A technical response to the debate over human art, generative models, and the “spark of humanity”
Pope Leo XIV recently argued that, in the age of AI, we need to distinguish human art from what machines produce. His argument points to an important issue: algorithms do not possess the human “spark” that gives art its meaning.
I agree with the underlying concern.
But I think the technical problem is more interesting than “humans create, machines calculate.”
That distinction is too simple for modern generative systems.
The real question is:
What happens to human agency when a computational system becomes an active participant in the generative process?
That is where the discussion should move.
- Generative AI is not simply a copying machine
A common description of generative AI is:
“It calculates statistical patterns from millions of examples and produces something similar.”
Technically, this is incomplete.
A diffusion model, for example, does not retrieve an existing training image and paste its pixels into the output. During training, the model learns a parameterized representation of statistical relationships in the data.
At inference time, the system operates on a new state and iteratively transforms noise toward an output conditioned by the prompt and other inputs.
Conceptually:
Training data
↓
Representation learning
↓
Parameterized model
↓
Conditioning
↓
Sampling / generation
↓
Novel output
The generated image may never have existed in the training corpus.
This does not mean the model is conscious.
It means something more precise:
computational novelty does not require subjective experience.
That distinction matters.
A system can generate novelty without possessing an inner experience of creating that novelty.
- Novelty is not intention
This is where the philosophical and technical questions intersect.
Consider four different properties:
Property Generative AI Human
Produces novel configurations ✓ ✓
Learns statistical structure ✓ ✓
Has subjective experience Unknown / unsupported ✓
Possesses intrinsic artistic intention Not established ✓
The mistake is to collapse all four into one concept called creativity.
A model can exhibit highly sophisticated generative behavior without us having evidence that it has:
subjective experience,
intrinsic goals,
personal meaning,
aesthetic desire,
existential concern,
or responsibility for its output.
Therefore:
generation ≠ intention
and
novelty ≠ meaning.
This distinction should become foundational in discussions about AI-generated art.
- But AI is not merely a digital paintbrush either
There is another problem.
Calling AI “just a tool” is also technically inadequate.
A paintbrush does not propose a composition.
A camera does not generate ten alternative visual concepts.
A synthesizer can transform sound, but a modern generative model can operate over an enormous learned space of possible outputs and actively influence which direction the creative process takes.
A simplified human-AI loop looks more like this:
Human intention
↓
Prompt / constraints
↓
Generative model
↓
Candidate outputs
↓
Human evaluation
↓
Selection / modification
↓
New intention
↓
Generative model
↺
Notice what has happened.
The model is no longer simply executing a deterministic command.
It participates in an iterative search process.
The human says:
“Not this. Try something darker.”
The model generates alternatives.
The human selects one.
The selected output changes the human's next decision.
The model responds again.
The creative process becomes a coupled system.
- The emerging unit of creativity may be the interaction
This suggests a more interesting model.
Instead of:
Human → Tool → Artifact
we may need:
Human ↔ Generative System
↓
Artifact
The artifact is produced by an interaction between:
human intention,
learned representations,
model sampling,
environmental constraints,
iterative feedback,
human selection,
and post-generation modification.
This doesn't make the AI an artist.
It makes the creative system different.
That distinction is important.
We don't necessarily need to decide whether the model itself is a “creator.”
We need to understand what kind of agency emerges from the human–model system.
- This creates an attribution problem
Once generative systems become active participants, a traditional authorship model becomes unstable.
Imagine three cases.
Case A — Pure generation
A person writes:
“Create a Renaissance-style portrait.”
They select the first output.
Human contribution:
low
Model contribution:
high
Case B — Iterative direction
The person generates hundreds of candidates, rejects most of them, changes composition, lighting and symbolism, combines outputs and performs extensive editing.
Human contribution:
high
Model contribution:
high
Case C — AI-assisted execution
The human develops the concept, composition, narrative and visual structure, then uses AI to execute technically difficult portions.
Human contribution:
very high
Model contribution:
instrumental
All three may produce visually impressive work.
But treating them as identical forms of authorship makes little sense.
This suggests that AI-era authorship should become process-aware rather than artifact-only.
- We need provenance for creative processes
This is where technology can provide something better than philosophical arguments.
