Enabling more fluid interactions between people and promising AI technologies also remains a critical challenge in Education, which has seen considerable progress in the same period. Though quality education will always require active engagement by human teachers, AI promises to enhance education at all levels, especially by providing personalization at scale. Interactive machine tutors are now being matched to students for teaching science, math, language, and other disciplines. Natural Language Processing, machine learning, and crowdsourcing have boosted online learning and enabled teachers in higher education to multiply the size of their classrooms while addressing individual students’ learning needs and styles.
The past fifteen years have seen considerable AI advances in education. Applications are in wide use by educators and learners today, with some variation between K-12 and university settings. Though quality education will always require active engagement by human teachers, AI promises to enhance education at all levels, especially by providing personalization at scale. Similar to healthcare, resolving how to best integrate human interaction and face-to-face learning with promising AI technologies remains a key challenge. Robots have long been popular educational devices, starting with the early Lego Mindstorms kits developed with the MIT Media Lab in the 1980s. Intelligent Tutoring Systems (ITS) for science, math, language, and other disciplines match students with interactive machine tutors. Natural Language Processing, especially when combined with machine learning and crowdsourcing, has boosted online learning and enabled teachers to multiply the size of their classrooms while simultaneously addressing individual students’ learning needs and styles. The data sets from large online learning systems have fueled rapid growth in learning analytics. Still, schools and universities have been slow in adopting AI technologies primarily due to lack of funds and lack of solid evidence that they help students achieve learning objectives. Over the next fifteen years in a typical North American city, the use of intelligent tutors and other AI technologies to assist teachers in the classroom and in the home is likely to expand significantly, as will learning based on virtual reality applications. But computer-based learning systems are not likely to fully replace human teaching in schools.
Today, more sophisticated and versatile kits for use in K-12 schools are available from a number of companies that create robots with new sensing technologies programmable in a variety of languages. Ozobot is a robot that teaches children to code and reason deductively while configuring it to dance or play based on colour-coded patterns. Cubelets help teach children logical thinking through assembling robot blocks to think, act, or sense, depending upon the function of the different blocks. Wonder Workshop’s Dash and Dot span a range of programming capabilities. Children eight years old and older can create simple actions using a visual programming language, Blockly, or build iOS and Android applications using C or Java. PLEOrb is a robot pet that helps children learn biology by teaching the robot to react to different aspects of the environment. However, while fun and engaging for some, in order for such kits to become widespread, there will need to be compelling evidence that they improve students’ academic performance.
Intelligent Tutoring Systems (ITS) and online learning
ITS have been developed from research laboratory projects such as Why-2 Atlas, which supported human-machine dialogue to solve physics problems early in the era. The rapid migration of ITS from laboratory experimental stages to real use is surprising and welcome. Downloadable software and online systems such as Carnegie Speech or Duolingo provide foreign language training using Automatic Speech Recognition (ASR) and NLP techniques to recognize language errors and help users correct them. Tutoring systems such as the Carnegie Cognitive Tutor have been used in US high schools to help students learn mathematics. Other ITS have been developed for training in geography, circuits, medical diagnosis, computer literacy and programming, genetics, and chemistry. Cognitive tutors use software to mimic the role of a good human tutor by, for example, providing hints when a student gets stuck on a math problem. Based on the hint requested and the answer provided, the tutor offers context-specific feedback.
Applications are growing in higher education. An ITS called SHERLOCK is beginning to be used to teach Air Force technicians to diagnose electrical systems problems in aircraft. And the University of Southern California’s Information Sciences Institute has developed more advanced avatar-based training modules to train military personnel being sent to international posts inappropriate behaviour when dealing with people from different cultural backgrounds. New algorithms for personalized tutorings, such as Bayesian Knowledge Tracing, enable individualized mastery learning and problem sequencing.
Most surprising has been the explosion of the Massive Open Online Courses (MOOCs) and other models of online education at all levels—including the use of tools like Wikipedia and Khan Academy as well as sophisticated learning management systems that build in synchronous as well as asynchronous education and adaptive learning tools. Since the late 1990s, companies such as the Educational Testing Service and Pearson have been developing automatic NLP assessment tools to co-grade essays in standardized testing. Many of the MOOCs which have become so popular, including those created by EdX, Coursera, and Udacity, are making use of NLP, machine learning, and crowdsourcing techniques for grading short-answer and essay questions as well as programming assignments. Online education systems that support graduate-level professional education and lifelong learning are also expanding rapidly. These systems have great promise because the need for face-to-face interaction is less important for working professionals and career changers. While not the leaders in AI-supported systems and applications, they will become early adopters as the technologies are tested and validated.
It can be argued that AI is the secret sauce that has enabled instructors, particularly in higher education, to multiply the size of their classrooms by a few orders of magnitude—class sizes of a few tens of thousands are not uncommon. In order to continually test large classes of students, automated generation of the questions is also possible, such as those designed to assess vocabulary,86 wh (who/what/when/ where/why) questions, and multiple choice questions, using electronic resources such as WordNet, Wikipedia, and online ontologies. With the explosion of online courses, these techniques are sure to be eagerly adopted for use in online education. Although the long-term impact of these systems will have on the educational system remains unclear, the AI community has learned a great deal in a very short time.
