There are plenty of ways in which machine learning is already being used in schools, but how might we best teach our pupils how machine learning works, and what its implications are?

There seems an ever-increasing interest in how we might bring AI (particularly machine learning) into schools, with promises of reducing teachers’ workload, personalising learning, or even of addressing the issues we face in teacher recruitment by getting machines to do at least some of the teaching for us. I’m far from convinced that teachers should, or even can, be replaced by robots anytime soon, but I suspect there are few of us who would mind a little help with the marking from time to time, or some helpful suggestions for what activities or resources might prove effective with individual pupils or some of our classes.

Using AI

Teachers, schools and education systems are starting to explore some of the ways in which machine learning can be used. Bringing Alexa, Siri or the (anonymous) Google Assistant into the classroom is one way, providing a response to questions from “What’s 37 x 28?’ through to ‘outline the causes of the first world war’. We’ve long used computers to help support assessment, but machine learning moves this beyond short response and multiple choice into automated essay grading; alongside this, machine learning applications of stylometry are helping to spot ‘contract cheating’, where a student buys an original piece of work off an essay mill in the hope of avoiding detection by traditional anti-plagiarism tools. Regression-based machine learning models can do a fairly sophisticated job of drawing on far more factors than SATs or MidYIS scores to predict outcomes. Platforms like Century and Maths-Whizz make impressive claims for their ability to create personalised learning pathways for individual students. Tracking tools allow students with individual difficulties to be identified early, including spotting unidentified SEN, and even Ofsted are using similar approaches to identify where changes in school performance seems to warrant an earlier rather than later inspection visit. IBM’s Watson has been used as a teaching assistant at Georgia Tech, getting much more positive evaluations than its human counterparts.

Why teach about AI?

As well as these and other uses of AI in education, a strong case can be made for bringing AI into the curriculum, teaching our pupils about AI: some of its uses, some of the principles and technologies on which it’s based, and some of the personal and societal implications of its increasing role. Informatics Europe and ACM Europe recently argued that,

In the near future, perhaps sooner than we think, virtually everyone will need a basic understanding of the technologies that underpin machine learning and artificial intelligence.

The House of Lords AI Select Committee recognised that children would need to be adequately prepared for working with and using AI. This means that all need to have the basic knowledge and understanding needed to navigate an AI driven world, including addressing questions about the ethical design and use of technology, and that some would need to go further still, receiving a thorough education in AI-related subjects.

Fortunately, the computing programmes of study in the national curriculum offer enough scope to address this. These have, as an aim, that pupils will be able to evaluate and apply information technology, including new or unfamiliar technologies, analytically to solve problems 5-7 year olds are taught to recognise uses of IT beyond school; 7-11 year olds learn to combine software to create systems to analyse and evaluate data and information; 11-14 year olds to undertake creative projects to analyse data and 14-16 year olds to develop and apply analytic problem solving skills.

What would an AI curriculum look like?

We might though want a more detailed framework for teaching pupils about AI. I think there are six broad areas that we might seek to address: using AI, understanding AI, considering the implications of AI, training machine learning systems, coding with machine learning models and developing machine learning applications. The first four of these could be at all key stages; by the end of Key Stage 2, pupils coding skills are well enough advanced for them to draw on machine learning models in their code; developing machine learning applications is a step up, but these could be highly motivating independent projects at A Level or EPQ, or indeed as a basis for an IB extended essay.

There are some great tools available now that allow pupils to start using, and learning about, AI themselves. Pupils can explore some common applications of machine learning such as image and voice recognition using tools such as Microsoft’s Seeing AI, Google’s Quick, Draw! and Google’s Voice Typing tools in Google Apps for Education. Google’s Teachable Machine is one of the nicest tools I’ve seen for introducing the idea of training a neural net to recognise images, and Slice of Machine Learning, also from Google, provides a good introduction to using decision trees for classification, in this case between pizza and not-pizza. MIT’s Moral Machine and the BBC’s ‘Will a robot take your job?’ allow pupils to start thinking through some of the ethical and societal implications of AI.

Scratch 3, out in January but already available in public beta, provides a number of extensions which draw on some elements of AI, including text to speech and machine translation, allowing pupils to make use of these in their own code. Dale Lane’s Machine Learning for Kids and Ken Khan’s eCraft2Learn work extend Scratch and Snap! by adding in blocks to make use of machine learning - Dale’s work makes of IBM Watson, providing an interface for pupils to collaborate in training their own models which they can then make use of in their Scratch projects. For sixth-formers wanting to take things to the next level, Kaggle’s competitions and data sets are great, and Microsoft offer free Azure hosting for machine learning, data science (and other) projects using the accessible Jupyter Notebook interface.

Whilst getting pupils acquainted with working with AI, and understanding some of the principles on which it’s based is enough for now, I wonder if we ought to start radically rethinking school curricula to better prepare our pupils for a future in which many jobs will be done by machines rather than people. What sort of education would best prepare our pupils for a future in which few, if any, might have jobs? I’d hope that human traits such as curiosity, courage and creativity would still matter in a future where machine intelligence dominates, and that such traits matter more than ever in our schools.