Microsoft Research Explores Tools for Thought and the Future of AI in Knowledge Work at CHI 2025

At CHI 2025, Microsoft Research unveils research and prototypes on designing Artificial Intelligence as a collaborative partner in human cognition, decision-making, critical thinking, and creativity.

Microsoft Research has announced a suite of new research papers and prototypes focused on the integration of Artificial Intelligence into knowledge work, presented at CHI 2025 under the ‘Tools for Thought’ initiative. The central theme explores how Artificial Intelligence can be designed not just to automate workflows but to actively enhance critical thinking, creativity, and decision-making for knowledge workers. The effort spans empirical studies, interactive prototypes, and a multidisciplinary workshop, aiming to redefine the relationship between humans and Artificial Intelligence in professional settings.

A key study surveyed 319 professionals to understand how generative Artificial Intelligence impacts cognitive effort and critical thinking. Results revealed a nuanced trade-off: for high-stakes tasks, workers spent more effort on critical evaluation with Artificial Intelligence, while routine or time-pressured tasks saw a decrease in such engagement. The research also highlighted that increased confidence in Artificial Intelligence could reduce critical thinking, while strong self-confidence boosted it, albeit at a higher mental cost. Microsoft recommends that Artificial Intelligence tools should motivate and scaffold critical thinking by offering explanations, guided critiques, and proactive prompts, thereby supporting knowledge workers as both ‘thought partners’ and provocateurs.

Three prototype systems were introduced to demonstrate practical applications. In decision-making, the RecommendAI and ExtendAI tools represented distinct approaches: the former provided actionable recommendations, sometimes at the expense of user engagement, while the latter promoted deeper self-reflection but could reinforce existing biases. In the context of meetings, AI-powered prototypes experimented with both passive visualizations and active questioning to help participants stay aligned with goals, carefully balancing information delivery and interruption. Lastly, the YES AND system used multi-agent Artificial Intelligence to enable diverse brainstorming by simulating conversations with persona-driven agents, helping users overcome groupthink and generate innovative solutions.

To advance the field, Microsoft Research is also facilitating a CHI 2025 workshop, uniting over 60 experts from academia and industry to discuss frameworks for understanding, protecting, and augmenting human cognition with Artificial Intelligence. The initiative reflects the growing momentum toward designing Artificial Intelligence not merely as an answer generator but as an enabler of deeper, broader, and more effective thinking in knowledge work. The goal is to foster a community and ecosystem in which Artificial Intelligence amplifies—not diminishes—our ability to think critically, creatively, and strategically.

72

Impact Score

UK and EU Artificial Intelligence regulatory outlook for May 2026

The UK is moving ahead with targeted Artificial Intelligence measures in policing, online safety, cyber security and copyright policy, while the EU is refining how the EU Artificial Intelligence Act will apply in practice. Consultations, new offences and implementation deadlines are shaping the next phase of compliance on both sides.

Germany sets out national implementation of the Artificial Intelligence Act

Germany has published a draft law to implement the European Artificial Intelligence Act through new supervisory structures, clearer institutional responsibilities, and measures designed to support innovation. The proposal puts the Federal Network Agency at the center of enforcement while preserving sector-specific oversight in sensitive fields.

ECB warns banks about new Artificial Intelligence security risks

The European Central Bank has called major banks to an emergency meeting over cybersecurity risks tied to advanced Artificial Intelligence models. Regulators want banks to speed up security updates as newer tools make it easier to find and exploit vulnerabilities.

Anthropic keeps Mythos restricted after vulnerability findings

Anthropic says its cybersecurity model Mythos is powerful at uncovering software flaws but remains too risky for broad release. Early testing found large numbers of vulnerabilities across major software and open source projects, while fixes have lagged far behind discoveries.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.