Artificial intelligence, machine learning and deep learning: definitions and differences
Deepak Jakhar, Ishmeet Kaur · Clinical and Experimental Dermatology · 2019
Artificial intelligence (AI) and its application is the next big thing in dermatological imaging, and includes, but is not limited to, image acquisition, processing, interpretation, reporting and follow‐up planning.1 2 Furthermore, there are additional benefits of data integration, data storage and data mining. In fact, the possible applications are so many that AI is expected to become an inseparable tool in a dermatologist’s life. However, most dermatologists are still illiterate in AI. While we are still trying to understand AI, two other terminologies: machine learning (ML) and deep learning (DL) are already the talk of academicians. Most of the medical schools still do not teach AI as a part of academic curriculum, and the interchangeable use of these three terms in literature provides no help to young researchers. The younger generation, which is at the forefront of facing this revolution, needs an awareness and clarity in terms of definition and differences of the three terminologies. AI refers to a field of computer science dedicated to the creation of systems performing tasks that usually require human intelligence.2 3 It can be loosely interpreted as incorporation of human intelligence into machines. In AI, machines complete the task based on the stipulated rules and algorithms. AI is an umbrella term for any computer program that has the touch of human intelligence and encompasses ML and DL (Fig. 1).4 Deep learning is a subset of machine learning, which in turn is a subset of artificial learning. ML is a subset of AI, which includes all the approaches that allow machines to learn from data without being explicitly programmed.5 The intention of ML is to train machines based on the provided data and algorithms. Using the processed data and information, the machines learn how to make decisions.4 ML is dynamic, meaning that it has the ability to modify itself when exposed to more data. The ‘learning’ aspect of ML means that the ML algorithms attempt to minimize the errors and maximize the likelihood of their predictions being true. In short, ML is simply a technique to realize AI. DL is a subset of ML, and incorporates computational models and algorithms that imitate the architecture of the biological neural networks in brain [artificial neural networks (ANNs)]. Whenever the brain receives new information, it tries to compare it with already known information to try to make sense of it. The brain deciphers the information through labelling and assigning the items to various categories, and DL employs the same concept. ‘Deep’ is a technical term, and refers to the number of layers in an ANN. There are three types of layer: the input layer (receives the input data), the output layer (produces the result of data processing) and the hidden layer (extracts the patterns within the data). A deep ANN differs from the superficial ANN (single hidden layer) by having a large number of hidden layers, which means it is able to perform more complex tasks.4 While the data move from one hidden layer to another, simpler features recombine and recompose to complex features. Simply put, DL works exceptionally well on unstructured data and has higher accuracy than ML, but requires a huge volume of training data, along with expensive hardware and software. There is a great enthusiasm and simultaneous fear for the development of AI systems in dermatology. The fear results from the perception that AI may become a threat to the medical fraternity in future. The existing distress is the greater because most dermatologists are still unaware of the basics and foundation of AI. An updated and upgraded dermatologist should be aware of the basic definitions and principles of AI systems. Accept or deny, we are already in the AI era, and now is the time to understand it. Conflict of interest: the authors declare that they have no conflicts of interest.