With so much buzz around AI in the press, we should see big payoffs in science. Yet it doesn’t seem like major advancements with chatbots and image generators in recent years have also extended to scientific breakthroughs. In this talk, I will discuss the concept of representation learning: the way that AI algorithms understand data, and a major bottleneck in repurposing existing AI algorithms to solve new problems. I will specifically highlight challenges and opportunities in materials science applications including molecular simulation, characterization, and small molecule design.