The Internet poses many opportunities and risks for university learning. Filtering and processing enormous amounts of information is a challenge for all students. Relying on the first few search hits and avoiding contradictory information, they may overlook or ignore facts and well-founded knowledge in the vast masses of information. To reduce complexity and increase time efficiency, students are moving towards outsourcing thinking processes and complex information processing (sometimes subconsciously) to online tools (e.g., it is easier to google something than to remember it). Knowledge acquired in this way is not only rather episodic, inert, not well interconnected, and difficult to retrieve, but also impairs a deeper understanding of subject content or contexts and leads to misconceptions. This phenomenon of insufficient information processing is referred to as negative learning (NL). NL usually occurs unintentionally and subconsciously and is difficult to avoid, particularly online, as it can easily be amplified through distorted or counterfactual information. Although learning research can describe what is (not) learned on the Internet, it cannot explain the underlying processes or how learning can be fostered. We do not know which factors promote or mitigate NL in the interactions between human learning and dynamic, partly artificially intelligent online learning environments. In PLATO, the Internet is modeled as a learning space that comprises both the theoretically available online information resources and the resources students actually use in their course of the study. On this basis, discipline-specific and individual ‘information maps’ are created and new integrative theories and models for the explanation and prediction of NL are developed, which are then tested in experimental and longitudinal studies. Using innovative approaches from the fields of psychology, neuroscience, computer science, pedagogy, media and communication sciences, linguistics, mathematics, and physics, PLATO models neural, cognitive, affective, and motivational states as well as behavioral patterns in learning on the Internet and examines the way they change during interaction and communication processes. These changes are investigated online and by means of narrative case studies, text and data mining, simulations, big data analyses, and other technologies. The results are then incorporated into the development of practical application tools to prevent and transform NL.