Lecture 1.2 – Introduction to AI Research (MIT How to AI Almost Anything, Spring 2025)

Paul Liang's lecture focuses on how to conduct AI research, covering idea generation through bottom-up and top-down approaches, literature review, research question formulation, experimentation, and report writing. The lecture outlines a timeline for bottom-up discovery, emphasizing the importance of scoping research areas, experimenting with state-of-the-art methods, analyzing successes and failures, and proposing new ideas. It also discusses various research themes, including AI with non-deep learning modalities, AI for sensor data, AI reasoning, interactive agents, embodied and tangible AI, socially intelligent AI, human-AI interaction, and ethics and safety, providing examples of research questions and resources for each theme. The lecture further explains how to do shared reviews and read papers using resources like Google Scholar, Papers with Code, and blog posts.