Following the first essay, consider creation a little further.
Why can the idea that AI participates in creation provoke resistance?
Perhaps because creativity is more than an ability in our culture. It is also an identity, a dignity, something that distinguishes humans from other beings. Machines can be faster, more precise, and less easily tired. Creativity still feels like the last fortress.
If machines can create too, what is special about being human?
Chapter 1: Similar words, different processes↗
Research on human creativity has moved beyond looking for an isolated creativity module. For example, Beaty and colleagues’ 2015 fMRI study found a distributed network during a divergent-thinking task, with coupling between default and executive networks changing across stages.
Memory, association, combination, evaluation: these words also appear in descriptions of AI. Familiar language does not establish identical mechanisms. Assigning the hippocampus, default network, and prefrontal cortex the fixed roles of store, associator, and evaluator oversimplifies their functions. It also leaves their proposed correspondence with model components untested.
Carbon-based neurons, silicon chips. Whether different substrates can support some similar functions is a question for further research. Saying that the substrate is the only difference is a stronger claim. A neuroscientific account of creativity does not establish it, nor does it settle metaphysical questions about souls.
Someone might still object: humans have consciousness, emotion, and intention; AI merely manipulates symbols without understanding what it does.
These are serious philosophical and scientific questions. I do not want to pretend they have simple answers. But consider some narrower ones:
- If AI-generated music moves you to tears, is the feeling real?
- If an AI-generated painting astonishes you, is the astonishment real?
- If a mathematical proposition suggested with AI assistance receives a valid proof, does its origin invalidate that proof?
The recipient’s response does not become false because of how a work was generated. Nor does the response establish that the generating system has the same feeling.
A work’s effect on a person and a system’s own experience are different questions.
Does this count as creation? The issue already extends beyond producing something new.
Chapter 2: That particular afternoon↗
One answer worth taking seriously is that human value remains connected to human experience.
A beautiful song can describe crying late at night, heartbreak, or the despair of feeling abandoned. Describing those events does not establish that the model generating the song lived through them.
A painting can show a particular afternoon, a particular light through a window, the recollection of a particular person. The image’s success and the memory’s belonging to its generator are still different matters.
A passage can be moving. Its effect, the training that made it possible, and any accompanying subjective experience each require an account. Statistical regularity does not make its effect unreal. Writing that resembles a memory is not evidence of having lived it.
Part of a human artist’s distinctiveness may lie in the connection between work and life. This does not require declaring that AI could never experience anything. Research on AI consciousness attempts to derive independent indicators from theories of consciousness. Artistic performance, or the sentence “I feel sad,” cannot substitute for such an assessment.
AI-assisted work can also carry human experience. Selection, prompting, revision, and use become part of its history. The question is whose experience it carries, which contributions come from the model, and which come from the person working with it.
Contributing to valuable new works and experiencing creation as a human does are not interchangeable claims. The scientific, mathematical, artistic, and musical examples in the first essay support particular judgments about capability. They do not collectively settle the second question.
Human and AI-assisted creation may have different textures.
Chapter 3: Back to the library↗
The library containing every possible book also contains an unimaginable amount of nonsense. A meaningful work does not reach a reader simply because its characters can be arranged.
Human writers search for something worth writing. Generative models produce outputs from learned distributions under supplied conditions. Both can be described as selecting among possibilities, but that description does not explain how each assesses meaning. A language model’s training objective is not an objective that identifies every truth and every beauty.
Can the process produce something new? Yes. But newness alone does not make a good work.
Margaret Boden’s account of creativity brings novelty, surprise, and value together. Under such output-oriented criteria, some AI-assisted achievements can meet some requirements. Judgment still needs a reference point: new relative to what, assessed by whom, and valuable for what reason?
If creation must come from nothing, existing alphabets, experiences, and traditions make the criterion difficult to use. Beginning with existing material does not restrict creativity to simple recombination, either. Boden distinguishes combinations, exploration, and transformations of conceptual spaces. Finding a route on a map and changing how the map is drawn are different achievements.
Perhaps creativity need not be a magic reserved for humans. It can become a set of processes to explain: combining, exploring, selecting, connecting. Those words do not erase differences in implementation or in the circumstances of a life.
Chapter 4: What remains human?↗
What humans have, I think, is being alive.
We breathe, hurt, love, and lose. We remember that particular afternoon, that light, that person.
AI can generate a song about heartbreak without the song establishing that its generator has been heartbroken. It can write about death without facing death in the human way. This distinction concerns the history of experience. It neither equates shutting down a machine with a human death nor declares experience impossible for every conceivable AI.
Does the difference matter?
I am not sure. It may matter to some listeners and not to others. Its importance may change over time.
Perhaps one day the maker will matter less than the work. Perhaps that day will never come.
I do not know.
But the question is already about more than a work’s capabilities. It concerns existence, provenance, and relationships. Its answers need not coincide.
Chapter 5: The wave and the poet↗
I want to end with an image.
Imagine standing on a beach, facing the sea. Waves keep arriving, each with small differences.
In a broad sense, the sea can be said to create new waves. That describes the appearance of new forms without attributing an artistic purpose to the ocean. Now imagine a poet on the same beach. She watches the waves and writes a poem.
Which is more creative, the poem or the wave? Perhaps that comparison leaves out what matters most: why the poem was written and how it enters another person’s life.
Perhaps creativity is more than an attribute. It is also a relationship between maker and viewer, work and mind.
The wave does not appear for a particular reader. A poem can be written for someone. What about AI-generated works?
They occupy a strange position among familiar categories. They are not undesigned natural processes: training, prompting, selection, and distribution involve human arrangements. Those arrangements do not establish that the generating system knows what its work means to someone.
Perhaps this is part of what unsettles us: a familiar effect begins to come from an unfamiliar history.
Humans and AI may take different directions in creation, or contribute to a work together. A life with a heartbeat enters a computable form; a computed form enters a moment in someone’s life.
The results may have different temperatures. What is cold novelty, and what is warm novelty? What is beauty calculated, and what is beauty made from living? These are not labels that can simply be read from a file, nor need they be mutually exclusive categories.
Perhaps the question is not only whether AI can create, but what kinds of creation we want.
The responsibility for answering it still belongs to people.