Education and Artificial Intelligence – critical optimism

We shouldn’t believe the hype or succumb to dystopian panic – it is vital to be simultaneously positive and reflective

Artificial Intelligence (AI) seems to be at peak media frenzy right now, with its effects on education frequently in the mix. My own reading in the last few days has included the first published output from Stanford University’s One Hundred Year Study on Artificial Intelligence, news of the American technology giants working together to address societal concerns around AI, and a review of Yuval Noah Harari’s new book Homo Deus, which follows on from his highly influential Sapiens to examine the future of mankind in the light of new technologies.

The attention is not new, even if the intensity and pitch seem higher. Education has long been touted as a sector where advanced machines can make a real difference – and as Audrey Watters regularly points out, this history pre-dates digital computers. “Personalisation at scale” is the holy grail of education technology, seemingly promising better learning with fewer resources. Education can be portrayed as an engineering problem, with the notion of “learning science” becoming prevalent and indeed key to some major corporations’ strategies and marketing literature.

The market is full of marketing, of course. Personally, I’ve become convinced of the promise of better analytics – providing aggregated and digestibly presented data to inform the decisions of students, teachers, parents and administrators. After all, the markbook (aka gradebook in the USA) has long been a key tool of the teacher in its paper and pencil form. However, I am instinctively sceptical the moment analytics become “predictive” or software starts to recommend or even enforce paths of action to/on people who are not empowered to understand, question or circumvent them. I am particularly dubious if a commercial vendor does not wish to explain the theory behind a software algorithm, and/or claims wide, generic application, and/or does not wish for their claims of learning improvement to be subjected to rigorous independent evaluation.

My knee-jerk reaction is clearly questionable. There will never be any clear boundaries between analytics, recommendations, or artificial intelligence – the moment you ask a question of data and choose how to present its results, you have prioritised that line of enquiry over many others. Some vendors have persuaded me – often via an enjoyable debate – that their algorithms and user experiences work effectively, but only for specific aspects of learning (for example, in delivering and reinforcing factual knowledge, but not in teaching critical thinking). There is thoughtful work being done on student retention, and interesting research taking place on evaluating adaptive technologies.

My concerns remain, however, based on the following:

  • Context is crucial in effective learning – and there are so many different contexts. There are plenty of commentators who talk as if a tool or technology is context-agnostic – a “magic pill”. Yet I have had murmurs of recognition pretty much every time I discuss what I call the “carrot and piece of string” issue: a truly great teacher can probably deliver a great lesson on any subject with a carrot and a piece of string, but a poor teacher will struggle even when equipped with every fancy tool available. This is why thoughtful implementations of technology in education focus on the contexts in which effective learning happens with the provided tool, and almost invariably include training (perhaps for both teachers and learners), evaluation and reflection as part of the rollout.
  • Computers can never capture all of the data which go into learning. In the past I have caricatured this as “the breakfast problem”: there was some research some years ago in the UK which (perhaps unsurprisingly) indicated that kids who didn’t have breakfast learnt worse during the day. I haven’t yet come across a system which asks kids about their diet, but even if one did capture this relatively simple data point, it would be impossible to gather the totality of the experience the child has brought to school that day and therefore to make the absolutely best recommendations for what and how (s)he should be learning. A teacher can’t capture all the data either, but they have many more things to work with, most importantly human empathy.
  • Misleading claims of scientific amorality. All algorithms are written by humans; and education is a world particularly freighted with values. Vladimir Putin may view – and teach – the recent history of Syria rather differently to me. The notion of “learning science” can be used to imply fixed, unarguable goals as much as it can be used to imply a critical, evidence-based approach to acknowledging the complexities of understanding and delivering “good” education.
  • Crass comparisons between education and other sectors, sometimes by people who should know better. Education is not retail or entertainment. To take one example, learning recommendations are sometimes compared to Amazon’s “you bought this, so you might want these things other people like you bought” functionality. Retail is binary, unitary and effectively irreversible. You bought something or you didn’t. (Yes, you can return an item, but you still bought it). Learning is none of these – you can partially understand a concept, you may need to learn other things before you can master it, and you can forget it too (just ask me to try and explain calculus).

