When you compare the typical 21st century classroom with that of the early 1900s, the differences aren’t terribly obvious. Teachers will be standing in front, giving instructions and sharing notes on a modern-day version of the old blackboard—say, an overhead projector or a shared computer display. Students will be sitting at their desks in the classroom or watching via online video-conferencing software. The technology has changed: A lot of the tools and processes have been digitized, some of it has been automated, and geographical barriers have been removed to some extent—but the actors and elements have remained much the same.
But thanks to advances in artificial intelligence (AI) and machine learning, a slow but steady transformation is coming to education, under the hood. In a few years, teachers will no longer be alone in shouldering the burden of training the young generation or the workforce at corporations.
Already, AI algorithms are helping enhance education by collecting, analyzing, and correlating every interaction that takes place in physical and virtual classrooms, and helping teachers to address the specific pain points of each student. This might be the beginning of a revolution in one of the oldest and most valuable social skills that mankind has developed, and an imperative in a world where humans live and work alongside smart machines.
Measuring Learner Progress
Instructors have to take into account every reaction to a lecture, every blank or attentive stare, every eager or hesitant response to a question, every assignment that is turned in early or late, and a lot more when assessing a student’s grasp of a concept. This is how they can find out where students are lagging and steer them in the right direction.
It is also why measuring a learner’s progress, an endeavor that is deeply social in nature, is one of the biggest challenges every teacher faces, and a task that is hard to accomplish with classic rule-based software.
“Course lectures, whether on a college campus or in a corporation, are predominantly one-size-fits-all, with the dominant mode being teachers speaking to students,” says Chris Brinton, Head of Research at Zoomi, an AI company that specializes in capturing and analyzing behavioral data in educational settings. “This is born out of necessity: it would be impossible, or at least inefficient from the standpoint of time, for the teacher to pause the lecture for prolonged periods of time and address each student concern individually to bring all to the same page. Instead, a student with many questions would usually be asked to follow up with the instructor outside of class time.”
However, machine learning algorithms, which are based on analyzing and finding patterns and correlations between data points, are proving to be an effective tool in helping teachers quantify a student’s understanding of a lecture.
“By analyzing specific student data, AI has the potential to help surface more quickly areas in which students may need more help, thereby improving student achievement and teacher support,” says Jessie Woolley-Wilson, president and CEO of DreamBox Learning, an intelligent math-learning platform.
Equipping the classroom with artificial intelligence is the equivalent of providing every student with a digital tutor, Brinton explains. “The algorithms driving AI can be trained to detect when a learner is struggling and what caused them to struggle, or when they are bored and what caused their boredom,” he says.
This is a shift from traditional learning software, which relied only on assessment responses to measure students’ grasp of the topics they study. “This data is often not available during a lecture, much less at the subsecond granularity in which a student may switch from a clear to confused point of view,” Brinton says.
There are now a number of AI-powered platforms that create rich digital profiles of each student by collecting live information from the user’s interaction with course material and context. In addition to keeping records of grades and scores, Zoomi, the platform Brinton helped develop, tracks micro-interactions such as viewing specific slides or pages on PDF documents, replaying a specific part of a video, or posting a question or answer on a discussion forum.
The data is then used to build a model that can give real-time insights into a student’s understanding and engagement with specific topics. Data models also help in finding common patterns among multiple students and performing predictive analytics, such as forecasting how students will perform in the future.
More advanced use of AI can involve the employment of complicated computer-vision algorithms to analyze facial expressions, such as boredom and distractedness, and link those to the other data gathered on students in order to create a more complete picture of a student’s learner model.
Finding and Addressing Gaps in Learning
There are multiple benefits to having a reliable digital model that represents a student’s knowledge. “The data can be used either automatically by an intelligent system to immediately engage students in learning experiences that specifically address those gaps in understanding, or by the teacher to identify—and respond to—those specific areas of need,” says Woolley-Wilson of DreamBox.
