We thank Dr. Burns for his comments on our article.[1] Many definitions of AI have been proposed. According to LeCun, AI “allows machines to perform tasks and solve problems normally reserved for humans”.[2] Since some machines aim to reproduce human tasks, particularly those related to diagnosis, it makes sense to compare how physicians and AI work, and to draw consequences for clinical practice. However, we need to agree on what should be called “intelligence”. Strictly speaking, a machine is not intelligent when it performs a task. It does not understand the task, the process or the result it produces. The intelligence of the machine lies in its ability to learn.[2] AI is therefore “efficient” in its ability to solve clinical tasks, and “intelligent” in its ability to learn things and improve its performance. In the case of deep learning, this “intelligence to learn” is linked to a particular architecture: all layers of neural networks are trainable and the learning performed by one layer is used by the following layers to form increasingly complex and abstract concepts.[3]Dr. Burns is right to point out the risks of blind use of machines, i.e. ignoring their functioning and limitations. However, we must acknowledge that there are still many gray areas in human cognition, and that most physicians are unable to explain how they make decisions.[4] Yet human cognition has been at the heart of medical decision-making for centuries.
We believe that the combination of human and artificial intelligence will make it possible to overcome the limitations (including biases) associated with the two intelligence. For the physician, it will not be a question of “using” the intelligent tools, but of “knowing how to use” them. The difference lies in being aware of the added value of the tool (e.g. saving time when confronted with many data, directing attention when dealing with atypical data, gaining precision when interpreting complex data), and using the tool for this purpose alone, without wanting to substitute it for medical decision (and physician’s intelligence).
Source: https://www.cmaj.ca/content/intelligence-artificial-intelligence-and-its-use
