Coronavirus and the future of learning: What AI could have made possible

Young girl with glasses, icons on head signifying artificial intelligence that could be used during covid-19 coronavirus

By Stéphan Vincent-Lancrin

Senior Analyst, OECD Directorate for Education and Skills

Before COVID-19 and the efforts to minimise its dire consequences became what we want to talk and read about, artificial intelligence (AI) was very much the flavour of the month – as a buzzword referring to anything digital capturing and using data in a “smart” way. It’s hard to believe that all the shouting about AI was just a few weeks ago…

Education and learning have been at the centre of many relevant AI conferences in recent months. Watch, for example, “Behind the scenes of AI” that took place during the Finnish presidency of the Council of the European Union. The coronavirus crisis has made the importance of digitalisation and AI in education even more obvious. AI-powered systems could have helped teachers, students and parents navigate the range of digital learning resources out there if they were more available and ready to use.

How could AI have supported learning during the coronavirus crisis?

One of the most advanced fields in digital education technology is the “personalisation of learning”. Personalisation is, in some ways, the holy grail of education. Imagine an education system where students could do exactly the kinds of tasks they need to make progress and overcome any learning difficulty they may have. In fact, that’s what teachers try to do in their classrooms, but they have many constraints – that is, many students with different levels of knowledge, interests and difficulties. On top of this, they also have a local or national curriculum that they have to teach.

Digital technology has made tremendous progress in the design of “personalisation” solutions based on students’ knowledge and learning difficulties. Granted, it still needs improvements, but three types of system would have been very helpful during this current crisis, had they been more widespread and advanced:

  • Systems that help teachers, students and parents navigate the existing digital resources freely available on the Internet as well as commercial ones.
  • Systems that help teachers give students the right types of tasks based on their knowledge and difficulties in their subjects.
  • Systems that help students with their homework in different domains.

The underlying principle of all digital personalisation systems is more or less the same. They first identify what students know and where they face difficulties; they make a (more or less accurate) diagnosis of what students should be doing next (for example improving their understanding of certain types of fractions); they recommend pedagogical actions, such as continuing to work on a specific difficulty or moving on to other kinds of tasks and problems. Most of the time, they do it within a discipline (for example, literacy or numeracy), but the same ideas could be applied at the overall curriculum level, for different contexts (work at school or at home) and taking a variety of information into account (knowledge, interest, motivation, etc.). In fact, developers are working on a range of different systems.

Do these systems actually improve learning?

Digital personalisation systems help teachers and other “educators”, including parents, do what they routinely accomplish by other means. Teachers and parents try to keep some balance between activities like reading, maths, physical activity and cultural activities so that children remain engaged, make progress and get a holistic education. Like in a video game, this sometimes involves a lot of practice and repetition. AI systems do the same, but they have more data (or observations) to crunch, more exercises they can potentially provide and generate. In some schools, in some countries, those digital systems are in use – although I am not aware of any place where they are mainstreamed. Imagine how these systems could have helped teachers, students and parents to know what to do while studying from home when schools (and universities) are closed.

I know what your next question is: do these “personalisation” solutions work? Well, it depends. There is evidence that some of these systems do actually work which is good news. For example, researchers have shown that some personalised solutions designed for maths homework did actually improve students’ learning – and they have done it with the highest levels of scientific evidence.

The recommendations of your favourite shopping, movie or music digital platform do not always work better than those of your friends, favourite magazines or website, do they? They may help nonetheless.

AI systems providing recommendations about what to read, watch or listen to are largely based on your “profile” – mainly what you have read, watched and listened to before. Educational AI systems have to answer more difficult questions as they try to make you move to the “next level” based on a diagnosis of your skills. Your favourite movie or music platforms, on the other hand, are agnostic about “level”: they are designed to ensure you keep buying or don’t cancel your subscription. Educational AI systems also have to answer a much narrower question, for which there is sometimes a “right” answer: can you solve multiplications of that difficulty? Can you read (and understand) text of that level of difficulty? Do you recognise these syllables? Profiles and recommendations can more easily be sophisticated and accurate as the problem is narrower, even though the general education problem is much more difficult.

Imagine how digital personalisation systems could have helped teachers, students and parents to know what to do while studying from home when schools (and universities) closed.

A new digital divide?

Another question pertains to inequity (equity deserves another article in itself: hopefully it will come). AI and other digital systems are just tools augmenting our human capacity, so what we achieve with them really depends on how we use them. Could educational AI systems and personalisation increase inequities? They certainly could if, after the coronavirus crisis, they become highly-priced commercial services affordable to a small share of the population, they will induce a “new digital divide”.

But let us be more optimistic. In the current COVID crisis, who would have benefited the most from those tools? Children from disadvantaged backgrounds: those whose parents do not have the cultural capital to support them; those whose parents’ energy is fully mobilised by making ends meet; those whose parents keep our supermarkets, hospitals, and delivery shops open. This includes parents whose parents cannot telework and just have to stay at home. Although in some countries the gap between children from advantaged and disadvantaged backgrounds has decreased over time (see this study), it starts from birth: on average, children from lower socio-economic backgrounds will rarely be exposed to the same amount of vocabulary, of nudging towards “correct” grammar, good arithmetic, ambition, etc.

AI systems do not care about your social background and, at least in principle, they can provide less-privileged children with the vocabulary, syntax, maths, widened interests, and the training that more privileged parents offer their children. Less-privileged children have much more to gain from these digital systems than privileged ones given that their parents’ socio-cultural status give them all they need to thrive in our societies.

AI and digital systems built by human beings to augment our intelligence are not perfect – but they should and will improve over the next decades. As the coronavirus crisis reminded us, we, human beings, are not perfect either – individually and collectively. However, as we try to improve and to learn, we should critically mobilise all the tools and wisdom of the past and of our time. That’s what education is about.

Read more: