Initiative 4: Artificial Intelligence and Personalization
There have been great strides in the science and the technology of education over the last several decades. Scientists now have a deeper scientific understanding of the cognitive, social, and cultural processes of learning. For example, the role of personalization, formative assessment, and metacognition in learning are much better understood today.
On the other hand, computing technologies of the internet and social media are also transforming education—for example, the development of entirely new categories of programs, including the Georgia Tech Online Master of Science in Computer Science (OMSCS) and OMS Analytics degrees. Further, the combination of the new understanding of learning and the availability of online educational materials has led to broad adoption of pedagogies such as flipped classrooms and blended learning.
In the science of education, there will be continued progress in the cognitive and learning sciences, especially cognitive neuroscience. A generation from now, there should be a much deeper understanding of how the human brain processes information and how the human mind learns.
We can also expect rapid progress in technology, especially in artificial intelligence (AI), which is beginning to impact education in a myriad of ways, including intelligent tutoring systems and question-answering agents. It is likely that current movement toward scale and personalization will not only continue but also accelerate, with AI acting as the key accelerator, as detailed in the CNE Report Supplement Exploiting Artificial Intelligence (AI) for Personalized Learning at Scale (Georgia Tech 2018c).
The internet has enabled the commoditization of knowledge, with virtually every fact or concept ubiquitously available at the touch of a button. The ability to guide students through complex content domains, arrange experiences that allow them to apply their budding expertise, and provide effective feedback that enables them to refine and improve cognitive models cannot be so easily commoditized. To do this at scale and with a high degree of quality, AI will be necessary.
The “Jill Watson” experiment
In 2017, Georgia Tech began its third semester using virtual teaching assistants (TAs) in an online course, a year after “Jill Watson” was introduced in the class on Knowledge Based Artificial Intelligence, a core course taught by Professor Ashok Goel as part of the Online Master of Science in Computer Science degree program.
“Jill,” originally implemented on IBM’s Watson platform, answers frequently asked questions without the help of humans. In spring 2016, the students didn’t realize her identity until they were told on the final day of the class.
Recent results show that “Jill” has a personality: she is conscientious, optimistic, and resilient, three traits often associated with effective teaching. These findings also indicate that interactions with “Jill” enhance student engagement that is often strongly co-related with student performance. This is because “Jill” provides more timely answers to student questions.
Virtual teaching assistants as illustrated by “Jill” were recently recognized as one of the most transformative technologies to impact college within the past 50 years by The Chronicle of Higher Education (Myers and Lusk 2016).
The “Jill Watson” experiment, which utilized IBM’s Watson system as a basis for an artificially intelligent teaching assistant called “Jill”, was widely hailed as a breakthrough in both AI and educational technology.
But for all her success, “Jill” is a question-and- answer tutor. The opportunity now exists to augment “Jill’s” skills to handle other tasks that are associated with personalized learning, allowing advisors to go beyond scheduling and keeping students on the path of timely completion of degrees to become powerful partners in learning. A multifunction virtual tutor can be deployed to advisors, coaches, and even mentors located in a Georgia Tech atrium™, as described in Initiative 5 below.
Using such facilities, trained specialists can deliver personalized learning services. The platform for these services is an AI-enabled personalized learning system. Such a system must be able to answer questions effectively and with a human touch. The system must also help design formative assessments, be a cognitive tutor, and provide metacognitive tutoring capabilities. The development of such a tutor will require advances in human-centered AI and the ability to apply it to specific domains.
In the near term, AI-based platforms for mastery learning, which have been tested across subject matter and student populations for more than thirty five years, can be married to online and adaptive learning platforms. These platforms, due to their ability to provide flexible learning, remove time as a critical variable to learning. Georgia Tech will explore these effective learning methodologies to raise general levels of achievement to those normally associated with the upper 10 percent of learners.
The Commission recommends pilot projects to test appropriate adaptive learning platforms that can be customized by instructional faculty. Some of these experiments may include interactive books, interactive videos, and AI agents like “Jill Watson” for many Georgia Tech classes, especially large, bottlenecked, remedial, and/or online classes. Some of these adaptive learning platforms can also be transferred to pre-college and post-graduate education.
To keep ahead of fast-moving innovation elsewhere, the Commission recommends that Georgia Tech develop a multifunctional virtual tutor that can combine cognitive and metacognitive tutoring tasks normally associated with human teachers, such as coaching on open-ended projects and critical thinking development. Such a tutor can be available in the next two to five years. A multifunctional tutor will push the envelope on personalized learning by tailoring the kind of assistance to the needs of individual students, offering formative assessment and metacognitive tutoring based on individual progress, and providing contexts for other functions. For example, a “Jill Watson”–like AI agent operating as part of a virtual multifunctional tutor would be able to answer more complex questions about concepts taught by the cognitive tutors.
In the longer term (up to fifteen years), the development of a multifunctional virtual tutor fully capable of supporting personalized learning at scale would require interactive AI agents whose interactions with humans use cognitive models of humans and contextual knowledge to enhance the quality of the human-AI interactions. Human-centered AI focuses on developing interactive agents that can live, work, play, and learn with humans.