New tool shows how humans can learn with AI

edited by Swati Mestri, reviewed by Andrew Zinin

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

Add to Preferred Sources Credit: Unsplash/CC0 Public Domain A world-first 3D modeling tool developed at the University of Sydney has demonstrated how users can develop a deeper understanding and learn when engaging with AI.

The Treellis system addresses growing concerns that AI users outsource their thinking, potentially inhibiting their learning.

A conference paper outlining the program received the Best Paper Award at the Association for Computing Machinery Symposium on User Interface Software and Technology (UIST 26) in Michigan.

Modern AI tools can turn a few words into a completed 3D design, but the result does not necessarily help users understand how it was made. Researchers in the university's Faculty of Engineering led the development of Treellis, in collaboration with the University of Cambridge and Singapore Management University, to explore how co-creating with AI can provide learning opportunities and develop independent skills.

"Treellis supports more transparent and collaborative human-AI co-creation," said Yixiang Chen, a doctoral student in the Faculty of Engineering and the lead author of the paper. "The tool does not just make things for you. It shows you how things are made."

The interface could be used by professionals with limited experience creating 3D designs, such as medical practitioners looking to develop personalized devices for people with disabilities, said Dr. Anusha Withana, an associate professor in the School of Computer Science who led the tool's development.

"Treellis allows you to control and manipulate the designs without completely relying on AI, and it even lets you make mistakes, which is a big part of how we learn," Withana said.

Rather than producing a final design from a few verbal prompts, Treellis works iteratively with users to refine the work during assembly through both manual and AI-assisted operations. The interface is open source and freely available to anyone with a browser and can be incorporated into any AI platform via an application programming interface (API).

The ability of Treellis to enhance users' learning was tested on two groups by a team led by Withana. All participants were tasked with designing three 3D objects: a cup, a desk and a computer monitor, in increasing order of complexity and with specific requirements. The baseline group built the objects using a standard prompt-based AI tool. The second group built the objects using Treellis.

Unlike the prompt-only tool, Treellis made individual parts of each model and their spatial relationships completely visible. Participants could directly select specific components of the 3D object, progressively adjust their size or position through a graphical user interface and immediately see how these changes affected the model.

Researchers observed that Treellis users actively explored how their 3D designs were assembled and tried small, reversible changes to modify them to match the target. Doing so helped them learn complex 3D design concepts intuitively. Conversely, participants using the typical prompt-only tool often did not understand how the 3D designs were assembled, and their attempts to refine parts of the model introduced new problems elsewhere.

Withana said Treellis helped address two challenges commonly faced by users of AI technology: the gulf of evaluation, or the difficulty of understanding how a system works, and the gulf of execution, or the inability to execute a desired task using the system.

"The tool's interactive functionality narrows both gulfs and prevents users from completely offloading the task to AI," he said.

Anne Roudaut (University of Bristol, U.K.), a professor and chair of the UIST 26 award committee, said, "The UIST Best Paper Awards recognize research that stands out for its originality, scientific rigor and potential to make a lasting impact on the field. The selected papers represent exceptional contributions that not only advance our understanding of human-computer interaction but open up exciting new directions for future research and innovation."

Withana and his team hope to apply the open-source interface in clinical settings. For example, clinicians could use Treellis to interactively design fitted ankle-foot orthotics (AFOs) for children with cerebral palsy, tailor-made wrist splints for patients with arthritis or highly personalized ergonomic grips for daily rehabilitation equipment.

"Treellis breaks the pattern of using AI as a black box," Withana said. "Using 3D content design as an example, it enables users to understand the internal workings of the output from AI, so they can interactively learn what is generated and change it."

Yixiang Chen et al., Treellis: Exposing Editable Assembly Structures for Progressive Human-AI Co-Creation of 3D Models, Proceedings of the 39th Annual ACM Symposium on User Interface Software and Technology (2026). DOI: 10.1145/3830398.3830694 Key concepts Computational additive manufacturing Human-centered AI interfaces Agentic consumer AI AI wearables Machine learning methodologies Provided by University of Sydney Who's behind this story? Swati Mestri Swati Mestri holds a bachelor's degree in Electronics Engineering and has worked as a content editor since 2019. She has experience editing research documents across technology, health care, and materials science, and has a particular interest in technology and space. Full profile →

Master's in physics with research experience. Long-time science news enthusiast. Plays key role in Science X's editorial success. Full profile →