AI Reimplements HCI Research Systems Directly From Papers, Raising Questions About Artifact Reproducibility

Most technical HCI research projects never release their software. This gap—where code and interactive systems disappear after publication—creates real friction for the research community, making it hard to compare work fairly, build on existing ideas, or verify findings.

The Problem With Missing Artifacts#

The absence of accessible software artifacts in HCI research creates cascading challenges. Researchers cannot easily extend or adapt systems for new contexts, new projects lack solid baselines for meaningful comparison, and independent verification of results becomes nearly impossible. This invisibility of the underlying tools undermines both the pace and rigor of knowledge production across the field.

A New Approach: "Revibing" Papers#

A recent arXiv paper from Eytan Adar and colleagues explores whether agentic AI—automated systems capable of understanding and executing complex tasks—can recover interactive software by reading research papers alone. The team calls this process "revibing," reimplementing systems directly from published descriptions. They developed a metric to assess success and tested the approach on recent work presented at UIST, the premier venue for interactive systems.

Promising Early Results#

The researchers interviewed the original authors of the papers they reimplemented and found the revibed systems often produced code usable enough to serve as strong baselines for follow-up research. While limitations exist, the results suggest that AI-assisted reconstruction of research artifacts is more than a theoretical possibility—it could become a practical tool.

Implications for HCI Research#

If this technique scales, it could fundamentally reshape how researchers produce, share, and evaluate artifacts. Rather than relying on often-absent source code, the community might routinely regenerate systems from published work, making reproducibility and comparative evaluation far more feasible. The work hints at a future where the absence of released code becomes less of a barrier to building on prior research.