Klaus Schwab of the World Economic Forum calls it the Fourth Industrial Revolution; artificial intelligence (AI) and machine learning (ML) are poised to penetrate every industry, discipline, and economy, and health care is no different. A survey of healthcare executives by AI market research firm TechEmergence revealed that 50% of respondents believe that ML will be ubiquitous in healthcare by 2025.[i] As soon as next year, ML technology will be incorporated in 30% of health systems as tools assisting physicians in making personalized treatment decisions.[ii]
Though ML represents a disruptive force in the economy, particularly in the organization of human resources, understanding the technology and its limitations is crucial for healthcare stakeholders to direct its advancement for the benefit of patient care.
What is Artificial Intelligence and Machine Learning?
The field of AI is concerned with creating computer systems that are able to perform tasks normally requiring human intelligence, including but not limited to: visual perception; speech recognition; decision-making; and translation between languages. ML describes one approach to AI that uses algorithms beyond condition- or rule-based structure and “learns” from data to improve its performance. ML is concerned with creating algorithms that give computers the ability to learn from data and make predictions or decisions without being explicitly programmed. Major software and hardware advances that have enabled ML include parallel computing, the computing of simultaneous processes using multiple CPU cores, and the cloud, which allows for unprecedented access to large amounts of patient data (necessary for the training of some ML algorithms) and provides the computational capacity necessary for the development of ML models. One example of ML models include neural networks that are meant to conceptually imitate the structural cognitive pathways of the human brain).
ML in Healthcare:
The adoption of ML in healthcare is driven largely by big data (i.e. electronic health records, EHR) and personalized devices, effectively engaging ML development in a positive feedback loop. As medical futurist Bertalan Mesko, PhD, states, medical data is expected to reach 44 zettabytes by 2020, and the sheer task of managing that data will be best achieved by the implementation of ML tools.[iii] ML relies on big data, which will necessitate advances in ML to extract cogent conclusions from big data.
There are applications of AI in anesthesiology as well, and it isn’t unreasonable to imagine a system, through the cognitive processing of patient data, probabilistically and/or algorithmically making efficient and accurate decisions in “monitoring the depth of anesthesia, determining the amount of anesthetic gas to administer, somatosensory evoked potential monitoring, classifying patients, and mitral valve analysis to coding and billing”.[iv] By enabling data-driven physician and consumer decision-making, ML may play a key role in lowering healthcare costs.
However, there are important questions to consider with ML, as there are with any major technological advance.
Issue #1: Ethics in Practice
If a machine makes a mistake, who bears the ultimate responsibility? From an ethical and legal standpoint, it is substantially more difficult to hold an algorithm accountable for a “wrong decision”. This is because ML models are not explicitly programmed to make decisions based on certain conditions but instead have been developed to learn from data in order to devise treatment pathways optimizing patient outcomes. For a health care professional non-conversant in ML or computer science, the worki