Discussions of the role of AI in learning have been given fresh energy, complexity and relevance by the dizzying advance of technologies which can be described with the general term “AI”, and by the extreme, often dystopian, predictions for the future which have been inspired by this rise (cue Martin Ford and Harari). Deep learning, NLP, expression recognition et. al. may, it seems, lead to a future where there are many fewer jobs, irrevocably deep societal divides, or even computers which are more intelligent than humans and present us with an existential threat.

Written (and it seems extensively debated) by a panel of genuine experts, the Stanford report is a useful counterweight to the prophets of doom. According to the report,

“Contrary to the more fantastic predictions for AI in the popular press, the Study Panel found no cause for concern that AI is an imminent threat to humankind. No machines with self-sustaining long-term goals and intent have been developed, nor are they likely to be developed in the near future. Instead, increasingly useful applications of AI, with potentially profound positive impacts on our society and economy are likely to emerge between now and 2030, the period this report considers.”

Frankenstein, mercifully, seems some way off. But the very next sentence in the text shows its nuanced approach, another impact on education, and how we all now have to step up:

“At the same time, many of these developments will spur disruptions in how human labor is augmented or replaced by AI, creating new challenges for the economy and society more broadly.”

Note that there is a deliberate choice not to say that there will be less jobs – there is a clear acknowledgement of the uncertainty around this matter elsewhere – but it is clear that jobs will be different. Here and implicitly threaded through much of the document is a highlighting of choice – the decisions that we have to make, collectively and individually, as these technologies become more prevalent and more powerful.

What does this mean for those of us that work in creating new educational products, services and companies, for those of us that research, teach and/or manage in educational contexts, for learners, or for the interested citizen? I’d like to see the following emerging:

  • Transparency from education software developers about the inputs and outputs used and produced by their code. A comparison here would be the approach taken by Android and iOS to contact information, location data, or Facebook posting – the user has to authorise apps to access external data and functionality, and therefore engages with what the software is doing behind the scenes.
  • Digital education products and services which actively include and react to input from their users as their algorithms produce recommendations – empowering teachers, learners, administrators, parents, and more to influence the real-world results (and potentially even improve the software). The Stanford report is worth quoting again here:

“Design strategies that enhance the ability of humans to understand AI systems and decisions (such as explicitly explaining those decisions), and to participate in their use, may help build trust and prevent drastic failures. Likewise, developers should help manage people’s expectations, which will affect their happiness and satisfaction with AI applications. Frustration in carrying out functions promised by a system diminishes people’s trust and reduces their willingness to use the system in the future.”

  • A marketplace for personalised learning algorithms based on an open operating system for education, so that software can compete on the basis of its learning outcomes in particular contexts rather than other barriers to market entry. I’ve written at length about this in my last blog post.
  • Investment and growth in the ecosystem of companies and other organisations aiming to up-skill, re-skill and motivate the widest possible range of people as we face the “disruption” in the job market due to AI anticipated by so many forecasters. (This is a key area of investigation for me right now as I explore my next paid job, so please get in touch if you are working on something interesting!)
  • A sophisticated ongoing debate around the deployment of AI in education, particularly amongst those developing new tools. Conference organisers please have a think!

In short, I believe in a “critical optimism” approach. AI in education is critical – it is essential for our future. We also need to be critical – we must carefully examine and challenge the actions of ourselves and others, including major organisations. And we need to be optimistic as this exciting set of technologies reveals its possibilities and pitfalls.

Note: The Stanford report is here. Audrey Watters is here. However, nearly every sentence in this piece could have had a reference to evidence or research, so I’ve left all other references out to make for clean reading. Please reach out to me via Twitter @nkind88 or in the comments, and I’ll post links.

Image credit: yumikrum on Flickr (CC-BY 2.0)

Leave a Comment

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s