Third Space Learning, an online education platform founded in 2012 to provide one-to-one math tutoring, is now leveraging AI algorithms to help improve the performance of teachers. Since its launch, Third Space has recorded data about thousands of sessions. In partnership with University of College London, Third Space is now engaged in a project to mine the data with AI algorithms in order to find successful learning and teaching patterns and provide real-time feedback to its online tutors about how their students are keeping up with lessons.
The AI learner model can also power intelligent tutoring systems (ITS). Intelligent tutors, which can work in a self-paced learning environment or in conjunction with human teachers, use a student’s historical and real-time data to provide them with personalized content tuned to their specific strengths and weaknesses. Providing a personalized learning experience is a goal that teachers have always struggled to achieve.
“AI-powered tutoring systems have shown to be effective at teaching well-defined subject areas, such as maths and physics,” says Rose Luckin, Professor of Learner Centred Design at the University of College London Knowledge Lab. “AI can currently relieve pain points by helping with record-keeping and with the selection and recommendation of resources for learners to use.”
An example is MATHIA, a AI-powered math learning platform developed by Carnegie Learning that mirrors the behavior of human tutors. MATHIA collects various data points and employs machine learning algorithms and predictive models to determine students’ knowledge and skill levels and estimate their performance in the future. The platform uses this data to adapt the learning path according to students’ learning processes.
“Each step in a problem, which could involve filling in a cell in a spreadsheet, plotting a point on a graph, etc., is associated with one or more cognitive skills,” says Steve Ritter, Chief Product Architect at Carnegie Learning. “Depending on whether the student does the step correctly or not, or asks for a hint, we adjust our estimate of the student’s knowledge on the associated skills.”
MATHIA uses “knowledge tracing,” the process of determining a student’s understanding of different concepts, as well as “model tracing,” the process of understanding a student’s approach to solving problems, in order to adjust the software’s support for the individual student’s thinking process instead of redirecting them to a standard approach that may not make sense to them. This helps provide personalized content, with possibly countless learning paths.
“Our hints, for example, change based on the order in which students complete the problem steps, if this ordering reflects different ways of approaching the problem,” Ritter says.
The evolution of intelligent tutoring systems can eventually lead to a richer self-paced learning experience. While it won’t be a replacement for human teachers, AI-powered online learning platforms can play a pivotal role in making high-quality education available in areas where there’s a shortage of teachers, and students have to learn by themselves.
“The combination of big data and AI could provide learners with their own personal analytics, which they can leverage to become the most effective learner they can be,” says Luckin.
Self-knowledge (knowing what you do and don’t know) and self-regulation (for example, being able to stop yourself from being distracted by what someone else is doing) are two skills that such systems can help develop, according to Luckin.
“AI can be used to scaffold (support) learners to develop these key skills by reflecting back across their personal data using carefully designed interfaces and visualisations,” Luckin says. “In this way all learners could be helped to be better at learning, which would be useful across all subject areas.”
One of the benefits of AI-powered learning systems is the seamless assistance they can provide. “The same intelligent technologies that help students and their teachers inside the classroom should always be leveraged to do the same outside of the classroom,” says Woolley-Wilson. “They can bring the same power of personalized recommendations wherever the student is. Learning opportunities and access should no longer be restricted to a certain time or place as they’ve typically been in our analog past.”
Corporate training can also benefit from AI personalization. Zoomi, which provides online tools for professional training, uses AI algorithms to recognize learner preferences and dynamically adapt course content to meet their needs. For instance, based on a user’s past behavior and reaction to different media types, the platform can decide whether course material should be served in a PDF or video format. Progressive Business Partners has been using the platform since 2016 to train HR professionals, resulting in a 12-percent increase in course completion and a 30-percent increase in revenue.
Finding and Addressing Gaps in Teaching
When students lag behind in a lesson, flaws in teaching methods and curriculum are often as much to blame as weaknesses in the students themselves. Was the cause of student misunderstanding something about the material itself, the manner in which it was presented, or the timing of the material within the flow of the curriculum? Was it that the student had the flu when some necessary concepts were covered previously? How did the student engage with the material—actively or passively?
