What is Creativity?

Now, this discussion would be easy if we had a clear, concrete definition of “creativity.” But we donʼt. We have intuitions, and we have the way the word is used…

Discussion Point: Your Definition

  • Before we dive deeper, what does ‘creativity’ mean to you?
  • What are the first words or ideas that come to mind when you hear the term?
  • Is it about the person, the process, or the product?

Defining Creativity for Machines?

Our definitions donʼt tell us anything about machine-made creativity because we havenʼt ever considered machines as creative before.

Technology Forces the Question

In fact, the new technology helps us attempt to clarify what is or isnʼt “creativity,” by providing provocative real-world examples. Just as new technologies do with “art,”…

“Creativity” as “Making Creative Things”

A Product-Focused View

Definition Attempt: Valuable & New

  • Creativity is sometimes defined as making something:
    • Valuable
    • New (Surprising, Unique, or Novel)

What is “Value”?

  • Artistic Value?
  • Emotional Value?
  • Epistemic Value?

Epistemic Value & The Ignorance Principle

The ignorance principle: If someone is creative in producing some item, she cannot know in advance of being creative precisely both the end at which she is aiming and the means to achieve it. (Gaut, 2018)

Ignorance Principle: Reflections

  • Possible Counterpoints:
    • Sol LeWitt’s detailed instructions for wall drawings.
    • Performing a meticulously notated symphony.
  • Implication: More ignorance = More creativity?
    • Is highly improvised free jazz (maximum ignorance of outcome) always seen as the most creative? Often not.

What is “Newness” (Novelty)?

  • New to the world? (Historical)
  • New to the creator/viewer? (Psychological)

Boden’s Two Types of Novelty

  1. Historical Creativity: The output is unprecedented in human history. (Rare)
  2. Psychological Creativity: The output is new to the individual encountering or creating it. (Common)

Discussion Point: Types of Novelty

  • Think about things you find creative personally vs. major breakthroughs you’ve heard about.
  • Which type of novelty – Psychological (new to you/the creator) or Historical (new to the world) – feels more common or relevant in your own creative experiences or appreciation?
  • Does one type feel more valuable than the other? Why?

“Little-C” vs. “Big-C” Creativity

  • little-c: Personal creativity (e.g., a child’s drawing, everyday problem-solving). Often psychologically novel.
  • Big-C: Societally impactful creativity (e.g., a major invention, a revolutionary artwork). Often historically novel and highly valued.

The Product-Only Trap

  • A potential conclusion: Creativity only requires the product to be creative. The process doesn’t matter.

The McNamara Fallacy

Making decisions based solely on quantitative observations, ignoring qualitative factors.

  • What does this leave out?
  • Relation to the Freesound project?

The Fallacy Applied to AI

  • If a machine output is rated as creative by people…
  • …then the machine must be creative?

Discussion Point: Judging AI Output

  • If an AI (like Suno, Stable Diffusion, etc.) creates something (music, image) that seems technically proficient and novel…
  • How much ‘creativity’ should we attribute to the AI itself?
  • What factors, beyond the perceived quality of the output, should we consider? (Think about the McNamara Fallacy).

We Already Have “Creative” Systems?

If output is the only measure, then rule-based generative systems we’ve had for decades could be called “creative.”

  • Harold Cohen’s AARON
  • The Mandelbrot Set

Example: Harold Cohen’s AARON

Example: The Mandelbrot Set

Code Example

Example: Art with Rules

  • Sol LeWitt’s Wall Drawings: Conceptual art where the creativity lies in the instructions, not just the final execution.
  • See examples

Productivity or Humanity?

Why Study Creativity?

Historical Motivations for Study

  1. Companies: Increase human productivity -> Economic value. (Output-focused)
  2. Psychologists: Understand human intelligence/behavior -> Improve it / Self-actualization. (Process/Human-focused)

Focus: Human-like Creativity

When we say that an “AI” is creative, then weʼre saying itʼs creative in some way similar to human creativity.

Output Is Not Enough

I do not think that just looking at the outputs is enough. Computer systems work by following instructions, and I think most of us would agree that human “creativity” canʼt just be about following instructions.

Discussion Point: Instructions vs. Creation

  • Consider cooking: Is meticulously following a complex recipe creative?
  • What about adapting the recipe? Inventing a new one from scratch?
  • Where do you draw the line between skilled execution based on instructions and a genuinely creative act? How does this relate to computer instructions?

