It is a time of remarkable transformation for education. Everyone recognizes the need to improve teacher effectiveness, to improve student engagement, and to create a twenty-first century education system that maximizes potential of every student. The challenges that must be addressed to make these improvements greatly exceed the scope of any single approach, whether it is educational technology, improved teacher training and better after school programs, and so on. In past research, AI--with its inextricable links to cognitive science, psychology, and mathematics--has proven a close fit for many of these challenging educational problems.
Educators have long known that the most effective teaching method is one-on-one tutoring. Ever since Benjamin Bloom's famous study, (1) education researchers have aspired to mimic the holy grail of one-on-one tutoring--to achieve a two-sigma improvement in student learning. Early AI researchers saw this as an opportunity to build intelligent tutoring systems (ITSs) that could adapt and tailor instruction to the individual needs of the student. Although today's systems fall short of the full two-sigma effect of a human tutor (roughly equivalent to two grade levels), Intelligent tutors have demonstrated remarkable progress in that direction. In fact, researchers have struggled to replicate the two-sigma effect suggesting that ITSs may already be as effective as human tutors. This suggests that the contributions of AI to education are perhaps more profound than previously believed and leads us to wonder why AI-based learning technologies are not in every classroom, every home, every library, and on every mobile device.
Today, ITSs are in widespread use in K-12 schools and colleges and are enhancing the student learning experience. As a specific example, 600,000 students in more than 2600 middle or high schools use Carnegie Learning's Cognitive Tutor mathematics courses regularly. Some full-year evaluation studies of Cognitive Tutor algebra have demonstrated better student learning compared to traditional algebra courses. Debate has ensued, however, involving both the efficacy of intelligent tutoring as well as whether the ITS approach can scale up to meet the broad needs of educational systems throughout the world. More broadly, studies have consistently shown that when a system models the knowledge it seeks to teach and uses that knowledge to assess, track, and scaffold learning, that it is more effective than non-intelligent counterparts.
We face unprecedented challenges in science, technology, engineering, and mathematics (STEM) education. Projections on the availability and needed competencies of STEM workers are disconcerting: current workers with STEM training are retiring and not enough of the younger generation will either be available or adequately trained to meet emerging needs. Recent reports suggest important links between education and health, civic involvement, criminality, and even eligibility for military service. Although these challenges transcend STEM education, increasing student interest and preparing students for STEM remain key challenges for the future of education systems throughout the world.
The goal of the special articles in the fall and winter issues is twofold; first, to present some of the best work at the intersection of AI and education in a way that highlights the power of AI to promote human learning; and second, to define the needs and challenges facing STEM education today and align those definitions with modern and emerging AI research.
The idea of putting these issues together and many of the articles included here originated from the first cyberlearning summit sponsored by the National Science Foundation. (2) The topics cover a significant cross section of AI and, in our opinion, represent some of the best interdisciplinary research in education today. Articles cover topics including uncertain reasoning, natural language processing, data mining, knowledge representation, explanation systems, automated reasoning, and more. …