I’ve been doing a lot of reading recently about the role of artificial intelligence in education for my subject Predict: Current and Future Trends in e-Learning. I approach the notion of AI with a fair amount of skepticism- over 20 years, I am yet to see an educational technology to fully realise its potential, and instead I’ve seen a lot of consultants and a lot of education technology companies make a lot of money off the back of gullible teachers and education departments. So, it is from that starting point that I engage with discussions about the incredible potential of artificial intelligence and learning analytics. The learning analytics part of this discussion is going to be covered in a fair amount of detail in Crunch: Learning Analytics for Performance Improvement, but there are parts that are relevant to this subject, too. One of the problems that I will face is the difficulty in making sure that I cover the information in sufficient detail within the time frame – AI is only one small part of Predict, so I think there will only be time for about 1 week’s work on the topic.
So, to being with, what are the promises made about Artificial Intelligence? There has been a move – towards practicality and sensibility if you ask me – in the last decade or so, away from the old notions of intelligent tutoring systems (that is, machines that actually undertake the teaching aspects) towards machines that inform teachers of specific characteristics and variables and allow teachers and learning designers to design better informed interventions to account for what the data is presenting. This is the artificial intelligence as computer assistant model. While this is eminently more practical than teaching robots and the like, there are still significant challenges in implementing any such model – not least of which is deciding what data is appropriate to collect, and how best to collect it.
One of the claims of this approach is that it might eventually make assessment – as in summation, test-based assessment – a thing of the past. The idea is that data can be collected on student performance almost continuously over time – rather than at any one point in time. Of course, the question then becomes, as Cope, Kalantzis and Searsmith write, in what ways do we represent learning – that is, what data do we collect and, if necessary, manipulate, in order to represent the learning that’s taking place. The danger, of course, is that some narrow identification of the data to collect will lead to a form of neo-behaviourism.
Having said that, there is also some interesting material about teaching robots – especially in places like China, as Hattie recently shared. In some cases, this is taking the form of Natural Language Processing – something that I will write more on later.