Instead of simply labeling an image:
AI GENERATED
we could record a richer provenance graph:
{
"human_intent": "original",
"model_assistance": "generative",
"human_selection": true,
"human_editing": true,
"iterations": 37,
"external_assets": 2,
"final_human_approval": true
}
Obviously, real systems would require much more sophisticated schemas.
But the conceptual shift is important.
The question is no longer:
Was AI used?
It becomes:
How was AI used?
That is a much more useful technical question.
- Human agency should become a measurable design property
This idea can extend beyond art.
Consider AI systems that generate:
software,
scientific hypotheses,
legal arguments,
business strategies,
architectural designs,
music,
political content,
medical recommendations.
In each case, there is a spectrum:
Human-controlled
↓
Human-directed
↓
Human-AI collaborative
↓
AI-assisted decision
↓
AI-delegated decision
↓
Autonomous system
The critical variable is not simply intelligence.
It is agency allocation.
Who:
initiates?
chooses?
evaluates?
rejects?
accepts?
explains?
takes responsibility?
This may become one of the most important questions in AI system design.
- The real risk is not that AI becomes an artist
I think this is where the debate becomes much deeper.
The most important risk isn't necessarily:
“Machines will make art.”
Machines already generate images, music, code and text.
The more consequential possibility is:
Humans gradually stop exercising the cognitive functions that make their actions meaningfully theirs.
If an AI chooses the idea, writes the argument, selects the evidence, creates the image, evaluates the result and makes the final decision, then the human may remain technically “in the loop” while becoming intellectually irrelevant.
That is a very different problem from AI replacing a profession.
It is a problem of agency erosion.
- Human-in-the-loop is not enough
AI safety discussions frequently use the phrase:
Human in the loop.
But the existence of a human somewhere in the pipeline doesn't guarantee meaningful human agency.
A human who merely clicks:
Approve
after an AI system has already generated and evaluated the options is technically “in the loop.”
But functionally?
The system may already control most of the decision space.
We therefore need a stronger concept:
Meaningful Human Agency
A system should preserve meaningful opportunities for humans to:
establish goals,
inspect alternatives,
challenge recommendations,
introduce new constraints,
reject system outputs,
understand relevant provenance,
and assume responsibility for the final decision.
This is much more demanding than simply putting a human somewhere inside an architecture diagram.
- The theological question becomes an engineering question
This is why I find the debate surrounding Pope Leo XIV particularly interesting.
The statement that algorithms lack the “spark of humanity” can be interpreted as a theological claim.
But it also leads directly into an engineering problem:
How do we design systems that amplify human agency rather than quietly replacing it?
That could mean designing AI systems around:
provenance,
explainable interaction histories,
controllable autonomy,
explicit human approval,
reversible actions,
uncertainty disclosure,
attribution,
and responsibility tracking.
In other words:
the future of human-centered AI may depend less on making machines appear human and more on making human agency computationally visible.
- Perhaps we are asking the wrong question
The debate usually asks:
Can AI create?
I think a more useful sequence is:
Can AI generate novelty?
Yes.
Can AI participate in creative processes?
Absolutely.
Does generation imply consciousness?
No.
Does creativity require subjective experience?
That remains a philosophical question.
Can humans create meaningful art with generative systems?
Clearly.
Can excessive automation weaken human agency?
Potentially—and this may be the most important question.
The deeper frontier
The boundary we need to protect may not be:
human vs. machine.
It may be:
agency vs. automation.
intention vs. delegation.
meaning vs. generation.
responsibility vs. optimization.
AI does not necessarily threaten humanity because it can generate.
It becomes threatening when humans stop asking why they are generating, what they are choosing, and who is responsible for the result.
Perhaps the goal should not be to keep machines out of creativity.
Perhaps the goal is to ensure that, even when machines become extraordinary generators, humans remain extraordinary authors of intention.
That is a much harder engineering problem.
And probably a much more important one.
What do you think?
If a generative model proposes, generates, evaluates, and iterates—but a human establishes the goal and accepts the final result—
where exactly does authorship begin and end?
Created by Seyed Alireza Alhosseini Almodarresieh
Top comments (0)