The future of AI in Healthcare — A conversation between Fei-Fei Li and Andrew Ng-April 2021

Andrew Ng is an American computer science researcher. He is an associate professor in the Department of Computer Science at Stanford University. He is the originator of the Coursera course AI in Healthcare.

Fei Fei Li is an American computer scientist and researcher specializing in computer vision. She is a professor of computer science at Stanford University. Fei Fei Li is co-director of @StanfordHAI: Advancing AI research, education, policy, and practice to improve the human condition.

Complexity of the medical sector

The diversity of players in the world of medicine and the complexity of healthcare systems around the world distinguish this sector from other sectors of the economy. Medicine is not just a capital-intensive activity; it involves human lives.

Usual issues

For Andrew Ng, the problems of deploying algorithms are the same as in commerce or industry. They concern privacy, user security, change management and also the generalization of models on new data. Andrew Ng underlines the difficulty of putting models into production and going beyond good performance on data sets.

For Andrew Ng, instead of trying to improve the engineering of models such as ResNet or U-Net, researchers should first work on the quality of the data used to train these models. It is necessary to move from model-centric AI to data-centric AI, focusing on the selection and quality of data.

Expanding the spectrum of data digitization

Fei Fei Li insists on the fact that the management of chronic diseases at home is not in the scope of health data production. Reports by nurses are rare and done by hand. Fei Fei Li believes that it is necessary to extend the use of data collection technologies to all care.

In hospital spaces, early applications of these technologies could enable more efficient clinical workflows and improved patient safety in intensive care units and operating rooms. At home, these AI-based technologies could extend the independence of older adults and improve the management of people with chronic disease.

For Fei Fei Li, Covid-19 will certainly have been a gas pedal in the awareness of the potential of telemedicine, the need to adjust the capacity of care in hospitals to face crises and the need to develop home care.

Ethics and Privacy

One of the most important aspects of medical research is ethics and privacy. In addition to working on algorithms adapted to medicine, Stanford researchers are working on ways to encrypt medical data (Homomorphic Encryption).

According to Fei Fei Li, in order to avoid data and models being biased, it is necessary to recognize the human responsibility and not the responsibility of the algorithms in generating these biases. Biases exist as soon as the data is collected or the algorithm is designed by the scientists.

Dialogue between engineers and doctors is necessary

For Andrew Ng, it is necessary to estimate the feasibility of each project as well as the value brought to doctors and patients before launching into an Artificial Intelligence project. To do so, it is necessary to bring together the different parties : Deep Learning engineers and practitioners, to share knowledge and objectives with humility.

Fe Fei Li advises all his students in Deep Learning applied to medicine to shut down their computers and accompany the practitioners in the field. It is necessary for the engineers to perceive the difficulties encountered by the medical staff on a daily basis.

Algorithms do not replace doctors, they help them in their relationship with patients

Andrew Ng says he helped design a model that predicts the end of a patient’s life. For him, the prediction was not the most important thing: it was the dialogue that this algorithm allowed between the patient, the patient’s family and his caregivers.

Diplodocus interested in the applications of artificial intelligence to healthcare. Twitter : @