Data sets were collected from massive scale online learning systems, ranging from MOOCs to Khan Academy, as well as smaller scale online programs, have fueled the rapid growth of the field of learning analytics. Online courses are not only good for widespread delivery but are natural vehicles for data collection and experimental instrumentation that will contribute to scientific findings and improving the quality of learning at scale. Organizations such as the Society for Learning Analytics Research (SOLAR), and the rise of conferences including the Learning Analytics and Knowledge Conference and the Learning at Scale Conference (L@S) reflect this trend. This community applies deep learning, natural language processing, and other AI techniques for analysis of student engagement, behaviour, and outcomes.
Current projects seek to model common student misconceptions, predict which students are at risk of failure, and provide real-time student feedback that is tightly integrated with learning outcomes. Recent work has also been devoted to understanding the cognitive processes involved in comprehension, writing, knowledge acquisition, and memory, and to applying that understanding to educational practice by developing and testing educational technologies.
Challenges and opportunities
One might have expected more and more sophisticated use of AI technologies in schools, colleges, and universities by now. Much of its absence can be explained by the lack of financial resources of these institutions as well as the lack of data establishing the technologies’ effectiveness. These problems are being addressed, albeit slowly, by private foundations and by numerous programs to train primarily secondary school teachers in summer programs. As in other areas of AI, excessive hype and promises about the capabilities of MOOCs have meant that expectations frequently exceed the reality. The experiences of certain institutions, such as San Jose State University’s experiment with Udacity, have led to a more sober assessment of the potential of the new educational technologies.
In the next fifteen years, it is likely that human teachers will be assisted by AI technologies with better human interaction, both in the classroom and in the home. The Study Panel expects that more general and more sophisticated virtual reality scenarios in which students can immerse themselves in subjects from all disciplines will be developed. Some steps in this direction are being taken now by increasing collaborations between AI researchers and researchers in the humanities and social sciences, exemplified by Stanford’s Galileo Correspondence Project and Columbia’s Making and Knowing Project. These interdisciplinary efforts create interactive experiences with historical documents and the use of Virtual Reality (VR) to explore interactive archaeological sites. VR techniques are already being used in the natural sciences such as biology, anatomy, geology and astronomy to allow students to interact with environments and objects that are difficult to engage in the real world. The recreation of past worlds and fictional worlds will become just as popular for studies of arts and other sciences.
AI techniques will increasingly blur the line between formal, classroom education and self-paced, individual learning. Adaptive learning systems, for example, are going to become a core part of the teaching process in higher education because of the pressures to contain cost while serving a larger number of students and moving students through school more quickly. While formal education will not disappear, the Study Panel believes that MOOCs and other forms of online education will become part of learning at all levels, from K-12 through university, in a blended classroom experience. This development will facilitate more customizable approaches to learning, in which students can learn at their own pace using educational techniques that work best for them. Online education systems will learn as the students learn, supporting rapid advances in our understanding of the learning process. Learning analytics, in turn, will accelerate the development of tools for personalized education.
The current transition from hard copy books to digital and audio media and texts is likely to become prevalent in education as well. Digital reading devices will also become much ‘smarter’, providing students with easy access to additional information about the subject matter as they study. Machine Translation (MT) technology will also make it easier to translate educational material into different languages with a fair degree of accuracy, just as it currently translates technical manuals. Textbook translation services that currently depend only upon human translators will increasingly incorporate automatic methods to improve the speed and affordability of their services for school systems.
Online learning systems will also expand the opportunity for adults and working professionals to enhance their knowledge and skills (or to retool and learn a new field) in a world where these fields are evolving rapidly. This will include the expansion of fully online professional degrees as well as professional certifications based on online coursework.
Broader societal consequences
In countries where education is difficult for the broad population to obtain, online resources may have a positive effect if the population has the tools to access them. The development of online educational resources should make it easier for foundations that support international educational programs to provide quality education by providing tools and relatively simple amounts of training in their use. For example, large numbers of educational apps, many of them free, are being developed for the iPad. On the negative side, there is already a major trend among students to restrict their social contacts to electronic ones and to spend large amounts of time without social contact, interacting with online programs. If education also occurs more and more online, what effect will the lack of regular, face-to-face contact with peers have on students’ social development? Certain technologies have even been shown to create neurological side effects. On the other hand, autistic children have benefited from interactions with AI systems already.
Lucubrate Magazine, Issue 51, December 28th, 2018
The photo on top: phonlamaiphoto
The article is the chapter called “Education” of the report: “One Hundred Year Study on Artificial Intelligence (AI100),” Stanford University, accessed August 1, 2016, https://ai100.stanford.edu. The Lucubrate editor has added on the pictures for the article.
Categories: Magazine, Education, Artificial Intelligence