In Qingtai county, a rural area of China, many K-12 students now have a non-human tutor. Sitting in front of a computer screen, an artificial intelligence (AI) system developed by a Shanghai-based company called Squirrel AI is helping them get up to speed with the school curriculum by identifying gaps in their knowledge and focusing on those areas. The students' scores on standardized tests generally improved, allowing them to achieve better results than many of their peers in the county.
Squirrel AI is gaining traction in China, where it now has over 2 million users, as well as plans for massive growth, but it's still early days for the use of AI in education, where many systems are still in research stages or being developed by start-ups and not yet available on a large scale. Although the Covid-19 pandemic has forced schools to transition to distance learning, which is likely to continue to some degree even after the health crisis ends, so far it has largely involved repurposing business tools such as platforms for group video calls and document sharing.
Few schools worldwide are using AI, although that could change in the next 10 years according to George Tilesch, a global AI consultant and author of Between Brains, a book about the current and future impact of AI. "I think AI systems will become the primary toolkit of educational systems," he says. "Ninety percent of the work will be done by machines, and 10%, focusing on the character development of the individual, will be done by teachers."
Platforms such as Squirrel AI, however, are appealing since they can personalize content. In a traditional classroom, it can be hard for teachers to cater to all levels of students, so high and low achievers often are left out. Furthermore, listening to a teacher talk is not effective for all learning styles, when some students retain knowledge better through hands-on activities or through visuals, for example. However, AI tools can choose topics and subtopics from the curriculum on which a student needs practice, based on initial assessments, and also can content in different formats, and even gamify lessons to cater to different types of learners.
Testing could also become more adaptive, so the same concepts are assessed in different ways. "We can leverage AI and data to generate assessments that are unique to [each] student and are more meaningful," says Jake Baskin, executive director of the Computer Science Teachers Association (CSTA). "It's not just everyone getting the same question on a test."
AI also has the potential to make education more accessible. Developing countries and rural areas often don't have the same educational opportunities as urban centers. A shortage of qualified teachers in many countries, including the U.S., is part of the problem. Tools available online could reduce inequalities between regions, as the use of Squirrel AI in Qingtai has demonstrated.
Tilesch say digitized curricula on AI platforms could easily be tweaked, as the skillsets needed to prepare students for future jobs change. "There is a significant mismatch in what schools are teaching and the expectations of enterprises or jobs in general," he says. "I think that's a massive problem that has been identified in the last two decades."
AI holds potential benefits for teachers, too. AI systems could take over administrative tasks, and the marking of assignments and tests, which takes up a lot of teacher time. "That's really going to make a big difference to the workload for teachers," says Danilo McGarry, an advisor to the European AI Alliance. "Teachers get burned out and there's a huge problem with retention of teachers nowadays, so hopefully that will make teaching more attractive."
Teachers may get to know their students better, as well. Especially when dealing with large classes, it can be hard to gauge how students are coping, and to identify their needs. AI systems will be able to gather large amounts of data, both about how students are performing in a specific subject and across the board. "I think there's going to be a lot more information that ends up getting collected and fed to the teachers," says McGarry. "It might step outside of pure education to monitoring the kinds of things [students] are watching or listening to."
There are concerns to address as well. Data bias can be a problem, illustrated by a grading scandal that occurred this year in the U.K. With exams cancelled due to the pandemic, an algorithm was used to determine the grades of students in their final year. It was thought that AI would give a fairer result than teachers but the algorithm, which took into account mock exam results as well as a school's past performance, gave many students lower grades than those suggested by teachers. It was also found to give a higher proportion of top marks to students from more prestigious schools.
There are ways to limit bias in algorithms. According to Tilesch, the system could have been better vetted by humans, for example by taking a closer look at the objectives it was using and whether they were appropriate. "In the case of this U.K. system, it has been put together in haste and there was not enough due diligence in the very beginning," he says.
Data protection is a concern as well. Large quantities of personal information about students will be stored online, which could be accessed not only by their teachers, but also by members of the school board, or even people at the companies producing the AI platforms. "Are these companies going to be advertising to kids; are they going to be collecting their data and using it for other purposes?" says McGarry. "This is something that has to be tightened up, especially for young children who are a bit more clueless about this stuff."
There is also a fear that students will miss out on certain skills if they are primarily being taught by a machine. According to the World Economic Forum, soft skills such as creativity, persuasion, and collaboration will be in increasing demand for future jobs, as automation becomes more widespread in the workplace. However, these skills are hard to acquire from AI systems. "AI is great at transmitting content, transmitting knowledge at scale, and gathering data and insights based on data, but teaching the skills that make humans better than machines is not in the system yet," says Tilesch.
As AI improves, that could change. Where machine learning systems look for correlations, AI is now moving towards extracting context, such as being able to perceive a user's environment and situation, which should make it slightly more human-like.
Teachers are likely to continue to play an important role in education. "I think there's incredible value in the relationship that teachers build with students, and I'm skeptical of our ability to entirely replace that," says Baskin. "It's essential that we're thoughtful in how we leverage AI, for teachers to do more of what they're best at, as opposed to trying to replace them overall."
Sandrine Ceurstemont is a freelance science writer based in London, U.K.