The New Future of Work

Large language model-powered tools like Copilot have the potential to increase labor productivity more than any technology in a generation. Motivated by the significance of this moment, researchers from across Microsoft have come together to measure and improve the productivity impacts of Copilot, as well as understand and mitigate externalities.

This workstream aims to examine how AI can and should influence software engineering – including its effects on developers, how it improves developer efficiency, how it can assist in keeping software safe, and what the potential risks are.

Through appropriate reliance, we aim to help people who use AI systems find a balance between over-trusting AI outputs and accepting them even when they are incorrect (aka overreliance on AI) and under-trusting—not accepting, or not using AI systems even when they perform well (aka underreliance on AI).