AI as a Tool for Thought: Defining the Modern Knowledge Worker

Overview

Our researchers focus on building AI tools that help knowledge workers to think through problems and generate responses, rather than AI tools that aim to do this autonomously. While a lot of the recent interest in AI is in generative language models, which can produce documents, images, and designs, it is becoming increasingly clear that such tools are good *for* generating outputs, but are not good *at* completing the work itself. This distinction matters: using AI to bypass thinking leads to shallow outputs and poor decisions. Instead, we focus on designing tools that support the thinking process—tools that challenge, critique, and guide users, while keeping them in control.

For instance, rather than asking a language model to write a strategy document from a prompt, we use it to ask probing questions, critique our ideas, and help refine our thinking—before drafting. The result is a document we own, with decisions we understand and can justify. Similarly, in medical imaging, instead of replacing clinicians, we design AI tools that intervene only when they detect high-confidence evidence contradicting a diagnosis—supporting, not supplanting, expert judgement.

We are applying the “AI as a Tool for Thought” approach across diverse domains—including medical imaging, defence analysis, and professional writing—through projects funded by the Australian Research Council and the Australian Advanced Strategic Capabilities Accelerator. Unlike typical AI research that focuses on accuracy and performance metrics, we design studies to ask: Is the tool genuinely useful? Does it improve quality or speed? Do users understand and trust it? Do they over-rely on it? Does it cause de-skilling? The answers to these questions consistently show that the tools we need to build look different from mainstream AI systems—but they’re better accepted, more trusted, and lead to higher-quality outcomes.