“Coming together is a beginning, staying together is progress, and working together is success.”

Henry Ford

In the last newsletter, we discussed examples of machine-to-machine or model-to-model collaborative learning in the form of meta models used in machine learning ensemble methods such as bagging (parallel), boosting (sequential), and stacking. We also discussed generative adversarial network (GAN) as two neural networks in the form of a generator and a discriminator and a third possible example of model-to-model collaborative learning as transfer learning, as this concept involves deep neural networks facilitating learning from a previous model that trained from a prior data set.

This week, we can discuss machine-to-human synergy. A key recent development in AI is the paradigm of cognitive elements being incorporated into existing AI tools (the so-called “smarter” AI).

Deep reinforcement learning is still not widely applied in clinical medicine, but is perhaps one of the most promising AI tools for the future, especially in complex decision making in the ICU or OR setting.

Deep Q Network (DQN) is an example of deep reinforcement learning in that it leverages two neural networks (Q and target networks) to maximize the Q value (or simplistically the maximum reward for each action).

Another important philosophical dimension is the machine-to-human synergy between the data scientists and clinicians during the planning and execution of AI projects. This collaboration should happen from the inception and throughout the entire process, not only “bookend” the project.

Ultimately, the entire AI in healthcare agenda will need to be driven by human leaders and their relationships and esprit de corps. The Alliance of Centers of AI in Medicine (ACAIM) convenes over 60 centers on a monthly basis to foster collaborations. This human-to-human network can translate to more collaborative learning strategies such as federated and swarm learning. Whilst the former requires a central server that connects participating models, the latter is a decentralized machine learning framework that uses blockchain technology and leverages distributed data.

We can learn about these collaborative efforts as well as other topics at our next AIMed Global Summit, scheduled for Spring of 2023 in southern California.

Hope to see you many times before then!