The Law of Mass action predicts that all adverse drug reactions are related to the concentration of the drug at the site of action, and therefore to the administered dose. In other words, there is no such thing as a non-dose-related adverse drug reaction. That being so, there are three types of adverse drug reactions in relation to the dose or concentration of the drug with which the reaction is associated, determined by the relation between the concentration of the drug at the site of action, determining the adverse reaction, and the range of concentrations expected to be associated with therapeutic benefit: those three types are hypersusceptibility reactions (at concentrations below therapeutic), collateral reactions (at concentrations in the therapeutic range), and toxic reactions (at concentrations above therapeutic). Different types of reactions also imply different degrees of predictability. Artificial intelligence (AI) is generally of no value in predicting adverse drug reactions of the different types in an individual, but it may be used in studies of the susceptibility factors that are likely to be associated with risks of harms. AI may also be useful in analysing large databases, such as Vigibase, Eudravigilance, and the US FDA Adverse Event Reporting System (FAERS). There is also as yet unrealised scope for using AI to analyse data in pharmaceutical companies’ clinical study reports and in clinical datasets, such as claims and complaints databases, electronic health records from hospitals and general practice records, and information from poisons centres.