Human-AI Co-Creativity: Advances, Opportunities, and Challenges

1 Introduction

• Realizing generative AI’s opportunities and addressing its challenges calls for a community spanning technical advances, system design and deployment, and firsthand creative practice.The workshop aimed to bring together researchers, industry professionals, and practitioners with these complementary forms of expertise.Source: §1 · p. 1
• The first thrust explored how generative AI can support open-ended creative tasks, including brainstorming ideas and creating content.The workshop brought together researchers and practitioners from creative fields alongside technical and industry communities.Source: §1 · pp. 1–2
• The second thrust identified challenges in integrating generative AI into creative workflows, including design fixation, idea homogeneity, copyright, and authorship.The workshop sought to foster collaboration among stakeholders to facilitate next-generation technologies that enhance human-AI co-creativity.Source: §1 · p. 2

2 Overview of the Workshop Activities

• The workshop bridged machine learning and real-world creative domains, with its program and venue supporting discussion and networking.The program included a panel and poster session.Source: §2 · p. 2

2.1 Topics of Interest

• The first topic thrust covered creative uses of generative AI, model creativity and diversity, creativity-focused methods and architectures, and evaluation benchmarks.Topics included creative work by models acting alone and in co-creation with a human user.Source: §2.1 · p. 2
• The second thrust covered workflow concerns, including creative professions, design fixation, idea homogeneity, content authenticity, and attribution of human creativity.These topics guided invited-talk selection and contributed-paper acceptance.Source: §2.1 · p. 2

2.2 Accepted Papers and Poster Session

• The workshop received 80 submissions and accepted 49 papers; accepted authors presented posters, with lightning talks offered to broaden dissemination.Submissions received at least three reviews from an 88-member program committee; Figure 1 visualizes accepted-paper titles as a word cloud.Source: §2.2 · pp. 2–3, Figure 1 · p. 3

2.3 Invited Talks and Panel Session

• Six invited talks addressed topics ranging from AI-content legitimacy and originality in art to design fixation and ways to diversify language-model outputs.The talks also discussed industrial-design orchestration and intentional friction, along with post-training and user simulation for exploratory tasks.Source: §2.3 · p. 3
• Other talks connected creativity with learning and cognitive thought partnership, and discussed creativity in NLP, analogical reasoning, and evaluating language-model outputs.The speakers also took part in a 60-minute panel session.Source: §2.3 · p. 4

3.1 Creativity Metrics

• Creativity metrics remain difficult to validate: evaluations are subjective, human annotations can disagree, and human-designed tests are not yet validated for generative AI.The section points to robust unified ratings aligned with domain experts and metrics calibrated to human creativity as research directions.Source: §3.1 · pp. 4–5

3.2 Benchmarks for Creative Tasks

• Creative-task benchmarks cover multiple domains, but tend to measure lower-order creativity and evaluate models acting alone, leaving human-AI co-creation uncertain.The section calls for benchmarks targeting transformational creativity and human-AI co-creation.Source: §3.2 · p. 6

3.3 Improving Creativity of Generative Models

• Generative models can produce homogenized outputs, and post-training can reinforce this while improving task accuracy at the cost of diversity.Research directions include maintaining output quality while improving creativity and diversity, and designing methods that support human-AI co-creation.Source: §3.3 · pp. 6–7

3.4 Authenticity and Authorship of Creative Artifacts

• Authenticity and authorship challenges include copyright, attribution, and content legitimization; research directions include internet-scale detection and fine-grained provenance over time.Provenance is relevant within a co-creation session and across sessions as content is generated or remixed.Source: §3.4 · p. 7

3.5 Human-AI Co-Creation for Creative Tasks

• AI suggestions can narrow users’ thinking; research directions include measuring fixation and over-reliance across domains, interventions, and balancing individual with population-level creativity.The section also points to optimizing for agency, transparency, and learning that transfers across co-creation sessions.Source: §3.5 · pp. 7–8

• Generative AI offers opportunities to support people in open-ended domains, while integrating agents into creative workflows requires new safeguards and technical innovations.Source: §4 · p. 8
• The workshop’s talks, papers, and discussions reflected community excitement, and its diverse perspectives highlighted the importance of drawing on ideas from multiple disciplines.Source: §4 · p. 8
• Continued collaboration among communities and stakeholders is needed in human-AI co-creativity.Source: §4 · p. 8