*4.3. Artificial Intelligence: Machine Learning (ML) and Knowledge Representation and Reasoning (KR and R)*

AI applications have been in use for at least 30 years in the medical and industrial contexts, but the use of AI in the social sciences is a current trend. In a social context, AI could provide useful tools for discovering new social phenomena and for testing existing theories. The basis for such AI systems can be found in the computational graph theory, which was proven to work well in computer networks [111].

Many AI-based methodologies have been put forward to construct models based on mobility patterns and predict the behavior of people individually or as a group [112,113]. These include stochastic models such as Markov Models MM) [114] and Bayesian Networks (BN), as well as non-stochastic models such as Artificial Neural Networks (ANN) and Decision Trees (DT) [115]. While researchers have used ML techniques such as ANN and DT [116], the stochastic models are preferred over these due to the uncertainty in or unpredictability of human behavior [116]. In several studies e.g., [115,116] researchers have also used Bayesian Networks.

ML and KR and R are two major techniques of AI that have the potential to provide solutions to matching transnational industrial partners through their connections to transnational academic partnerships. While humans learn things by using their brains, in ML algorithms are used by computers and robots to learn automatically (without explicit instructions) [117]. ML algorithms are based on training data, a kind of sample data. Data mining, as a field of study within ML learning, "is one of the most effective alternatives to extract knowledge from the great volume of data, discovering hidden relationships, patterns and generating rules to predict and correlate data" [118] (p. 687). Such a technique has often been applied in social network analysis [119,120].

KR and R [121] focuses on implementing knowledge repositories built by semantic descriptions that can be interpreted by both humans and machines. There are many KR and R formalisms used to implement semantic models, such as ontologies, databases and semantic rules. The former formalism is getting attraction in several fields, such as in the industrial automation field [122,123], needing to build semantic models including information from different and interrelated concepts. One of the major benefits of using such technic is the possibility of adding a layer of reasoning in order to discover implicit knowledge from the explicit data graphs that are fed to the model. Further, ML and KR and R can be integrated to work on different layers of implementation, i.e., using them for different processing data format, syntax and semantics.
