This post was originally published by the Institute of Art and Ideas and is republished here with permission as part of the Blog of APA’s partnership with the Institute.
Many claim informational systems cannot be truly creative in the way human artists are because they lack human emotion and originality. But philosopher and cognitive scientist, Hanne De Jaegher, argues the issue is deeper still: AI is not alive. It has no will to live and to ensure its survival. Unlike humans, AI doesn’t desire, need, relate, or make sense of the world and what matters in it. True creativity arises from things mattering; even when—especially when—those things are contradictory, nonsensical and ambiguous.
Dutch musician Eric de Jong (Spinvis) is on the Belgian classical radio, giving an in-depth interview about his lifelong love of music and music-making, in one of those programs centered around favorites from the interviewee’s record collection. Towards the end of the two-hour program, de Jong announces a piece of music without much introduction. We listen. It’s a Baroque song. Afterwards, he surprises the audience and the host by saying the piece was made by AI. More surprising to me is the explanation he gives for why he likes it. He says this music is generic, and explains how the AI system does not know anything, but has access to all the music ever made. “You ask it to create a piece of Baroque, and it generates a piece that is a kind of average of all the Baroque music, which it takes as input.”
de Jong says he finds this “very good news”. He considers AI a kind of mirror that tells us how generic music is, including the music made by humans. He says, “Much is made of ‘the soul of the composer’. But we are just machines. We make on the basis of what we see and hear. And so, we also make generic things. We are much more generic than we think.” Earlier on, he had talked about how, when recording a piece of music in a studio, it is like making a snapshot of a set of circumstances, and it will be a different snapshot when you record it another time. All kinds of accidents happen, like the sound of a dog barking outside, or a microphone dropping. The real world comes in and shines through, and this makes each recording unique.
The interview had started with de Jong talking about his own way into music. He talked about how, when he was very young, he heard Bach’s Badinerie on the radio, played by French trumpeter Maurice André. de Jong had been too young to know how to write down these names, but had tried his best, and when he told his mother about it and showed her the names, she took him out to buy the record that same day. His mother, he said, seemed to have understood that this had been important. This was his initiation into music, and he said he still gets goosebumps when he hears this piece.
In their long conversation, host and interviewee talk about many of these kinds of specificities of de Jong’s encounters with particular pieces of music, people, instruments, happenstances, and how he considers himself, as an artist, as a medium. He serves the music, and what the world, others, and his circumstances give him, and that’s how he produces music. I am surprised by the contrast between how he speaks about what Runway, the AI system he used to generate the Baroque piece we heard, means about human music-making and how he speaks about his own life and journey in music. He adds something else about the AI. It’s “like ‘the end of history,’” he says, and “we now know that human artists are also generic, machine-like generators.”
De Jong says this gives him hope, because after learning this, we can go back to making mistakes. To imperfection. Like a mother who sings to her baby, without worrying about being able to sing. The contrast in his way of speaking about AI-generated music and his own way of making music lies, I think, in the creativity that underlies both. Or rather, underlies one and not the other. It is obvious that knowing and creativity are much more like how he describes his life, his encounters, the accidents and unexpected connections in music making, than what he says about AI. I find it hard to accept his enthusiasm for AI-generated music. But I sense there is a playfulness in it. In a way, maybe his thoughts about AI are themselves a kind of creative, hopeful dealing with some lesson he thinks informational systems seem to be teaching us. Overall, the lesson is perhaps not that it is good, or even true, that AI shows us that how we humans generate music is generic. I think it is de Jong’s finding his way back to making mistakes, to imperfection, that is important here.
I recently happened upon a video of David Bowie talking about his admiration for The Pixies. In that interview, he talks about imagination, which he describes as not being about fantasy, but about “being able to understand the affinities of something and have those affinities illuminate the subject.” What goes into illuminating a subject?
Let’s think back to de Jong’s illustration of a mother or primary caregiver singing to their baby. What a mother does when she sings to her baby, and what the baby does when they move with the mother’s singing rhythm is, of course, relating. It is not about making mistakes or not, it is about something else entirely. “Mistakes” don’t even matter to what this is really all about. What is at stake, what mother and baby care about and enact in these interactions, is connecting, and they do so, even at a very young age, in demonstrably sophisticated ways. You might say they are working on affinities, illuminating a subject. Affinity, according to the dictionary, is firstly “A natural liking for somebody or an inclination to think and feel alike: this mysterious affinity between us; an attraction to or a taste for something.”
Affinity has to do with nearness. It also has something to do with understanding, with a certain kind of knowing. Some further definitions of affinity are “relationship by marriage; resemblance based on relationship or causal connection; a chemical attraction between substances, causing them to combine; a relation between biological groups with similar structure and common ancestry.” It comes from Latin, ad + finis, bordering on, near the edge. We understand things that have an affinity to belong together in particular ways. We understand things that are near to each other or that we are near to. This understanding is not always direct. It is a different nearness than that between the training data and the outputs of an AI system. AI or informational systems place things next to each other (they output strings of tokens) that are statistically predicted to be likely to occur in sequence.
But the nearness humans and other animals experience and look out for is much more complex. It is variegated, unpredictable, often ambiguous, sometimes contradictory, paradoxical, surprising. None of these things you will find in an AI model, though you do sometimes find it in interactions with AI. And this is where one of the cruxes lies. It is in interactions with virtual systems that we can experience human meaning, maybe even illumination. Like in de Jong’s hopeful return to human connection, beyond re-allowing mistakes—to the cracks through which the light gets in. Affinity is, even in the dictionary, referred to as something mysterious. While there is something mysterious about affinity, and it is important to maintain that, for reasons of, well, mystery, I don’t think it is just mysterious, and understanding it better is important for distinguishing and being careful about what AI systems may or may not only show us, but do with us.
In the functioning of AI models, there is prediction, averaging, and generic-ness. What it averages over and what it predicts depend on what goes into it. No novelty is created in an AI system by listening to the richness of the real world, let alone listening to something that matters. For humans (and other living beings), meaning arises as particular and concrete things matter to them. Living beings actively contribute to, even participate in, this mattering and meaning. This is because living beings must continuously build themselves out of the world. Things matter, most fundamentally, to living beings, out of a concern with staying alive. The most primary way in which this happens is through metabolism itself. Living beings need sustenance, and to avoid what is detrimental or poisonous to us. Things are immediately, bodily, pertinent for our survival, and this makes certain things relevant, and others irrelevant. Things literally matter to us, and we constantly need and seek out this mattering.
This happens not just at the metabolic level. It happens also when I try to build a new habit. Like when I’d like to go on forest walks more often. Certain things in the world will now stand out as relevant that I hadn’t noticed before. I will need to go look for hiking boots, say, and bear spray, as I live in the Rocky Mountains. I will have to dwindle other habits, such as too much time on devices. I will restructure and rearrange how I spend my time, what I attend to, and who I do things with. I may call a friend and see if she wants to go with me. I will be arranging my affinities differently, and there is happenstance, context, surprise, and complexity in all of this.
And when an artist speaks on the radio about their love of music, they will speak in certain ways, and not in others. They will show us surprising, hopeful things about AI-generated music, and the meaning of what they say will go beyond what they have literally said. Because humans like illumination. We like to live and learn, to love and know—human knowing, which is living, loving knowing, and is engaged and engaging. We are inevitably changed by what and how we know. And we can and do help each other in this as we interact with each other, as we sing well or badly together, and as we illuminate imagination for each other. In short, as we participate in each other’s sense-making.
The post Information and Data Will Never Deliver Creativity first appeared on Blog of the APA.
Read the full article which is published on APA Online (external link)
