The online interactive magazine of the Association for the Advancement of Artificial Intelligence

By Nicola Capuano & Santi Caballe

In his annual survey, the learning technology expert Donald Taylor asks more than 2,000 industry experts from different countries to estimate the most popular topics in workplace learning1. Since 2017, adaptive learning has always been at the top of this ranking, barely overtaken by learning analytics only in 2020. From the higher education perspective, the EDUCAUSE Horizon Report 20182 included adaptive learning among the six most impactful educational technologies in the five-year horizon for higher education. This is confirmed by a recent survey3 where many chief academic officers consider adaptive learning as one of the most promising initiatives for improving the quality of student learning.

But, what’s it about? Adaptive learning refers to technologies that dynamically adjust to the level or type of course content based on an individual’s abilities or skill attainment, in ways that accelerate a learner’s performance with both automated and instructor interventions4. It is closely related to AI and is generally included in the area of Artificial Intelligence in Education. Its origin can be found in 1970s with the application of the first AI techniques, such as semantic networks and expert systems, to computer-based training systems to emulate aspects of human teaching. Nevertheless, the modern concept of adaptive learning was shaped only in the 2000s with the rise of online education and a growing attention of policy makers on the topic of personalized education that led to the first specific funding initiatives for this theme.

Adaptive learning systems enable the development of individual learning programs and supports students’ engagement so that their potential and success is maximized. They help in contexts with heterogeneous learners, i.e. when learners have different cognitive backgrounds and learning preferences. They are also useful when learning needs are not standardized but each learner has specific requirements, such as in vocational training, training on the job, and so on. They support both individualization as acquisition of skills through a diversification of teaching paths and personalization as achievement of cognitive excellence based on the specific intellectual potential of each learner.

From the methodological point of view, several results from AI have been used to build effective adaptive learning systems. Formal knowledge representation languages have been used to describe and organize the topics to be thought. Concepts from human-computer interaction, such as user modeling and adaptive hypermedia, have been adapted to the representation of students’ knowledge state and learning preferences. AI planning and optimization algorithms have been used for curriculum sequences generation. Classification techniques based on machine learning have been applied to data gathering during the learning process with the aim to predict training outcomes and prevent failures and dropouts in time.

Recent research advances in AI-based adaptive learning include conversational agents5 with natural language processing abilities, which have the ability to detect the student’s cognitive state during the discussions in course chats and forums and intervene accordingly to guide, support and animate students while engaging them in productive discussions on course topics. In addition, sentiment analysis techniques have been used to detect student emotions and react with the appropriate affective feedback while recommender systems have been employed to suggest additional learning material based on previous knowledge or similarity with other students. Finally, machine learning and group decision making techniques have been adopted for the automatic e-assessment of regular assignments, such as open-ended questions and also complex assignments, such as dissertations, mathematical proofs and computer programs.

In addition to research initiatives, there are companies, technology start-ups and academic publishers that now offer adaptive learning technology resources. For example, CogBooks6 distributes an adaptive learning solution for training providers that can be used stand-alone or integrated with third-party learning systems, such as Moodle or Blackboard. Smart Sparrow7 distributes a system with course adaptation capabilities based on instructor-defined rules and also supports emotion detection. Some long-established academic publishers are also entering this market. For example, the SmartBook Adaptive Learning8 solution distributed by McGraw- Hill Education leverages data collected during e-assessment sessions to adapt the learning experience based on student scores.

At this point, we can ask ourselves how far these technologies are really present today in the global panorama of education and training. Recently, the adoption of high-quality adaptive courseware has been strongly encouraged by a grant from the Bill & Melinda Gates Foundation, administered by the Association of Public and Land-grant Universities (APLU)9 involving eight public US universities across several disciplines. In addition, several self-funded pilot initiatives are taking hold around the world. For example, in 2017, National University in California launched the Precision Education Initiative10 to create a comprehensive educational environment integrating adaptive learning technologies. In the same year, Oregon State University collaborated with Smart Sparrow to build an adaptive online chemistry lab11. The University of Central Florida has recently made a strategic investment in the use of adaptive technologies to address students’ academic success and improve completion rates12. This only to mention some of the most recent initiatives.

However, while adaptive learning awareness in universities and colleges is higher than ever and numerous institutions have piloted this technology, large-scale implementations are still very few. In the new issue of the EDUCAUSE Horizon Report13 it is argued that one of the biggest challenges is the investment in terms of time, money, resources and vision needed to implement and scale these courseware products. Furthermore, it is necessary to find the optimal ways of integrating these technologies into the existing learning processes. Apart from a few isolated initiatives, currently it appears that most institutions are waiting and observing while early adopters share what they have learned.

Meanwhile, adaptive learning technologies are being asked to face new challenges coming from Massive Open Online Courses (MOOCs), that involve thousands of simultaneous participants, with some courses offered by Coursera and Udacity, just to mention some of the most popular providers, exceeding 100,000 registrants. Due to their scale, MOOCs introduce new technical and pedagogical challenges for adaptive learning. Indeed, MOOCs are intended to serve an indefinite number of participants with a high heterogeneity of contexts, profiles, and starting points. Furthermore, due to the high number of students and the relatively small number of tutors, their involvement during the delivery phases must be limited to the most critical tasks.

What could be the result of matching MOOC and adaptive learning? At the moment, this direction although promising is still underdeveloped. Some research projects are beginning to explore this field. Let’s wait and see, some interesting results will come out soon.

5. The European colMOOC project: