Learning paths in adaptive and artificial-intelligent learning programs: A critical analysis from a media didactics perspective.
The paper contrasts interactive, adaptive as well as artificially intelligent learning programs. Adaptive and AI-based applications prove to be more complex in development and have not been able to establish themselves so far. To assess the opportunities of these technologies, their capabilities are compared. Adaptive learning programs are particularly suitable for skill acquisition, AI-based solutions when an expertise model cannot be explicated. Didactic and pedagogical challenges that AI-based learning applications will have to face in the future are named.
~
KERRES, Michael, BUNTINS, Katja, BUCHNER, Josef, DRACHSLER, Hendrik and ZAWACKI-RICHTER, Olaf, [no date].
Digital learning applications are based in the first approach on analog media, such as books or other recorded learning offerings that can be played back via media devices and broadcast educational programs received via radio and television. With the computer, interactive media emerged on this basis in the 1990s, breaking up the linear structure of these media and allowing branching in playback ("hypertext"): Learners can retrieve linked information units interactively and generate their individual learning Path (Kuhlen, 1991).
KUHLEN, Rainer, 1991. Hypertext: ein nicht-lineares Medium zwischen Buch und Wissensbank. Berlin: Springer. Edition SEL-Stiftung. ISBN 978-3-540-53566-9.
The second approach refers to digital systems that present information adaptively: They detect the learner's level of knowledge in order to adapt the progress of the content presentation to it. The idea of adaptive learning programs – initially inspired by behaviorist learning theories – has existed since the mid-20th century (Frank, 1964); since the 1980s, so-called "intelligent tutorial systems" have been developed, mostly based on rule-based systems that control the learning process on the basis of a model of ideal performance (Anderson et al., 1985). Learning is understood as a Control Loop in which the presentation of the subject matter is closely linked to the learning process and the performance of the learners.
The third approach includes current developments that - mostly associated with the label "learning analytics" - use machine learning, neural networks, and similar software technologies to optimize learning processes. They evaluate the behavior of learners while working on a task in order to identify patterns of successful learners (Ferguson et al., 2016). The pattern of "ideal performance" is not derived from a theoretically or empirically based competency model, but from the behavior of many learners (on the Internet) whose learning outcome can be classified as more or less successful. An AI-based learning program can then compare the behavior of learners with patterns of successful individuals. The system can either control the progress of the content presentation by selecting and presenting certain information or tasks, or recommend their completion ("advice systems"). The goal of such recommender systems is to increase learning engagement and stimulate reflection. KERRES & BUNTINS (2020) analyzed the didactic design of existing systems in a review (see also Drachsler et al., 2015). Learners may receive the following types of cues in such systems, for example: [...]
This also reveals the application scenarios and teaching content for which these approaches are designed: Different behavioral parameters during the learning process are to be recorded and compared with a model of "ideal performance", on the basis of which feedback is given in the learning process. For adaptive systems, a theoretically based and empirically validated competency model describing ideal performance is required. In AI-based systems, this model is formed by observing individuals with high expertise in processing a task.