To the Editor:
It was a pleasure to read the editorial by Dr. Hemmerling, and it was very promising to see open discussion about technological innovation affecting the field.1 I agree that automated robotic systems appear inevitable. However, I would wish to comment that it is important to highlight both artificial intelligence and robotics as two distinct innovations that work synergistically together in this context. In a future where robots are responsible for delivery of anesthesia in theater, both innovations will require substantial development to ensure that the system can adequately learn from and respond to variation. Assuming any rate of improvement to these systems, they may soon begin to outperform humans, and the input of the human anesthetist may gradually shift toward a supervisory role.
One of the main advantages of machine learning algorithms is that they are capable of outperforming human decision-making, provided that their datasets are reliable and their decision-making has been refined appropriately before use. Dr. Hemmerling illustrates a good example of what anesthesiology in theater may look like in 2030, assisted by a robot. In this example, it is the human who offers the instructions and the robot executes them. This contrasts with other specialties, such as radiology and dermatology, where deductive artificial intelligence systems may offer diagnoses for the human to confirm and then act upon. If artificial intelligence systems are trained on large, diverse, and clean datasets, they should, in theory, be able to make decisions on the type of anesthesia to be performed and the various target parameters. Moravec’s paradox dictates that programming artificial intelligence systems to complete these complex cognitive tasks is often relatively straightforward when compared to simple physical robotic tasks.2 Before we see robots take over the delivery of anesthesia, they may begin to take over the instruction of what to deliver. Dr. Hemmerling compares the technology to the emergence of self-driving cars, which require similar alliance between artificial intelligence and robotics. This is a particularly appropriate comparison considering that issues of data homogenization, accountability, and unrepresentative datasets must be resolved before both of these technologies populate our highways and operating theaters. The timescales are uncertain, but that does not make them impossible.
The author declares no competing interests.