Artificial intelligence (AI) is a comprehensive term used to generally decribe the ability of computer systems to perform complex tasks traditionally associated with human intelligence, such as decision-making and problem-solving (Joda, Waltimo et al. 2018; Shan et al. 2021). Technically, AI can be divided into two main categories: “weak” and “strong” AI. The former, also known as artificial narrow intelligence (ANI), consists of the ability to complete specific tasks such as identifying a particular target in a series of pictures (computer vision) without any intrinsic cognitive abilities. Strong AI, on the other hand, is defined as a system that replicates typical human mental abilities and consciousness, working exactly like a human mind (Scerri and Grech 2020). Whilst no form of strong AI exists yet, the application of weak AI systems in daily life has considerably grown, revolutionizing the way humans interact with technology and approach complex tasks. AI research and applications have also continuously grown in medicine and dentistry.

Today, most AI applications in medicine are based on machine learning (ML) and focus on the analysis of radiological imaging. After training on existing datasets, ML algorithms can provide accurate predictions on new input data (Bernauer et al. 2021). In classic ML models, human intervention is required to design and provide the system with a labeled dataset, on which the algorithm can be trained. 

In this context, a subset of ML is known as deep learning (DL). Compared to ML, DL models can observe patterns in the data and autonomously hierarchize the raw information of the training dataset without the need for human intervention (Ren et al. 2021). Artificial neural networks (ANN) constitute the backbone of DL algorithms. Neural networks are made up of node layers: one input layer, one or more hidden layers, and an output layer, and each node is an artificial neuron connected to another neuron. ANN with more than 3 layers can be considered a DL algorithm (Fig. 1). ANN can learn and improve their accuracy through training datasets, however, this process requires large amounts of data (Bernauer et al. 2021).

Various AI-based applications and tools have been developed and introduced to help dentists in daily practice, such as identification and classification of pathologies on digital imaging (Litjens et al. 2017). Intraoral 2D X-rays, cone beam computed tomography (CBCT), and optical scanning are therefore of particular interest in dentistry. JPG, DICOM, and STL data are routinely collected for diagnosis, treatment planning, restorative design, and re-evaluation, and therefore offer an interesting source for the development and use of AI technology (Hung et al. 2020). AI diagnostic models have shown successful results in the radiological identification of dental pathologies such as interproximal carious lesions, root fractures, periapical lesions, and periodontal bone loss (Fig. 2) (Devito et al. 2008; Johari et al. 2017; Chang et al. 2020; Endres et al. 2020).

AI models have also been successfully introduced in orthodontics, where knowledge-based algorithms and computer vision methods have been demonstrated to be a reliable way to identify and measure anatomical landmarks on cephalograms (Gupta et al. 2015; Park et al. 2019). 

The aim of this narrative review is to describe and summarize how AI is currently used in implant dentistry. The structure of this overview follows the digital implant workflow: diagnostics, clinical treatment protocol (including the laboratory workflow), prognosis and maintenance (Fig. 3), and describes the work steps in which AI can play a role today or in the future. The focus is on the areas of dental imaging, treatment planning, guided implant surgery and patient monitoring.