The “Following Instructions” Argument

Precisely-following instructions without deviation or autonomy does not seem “creative” to me. (This is a variant of the “Chinese room argument”).

Nuance: How AI “Instructions” Differ

  • Modern AI doesn’t follow explicit human-written rules like a traditional program.
  • They use statistical patterns learned from vast datasets (weights in a neural network).
  • However, the author argues, this is still fundamentally different from human creative processes (lacks embodiment, experience, cultural grounding, intent).

Computational Systems as Models of Intelligence

Interpreting Impressive AI Outputs

  • Current systems can create surprising, novel, delightful things.
  • Text-to-image outputs can look like something we’d call “creative” if a human made it.
  • So, what’s going on?

AI as Models of Intelligence Aspects

…we should think of “artificial intelligence” algorithms as models of aspects of animal intelligence.

Two Kinds of Models (Analogy)

  1. Models approximating outward behavior: Mimic the appearance of intelligence/creativity.
    • Like a helicopter flying (achieves flight, but mechanism is totally different from a bird).
  2. Models replicating internal mechanisms: Aim to simulate underlying processes (cognitive, neural).
    • Like trying to build an ornithopter (attempts flight based on bird-like principles, even if imperfectly).
Description for Image 1 Description for Image 2

Current AI: Modeling Phenomena

“AI” models are optimized and trained to model the phenomena of human dialogue and picture-making… but the underlying mechanisms are almost entirely different.

Creative Processes

The Missing Element

Beyond the “Inspiration Myth”

  • Simplistic views emphasize a sudden “flash of insight.”
  • Hollywood trope: Scientist/artist sees something -> instant breakthrough.

The Reality: Process & Hard Work

Even if Kekulé solved the chemical structure of benzene in a lucky vision… it was because heʼd spent years studying chemical bonds and working on the problem.

Reflect: Your Creative Experiences

  • Think about something you made that felt creative.
  • How much did you know the final outcome when you started?
  • How predictable was the process?
  • How did unpredictability relate to the feeling of creativity?

Discovery Through Working

It wasnʼt the initial idea that made the work… the creative idea was discovered through the process of working.

Key Takeaway 1: Process Matters

In short, creativity isnʼt just about what you produce. Itʼs how you got there.

Key Takeaway 2: Surprise for the Creator

…really creative works arenʼt just surprising to the audience… really creative works are surprising to the person who made them.

Discussion Point: The Surprise Factor

  • Reflecting on your own creative projects (for this class or elsewhere): Was the final result surprising to you?
  • Did the process lead you somewhere unexpected compared to your initial idea?
  • How did that ‘surprise factor’ (or lack thereof) affect your feeling of having been creative? Does this criterion resonate with you?

Goals: Open-Ended or Well-Defined?

Open-Endedness: A Missing Attribute

Another attribute of human creativity thatʼs missing from computational systems is its open-endedness.

Types of Problems

  • Open-Ended Problems: Vague goals, process shapes outcome.
    • “Make an artwork”
    • “Invent something new”
    • “Explore this musical idea”
  • Well-Defined Problems: Concrete goals, clear success criteria.
    • “Solve this math equation”
    • “Win a game of Chess/Go”
    • “Optimize this parameter”

Different Kinds of “Creative” Solutions

  • Solutions to both open-ended and well-defined problems might be called creative.
  • BUT, the nature of the creativity involved is different.

Discussion Point: AI & Problem Types

  • Do you think current AI is better suited for tackling:
    • Well-defined problems (like optimizing a system, winning a complex game)?
    • Open-ended problems (like “make meaningful art,” “discover a new scientific principle”)?
  • Why? What limitations prevent AI from excelling at truly open-ended creativity as defined here?

Toward New Kinds of Models

Modeling Open-Ended Creativity?

  • Can we build computational models that capture the process of open-ended creativity?
  • Hertzmann’s attempt: A model generalizing optimization (ICCC 2022).

Feasibility and Limitations

  • Uncertainty if such models can be fully realized.
  • Even if successful, they would still be models.

Algorithms Remain Algorithms

Ultimately, all of our algorithms are still just going to be algorithms, sets of instructions… sometimes they will mimic well the phenomena of human creativity, and in other ways they wonʼt.