interesting when I think about AI. One of its great promises is its ability to help us think. We can ask it to challenge an argument. Find contradictions. Offer another perspective. Identify assumptions. Ask questions we haven’t thought to ask.

And it can search, process and connect information at a scale no individual human can. That feels like an extraordinary expansion of the library.

But it still has a boundary.

An AI system has access to some information and not other information. It learns from data and, depending on how it is built and used, may also retrieve information from additional sources.

But retrieval still requires something to retrieve.

Available information represents some experiences and perspectives better than others. Some things have never been recorded. Some sources are inaccessible. And the questions we ask are themselves shaped by what we think might matter.

NIST’s guidance for generative AI explicitly recommends examining the accuracy, representativeness, relevance and suitability of the data used throughout an AI system’s lifecycle.

The library matters.

Which raises a possibility I find uncomfortable.

AI might brilliantly challenge our thinking while still sharing a blind spot with us. Something simply isn’t in the library. And neither of us knows it is missing.

When good answers become dangerous

The better these tools become, the easier it may be to confuse the quality of an answer with the completeness of the search.

A thoughtful, beautifully reasoned challenge feels comprehensive. But fluency is not completeness. Neither, for that matter, is intelligence.

We have the same problem ourselves. Put five very smart people around a table and they may challenge one another brilliantly. But if they have read the same things, worked in the same places, learned from the same people and built broadly similar models of how the world works, there may still be a question nobody asks. Not because they aren’t clever enough. Because the question isn’t in the room.

Perhaps the job isn’t simply to ask what is wrong with this conclusion? It is also to ask what might be missing from the world in which we reached it? Whose perspective isn’t represented? Which sources haven’t we searched? What evidence would change our minds? What assumptions are so embedded that nobody has thought to name them?

And then there is the question I don’t know how to answer. What wouldn’t we know to ask?

The songs we cannot hear

Tomorrow morning, I will probably wake up with another song playing in my head. I still won’t know exactly why that particular one appeared. I like that.

There is something wonderful about a mind capable of reaching backwards, retrieving something forgotten and making a connection I hadn’t consciously made.

But I am becoming more interested in the boundary around it. Every song my mind plays tells me something about what has entered my world. None can tell me what hasn’t.

We spend a lot of time trying to become better thinkers. And now we are building machines to help us think better still. But perhaps the harder task is noticing the boundary around what both of us know.

Which leaves me wondering about everything outside it.

How do you go looking for a song you’ve never heard?