“Artificial intelligence is the Apollo program for the 21st century.”
-Demis Hassabis, Founder of DeepMind
One of the most renowned AI efforts is Google’s DeepMind, founded by neuroscientist Demis Hassabis. His philosophy is to “solve intelligence then use that to solve everything else”, and his DeepMind enterprise has soundly defeated human champions in the games. Go (AlphaGo) as well as other games (AlphaZero).
Clinical medicine, especially fast-moving subspecialties (such as intensive care, surgery and anesthesiology, emergency medicine, and cardiology), does not resemble the aforementioned gameGo. Instead, clinical medicine much resembles real-time strategy (RST) games like Starcraft II, which have innate complexities such as imperfect information, long term planning, real-time continuity, and large action space. Despite these challenges, DeepMind has been able to take on this game genre with its AlphaStar, which uses deep neural network that learns from game data via supervised and reinforcement learning as well as a multi-agent learning algorithm. This recent development in deep reinforcement learning is particularly encouraging for real-time decision-making so often encountered in the complex milieu of clinical medicine.
Another new development at DeepMind is generative query network (GQN), composed of a generation as well as a representation network. This dyad of networks learns like an infant by making sense of its observations with its own sensors without human labeling.
This concept of an intelligent agent learning about its own environment without human supervision is very innovative. The implications of this astonishing representation learning technology for biomedicine, particularly medical imaging, will be profound.