The term “artificial intelligence” was coined at the 1956 summer research workshop at Dartmouth College (Hanover, NH, USA) . 6 years after Turing had asked the question “Can machines think?” . The 1956 workshop aimed to develop a ma- chine that could pass the Turing test, i.e., exhibit intelligence at the level of hu- mans. Machines that actually think have not yet been developed, but AI also in- cludes machines that perform simulated thinking, i.e., solve tasks that would other- wise require human intelligence. As such, an automated ECG interpretation software coded line-by-line is within the field of AI Machine learning is the subset of AI that relates to self-learning algorithms trained with data, although at times AI is used synonymously with machine learn- ing. Deep learning is a special case of machine learning, whereby information is passed from layer to layer of artificial neu- rons in a neural network, a process that was inspired by human neuronal comput- ing. The main advantages of artifi- cial neural networks over other machine learning methods are that extraction of specific features is not needed and that the networks are flexible to comprehend complex data. The main limitation of neu- ral networks over other machine learning methods is the need for large, annotated data sets for the model to be successful (i.e., to converge). Many different neural architectures exist; of special interest to AF are recurrent neural networks (RNNs) and convolutional neural networks (CNNs). RNNs are ideal for long time series and fore- casting, whereas CNNs are used for image and signal analysis.

Although AI technically includes pro- grams that are not classified as machine learning, this review will focus on the ma- chine learning part of AI algorithms, and special emphasis is put on neural networks (deep learning).

Machine learning (ML) constitutes a subdiscipline of arti‐ ficial intelligence (AI) and is characterized by the auto‐ mated detection of patterns from data. Its use is particularly advantageous in the detection of non‐linear associations and modern computational power allows for the analysis of large datasets. The use of artificial neural networks has already led to transformation of scientific studies and understanding of vastly different areas such as language processing, object recognition, and predictive analysis. One important aspect is the increasing availability of large standardized digitized datasets, for example, electronic health records or imaging and ECG databases