What Happens When Art Meets AI? | Benedikte Wallace | TEDxArendal
The speaker, who dislikes traditional arithmetic, argues that the synthesis of human intuition and computational logic offers a new creative frontier for art and science. Using artificial intelligence trained on motion capture data, the speaker demonstrates the model's capacity for subjective, non-linear creativity when guided by human partnership. The central claim is that the synergy between human input and AI processing yields results that surpass the capabilities of either component alone.
## Speakers & Context
- Unnamed speaker, currently employed at the faculty of mathematics and natural sciences.
- Speaker admits to disliking arithmetic, though they are speaking about AI.
- The speaker recounts initially feeling trapped between perceived oppositions: art and science.
- The speaker posits that the collaboration between self and computer is *"truly empowering."*
## Theses & Positions
- Thinking of art and science as inherently separate opposites is a false dichotomy.
- The combination of human capability and computer processing power (speaker + computer) exceeds the sum of the individual parts.
- True creative potential resides in the collaboration between human intuition and computational mechanisms, which the speaker calls the "subjectivity sweet spot."
- AI should function as a "suggestion giver" or collaborator, not as a replacement for human expertise.
## Concepts & Definitions
- **Computational Creativity** — The field of study addressing whether and how machines can generate art.
- **Supervised Deep Learning** — The method used: building an artificial neural network with recurrent connections to predict future states based on observed data.
- **Subjectivity Challenge** — The difficulty in defining a precise, measurable metric for evaluating whether an artifact is "good art."
- **Optimal Care (Medical Context)** — The potential combination of human doctor expertise compared against vast AI datasets (millions of global patients) to improve diagnosis.
## Mechanisms & Processes
- **AI Training for Movement:**
- Shows captured data as giant matrices, where each cell represents a reflective marker's coordinates over time.
- The AI learns patterns from this data, predicting future marker positions.
- **Supervised Deep Learning:** The network builds recurrent connections to maintain "short-term memory" for predictions.
- **Predictive Task:** The AI is asked to predict the position of markers in the next frame, measured by comparing its prediction to the actual movement (a task with only one correct answer, encouraging imitation).
- **Subjective Guidance:** Encouraging the AI to do things differently moves it away from imitation toward chaos or creativity.
- **Collaboration Model:** Suggests AI should function as a tool to explore a repertoire (like using AI dancers for a human dancer) rather than executing the entire creative process.
## Named Entities
- **Faculty of Mathematics and Natural Sciences** — Speaker's place of employment.
## Tools, Tech & Products
- **Motion Capture Suits:** Used to capture data of dancers improvising.
- **Reflective Markers:** Attached to the motion capture suits, creating quantifiable points in space.
- **Artificial Neural Network:** The core computational model built for the task.
- **AI Dancers:** The output of the training process, demonstrated in various stages (wiggly, shapeshifter, realistic).
- **Advanced Spell Checkers:** Cited as an existing example of helpful AI technology.
## References Cited
- None.
## Trade-offs & Alternatives
- **Predictive Task (Imitation):** Setting a single correct answer encourages imitation, leading to potential mediocrity.
- **Chaos vs. Mediocrity:** The challenge lies in defining the "sweet spot" between these two poles of AI output.
- **AI vs. Human Expertise (Medical):** While AI can compare symptoms to millions of global patients instantly, the speaker suggests the doctor working *with* AI offers the optimal care.
## Counterarguments & Caveats
- The speaker notes that the initial, "wiggly guy" output was scientifically unsurprising, as it only processed numbers.
- The speaker acknowledges that the AI's output is not always "very human-like," and the process is not fully understood ("we don't even agree on what is good art made by humans").
## Methodology
- Utilizing motion capture data to create numerical matrices of human movement.
- Implementing supervised deep learning with recurrent connections.
- Systematically manipulating the task constraints—from single-answer prediction (imitation) to open-ended generation (potential chaos).
## Conclusions & Recommendations
- The solution to the subjectivity challenge is collaboration; the AI should be taught to work *with* the human, not replace them.
- The speaker concludes that together, "one plus one is equal to three."
## Implications & Consequences
- AI advancements will impact fields like medical diagnosis, offering rapid, global comparison of patient symptoms.
- AI tools can democratize art by providing advanced suggestion generation, allowing human artists to explore new movement repertoires.
- The fundamental synergy between human intuition and computational power will define future creative and problem-solving processes.
## Verbatim Moments
- *"i hate math"*
- *"more often than not we tend to think of the world of art and the world of science as inherently separate opposites"*
- *"i found that me plus my computer friend was greater than me alone"*
- *"can i teach my computer friend to be creative"*
- *"all it sees are these giant matrices and here each cell contains a number which represents one of these reflective markers"*
- *"let's have another look at this task that i've given my computer friend"*
- *"there is only one correct answer what the dancer did everything else is wrong"*
- *"how do i define this subjective sweet spot between chaos and mediocrity"*
- *"the ai and the doctor together has the potential of giving the optimal care"*
- *"one plus one is equal to three"*