Adaptivity and adaptation

Which complementarities in a learning personalization process?

Abstract

Within the iClass (Integrated Project n° 507922) and Elektra (Strep n°027986) European projects, the authors were requested to harness their pedagogical knowledge to the production of educational adaptive systems. The paper pinpoints and documents difficulties and limitations of this work as well as the possible fertility of such kind of interdisciplinary joint research. Identification of such pitfalls is a prerequisite for better interdisciplinary approaches in Technology Enhanced Learning (TEL) and safer design of usable and useful e-learning tools.
In both projects, it was assumed that e-learning technologies have a major role to play in providing such a personalised experience, the definition of personalized learning paths can possibly be left to three agents: the teacher, the learner and the machine. In this latest case, proponents of adaptive systems come usually from the technical side, they require from pedagogues that they provide rules deemed to inform this initial modelling and all aspects of the adaptive process so that the adaptation engine can determine the next step for the learning process. Unfortunately, in most cases, a systematic and automatic application of pedagogical principles turns to be disappointing, and even dangerous.
For the authors, the pedagogical added-value of adaptive tools is more likely to be found in the support of human decision-making regarding autonomy development and metacognitive training than in the provision of highly-technical automatic customization devices. The paper documents four reasons underpinning this position:
1. Pedagogy remains an unstructured field of problems
Those problems are characterized by an unlimited number of facts, features, and situations that produce a huge number of combinations and interactions, the effects and interplay of which on learners’ achievement and motivation is not clearly known. It means that adaptive systems are rarely self-sufficient and that the major challenge lays in their articulation with non adaptive components of the learning process. Forced to this acknowledgment, the iClass project decided finally to bend its initial automatic adaptive credo and accept that user's decision would compensate what the system could not thoughtfully decide by itself.
2. Rule's transparency as a condition of acceptance
The difficulty of creating automatic adaptive systems may not be the only reason that such systems are not found in every classroom. Legitimate resistance by teachers might be another one, underrated by adaptive systems proponents. If any teacher is to accept devolution of her teaching responsibility to a machine, she will be willing to understand how that individualisation occurs in order to accept it.
3. Pedagogical return on technological investment
Producing an hour of instructional material dedicated to an Intelligent Tutoring System is estimated to take between 300 and 1000 hours. As for the return on investment, namely an educational benefit resulting from personalization of learning obtained through adaptive systems, the question stays open.
4. The behaviourist tropism of the adaptive systems we experienced
On iClass and Elektra, LabSET worked on adaptive tools based on Knowledge Space Theory (Doignon & Falmagne, 1999), namely a skills-based cognitive engineering. From a pedagogical viewpoint, it confirmed that such systems are closely linked to a behaviouristic paradigm, limited in their scope and obviously premature to cater for the educational needs in respect to full blown personal learning.

Finally, the paper discusses the spectrum defined by two radical views on personalization: the "Summerhill personalization" (the student decides for everything) and the "Robocop personalization" (the student decides for nothing). It advocates for more development and research taking place in the median part of the spectrum, in the zone wherein adaptivity can support autonomy development.