Those are some of the questions that every teacher has to answer when assessing the quality of a delivered lesson and investigating the root causes of problems in learning.
“Great systems can leverage huge data sets to assist teachers in finding both weaknesses in curriculum and in finding struggling students,” says Woolley-Wilson. “And it’s important to remember that the amount of help provided to the teacher depends on the quality of data available informing the analysis.”
DreamBox’s online adaptive-learning platform uses the data it collects from students to uncover learning gaps and then helps teachers to address them at the class level or for specific groups or individual students. This can include creating strategy groups, personalized learning plans, or focused assignments that address specific gaps and complement the core curriculum.
AI also helps teachers in assessing the relevance of their teaching material. “While the content is delivered ‘live’ in a classroom setting, most instructors prepare their materials electronically,” says Brinton, the researcher from Zoomi. “As a result, it’s possible for AI technologies to interpret the material, determine the topics covered, and even analyze the course assessment materials to gain insight [as to] how well the assessment covers the course content.”
Zoomi uses Natural Language Processing (NLP), the branch of AI that parses the content and context of written material, to weigh the quality of a teacher’s course material. Zoomi’s algorithms remove content that does not have a positive impact on the learning process. The company is also working on algorithms that augment the learning experience by finding complementary content and repurposing it to fit within the context of a particular lesson where a student is struggling.
“Soon, algorithms may be able to modify sentences for clarity, and even author new material on their own just as a human would do,” Brinton says.
Content Technologies, Inc (CTI), an artificial intelligence research and development company based in California, has developed AI that automatically generates customized educational content. CTI’s engine uses deep learning to ingest and analyze syllabus and course material, master the knowledge and generate new content such as custom textbooks, chapter summaries, and multiple choice tests. The technology is being used by a number of companies and educational institutions.
Education Will Remain a Social Experience
While we’ve seen impressive efforts in the application of artificial intelligence in education, the results pale in comparison to other domains where AI algorithms are causing major disruptions. The reason is that education and learning are fundamentally social experiences that are extremely hard—if not impossible—to automate.
“AI cannot replace teachers, because it has no self-awareness or metacognitive regulation, and it also lacks empathy,” Luckin, the professor from UCL Knowledge Lab. “However, AI, when its design is informed by what we know about learning and teaching (i.e. the learning sciences), can be combined with big data about learners to unpack the black box of learning and enable learners, teachers, and parents to track progress across multiple subjects, skills, and characteristics—this can provide vital information to support learners to become more effective as learners as well as to help them learn knowledge and skills.”
The augmentation and assistance that AI provides to the education and learning process will make teachers even more productive and efficient. “Teachers will be able to focus on what they can do best: create excellent content, deliver strong lectures, and address the most pervasive pain points both in person and remotely, individually and in groups,” says Brinton.
Another social aspect of education is collaboration. Students often learn more from working in groups and with each other as they do from listening to lectures and solving problems at their own pace. “The goals of education include more social interaction, such as learning to be a good collaborator or to communicate with others,” says Ritter, the product architect from Carnegie Learning. “So a challenge in personalizing instruction is to balance seeing a student as an independent learner who can proceed at his or her own pace with the need to work collaboratively with others.”
But AI might also become a facilitator in collaborative learning. Intelligence Unleashed, a joint research paper by UCL and Pearson, which Luckin coauthored, explains that AI can support collaborative learning by comparing student learner models and suggesting groupings in which participants are at a similar cognitive level or have complementary skills and can help each other out. AI can also take part in learner groups as a member and help sway discussions in the right direction by providing content, posing questions and providing alternative viewpoints.
The ubiquity of AI across the learning process will eventually revolutionize education. According to a Stanford University report, in the next fifteen years, it is likely that human teachers will be assisted by AI technologies that will result in better human interaction both in the classroom and in the home.
The classroom might remain more or less as it is today, but thanks to digital assistants, AI algorithms, and more capable teachers, future generations will hopefully have access to higher quality education and will be able to learn at a much faster pace.