ACM CHI, the largest annual human-computer interaction conference, ran April 13 through 17 in Barcelona this year, with a newly refactored program spanning six submission areas — from long-form academic papers to hands-on exhibitions of research in action. Buried in the program is a cluster of papers that matter a lot more to working creative technologists than their conference-abstract titles suggest.
The actual research question
A meaningful share of this year’s creativity-focused papers weren’t relitigating whether generative AI can produce creative output — that’s treated as settled. Instead they’re pointed at a harder design problem: in a genuinely collaborative writing or making session between a person and an AI system, how do you keep human agency legible and intact rather than letting the AI’s contribution quietly dominate the result? One accepted paper examines proactive AI specifically as a catalyst for human creativity rather than a replacement for it; another looks directly at balancing human agency against AI contribution in collaborative story writing — the exact tension anyone building a co-writing or co-designing tool runs into the moment a beta user actually tries it.
Why this is the right question to be asking
Every creative tool that’s added an AI assistant in the last two years has quietly made a design decision about this same trade-off, usually without naming it: does the tool wait to be asked, does it suggest continuously, does it ever act without confirmation? Treating that as a first-class HCI research question — rather than a product decision made ad hoc by whichever team ships fastest — is what separates CHI’s contribution from the surrounding wave of AI creativity-tool marketing. A related paper on expressive behaviors as a bridge to human-robot creative collaboration extends the same question into physical and embodied contexts, suggesting the “who’s actually driving” problem isn’t unique to text and image tools.
What to actually take from it
If you’re building anything with an AI-assisted creative workflow — a co-writing tool, a generative design assistant, a livecoding aid — the practical takeaway isn’t a technique to copy, it’s a question to ask honestly about your own interface: at every point the AI contributes something, does the person using it know that’s what just happened, and can they say no to it as easily as they said yes?