MIND (Mixed-Initiative Next-gen Design): Workshop on Blending Agents and Direct Manipulation for Harnessing LLMs

Abstract

Since the 1980s, a key debate in human-centered computing involving machine learning at IUI is between agent-driven systems and direct manipulation. The explosion of Large Language Models (LLMs), particularly auto-regressive as agents serving as chatbots, generative search, and work automation tools, has also brought with it inherent limitations. We posit that efforts to address and alleviate these LLM challenges—hallucinations, unpredictable outputs, lack of transparency, and difficulties in customization—cannot be solved through algorithmic improvements alone but require elevated mixed-initiative interface design at the heart of the IUI community. This workshop aims to bridge the gap between agent-driven automation and direct manipulation by exploring mixed-initiative interaction models that blend the strengths of both paradigms to empower end-users seeking to harness LLMs.