Changing What and How We Learn: The Future of AI and Education
AI and automation are predicted to result in up to 800 million jobs disappearing by 2030--at the same time, new jobs will be created on a massive scale. New technologies have always been disruptive, but the predicted impact of AI is unprecedented. Google CEO Sundar Pichai has gone as far as to say that AI “is more profound than … electricity or fire.” From Siri to medical diagnosis tools, AI is already in the building and it's here to stay.
How should we adapt? How should our schools and universities adapt? What skills and knowledge will serve us, the humans, in the 21st century?
1. Changing what we learn
In the time that you can solve a simple math problem (solve for x: 2x + 4 = 10), a computer can solve billions. So, are computing machines just that much smarter than you?
For the time being, AI conforms to something known as Moravec’s paradox: what humans find hard, computers find easy, and vice versa. In more concrete terms: robots can crunch billions of numbers, but will struggle to pick up a mug of coffee.
So while barista work is for the time being safe, AI has a deeper weakness. While the predominant paradigm of modern AI, Deep Learning, has gotten good at classifying things, the extent to which it 'understands' what it learns is far from deep. In fact, Shallow Learning may be apter ('deep' refers to a technical aspect of this technology, rather than being conceptually 'deep'). As AI pioneer Judea Pearl puts it: current AI may be able to classify, but it can’t understand why.
So in imagining a future where humans and intelligent machines coexist, we ought to think about how we can cooperate; instead of replacing our mortal minds, can the machines be thought of as extending them? AI pioneer Terrance Sejnowski refers to this usage of AI as 'cognitive appliances' in his new book 'The Deep Learning Revolution'.
So what is the future of work in a workplace infused with intelligent machines? And what is the role of education in preparing us for such a world?
2. Changing how we learn
So machines may influence what we learn, but can they change how we learn? In the 1920s, Sidney Pressey, an early cognitive psychologist, built one of the first ever examples of a 'teaching machine': a machine that implemented multiple-choice questioning. Pressey offered the following justification of how such a machine might augment standard teaching:
"The procedures in mastery of drill and informational material were in many instances simple and definite enough to permit handling of much routine teaching by mechanical means. The average teacher is woefully burdened by such routine of drill and information-fixing. It would seem highly desirable to lift from her shoulders as much as possible of this burden and make her freer for those inspirational and thought-stimulating activities which are, presumably, the real function of the teacher"
And rebelling against the then predominant behaviourist psychology of time, Pressey stressed the importance of engaging with misconceptions and using incorrectly answered questions as opportunities for "cognitive clarification", rather than the behavourist approach of "rote reinforcings of bit of learnings".
In the context of AI and education, Pressey's great insight was that the problem of instruction could be decomposed into tasks suited to 1) for scalable machine intelligence, and 2) flexible human intelligence. And although Pressey's approach to teaching wasn't perfect, this general heuristic is still useful today.
In more modern times, the idea of Intelligent Tutoring Systems became popular in the 1980s, making use of the incrementally growing powers of what we now call 'old-school AI'. This movement sought to further decouple learning from the classroom, and increasingly automate instruction. The Intelligent Tutoring Systems movement still has considerable influence today, being the intellectual foundation of educational software such as the popular language-learning platform DuoLingo.
What is the next frontier of machine-augmented education? Today, as the powers of AI continue to grow, it is likely we will be able to find more and more aspects of learning that can be automated by machines. For example, Natural Language Processing (the ability of machines to understand day-to-day language) has seen rapid improvement in recent years. This may one-day allow AI augmented assessment in schools and universities. Further, AI systems could one-day help with the generation of learning materials by dynamically producing questions tailored to a specific learner. More generally, schools and universities are often bogged down bureaucratically. Could machines help by automated aspects of this too, freeing up the time of academics and teachers? And yes, there may be robots in the classroom soon to help students with special needs.
But we must also be wary of ethical issues in the use of AI. As AI applications become increasingly ubiquitous, we are likely to become increasingly unaware of their presence and the power of influence they might exert, or the unchecked weaknesses they may bring. And perhaps more worryingly, one of the challenges in modern 'neural network' style AI architectures is the lack of transparency in how they operate, even to their creators. Related to this, such neural network systems can fall victim to any number of ethically questionable biases without their creators intending for this to happen. There is some progress in addressing these issues in AI design, but there is still a way to go.
So how can we effectively and ethically deploy AI to improve education? How can we ensure cross-talk between technologists and educational researchers? What are the most exciting and impactful applications we can be working on now? What are the risks that we need to plan for going forward?
What do you think?
We want to know what you're excited or concerned about. There is a lot of hype, but also a lot of promise. What research is currently going on? What ideas do you have? What are your opinions? And perhaps most importantly, what sorts of research needs to be done going forward? We don't just want opinions from technologists but from everybody.
We would also appreciate it if you filled out this short survey (~3mins) on AI and Education