Symbol-Based Artificial Intelligence

The authors deserve many thanks for their excellent review article (1). They describe “neural networks” appropriately as “a form of artificial intelligence (AI).” This is the form that is predominant in the public perception. For this reason I wish to describe as an addition a different form of AI that has great potential in the area of medicine.

Further to “neural networks,” there is “symbol based AI.” In difference to machine learning, in symbol based AI, knowledge is transformed in an explicit form that a computer can process. No imitation of biological neurons takes place and no neural networks are used. Rather, knowledge is made interpretable for a computer by applying logic, rules, and semantic networks.

Especially in the area of clinical decision support systems, such symbol based AI has a big role. Existing rules—for example, from medical guidelines, can thus be made interpretable for computers and applied to concrete patients. A recent publication by Lichtner et al. (2023) is an example of this (2).

By contrast to neural networks, decisions based on symbol based AI are reproducible and transparent. Compared with machine learning approaches this facilitates their use especially in the clinical sector and in life-critical decisions.

As the authors emphasized in their article, doctors’ professional competence is required for a responsible use of AI systems of any shape in the medical setting. The German Medical Association’s ethics committee published a relevant position paper in 2012 (3). Accordingly, doctors are obliged to acquire knowledge of how to handle this new technology competently, for example, including its limitations.”

The excellent CME article and its prominent placement in the title page of Deutsches Ärzteblatt make a valuably contribution to this. My sincere thanks go to the authors.