A recent report written by Professor Sang Wan Lee from the Department of Bio and Brain Engineering highlights how discoveries in “human decision neuroscience can offer valuable insights into action control systems for robots”. Professor Lee worked in collaboration with researchers from the University of Cambridge, Japan’s National Institute for Information and Communications Technology, and Google DeepMind to publish “Decision-making in brains and robots — the case for an interdisciplinary approach” in Current Opinion in Behavior Sciences. The report tries to encourage greater interdisciplinary studies between computer scientists and neurologists, and it serves as a great example of the research published by KAIST’s many laboratories.

Professor Lee leads KAIST’s Laboratory for Brain and Machine Intelligence, which explores the intersection between artificial intelligence and neuroscience. The lab aims to understand the modes of learning used to guide a brain’s behavior, listing “learning, inference, and cognitive control” as its three key areas of research.

Lee’s report criticizes the dominance of reinforcement learning (RL) techniques in computer science. Generally, RL is the process where computers improve their actions over time through trial and error. These techniques are used in the decision-making process of autonomous robots, such as Google DeepMind’s AlphaGo, the first computer to beat a professional Go player.

Lee sets out to encourage greater interdisciplinary interaction in automated decision-making by showing how advances in neuroscience could supplement the failures of RL. For example, Lee’s report demonstrates how the human brain’s ability to evaluate its own performance, known as metacognition, is useful in overcoming the overfitting or underfitting of complex data in machine learning. The report states, “Metacognitive learning allows for rapid adaptation [sic] to context change while maintain [sic] robustness against environmental noise.”

Lee’s work provides a framework for future interdisciplinary studies of automated decision-making and displays the potential for groundbreaking research. Responding to KAIST’s PR Team, Lee stated, “Copying the brain’s code could greatly enhance the flexibility, efficiency, and safety of robots.”