In ‘Be Right Back’, an episode in Charlie Brooker’s Black Mirror, a grieving woman finds herself able to communicate with her late boyfriend through a replica of his personality created through his social media footprint. Whilst this seems far-fetched, the genius of this and other episodes of the TV show lies in the fact that we are closer to this than we think. Machine learning is here, and it is only going to become more embedded into our lives. As educationalists, we need to work out what this means for our schools and colleges.
There is broad agreement that AI isn’t at the Skynet/Terminator/take-over-the-world stage just yet. It may be great at creating bespoke Netflix playlists or offering greater efficiencies to the Amazon supply chain, but ask it to differentiate between a puppy and a muffin and it struggles. That said, Elon Musk took to Twitter this week to say that AI is a greater world threat than North Korea, and Mark Cuban has admitted that it ‘scares the shit’ out of him. Clearly we need to proceed with caution, lest the singularity be upon us sooner than we think.
Thus far we have seen little inroad into education. This may be because there is less immediate return on investment and the school is a more conservative space that tends towards the status quo wherever possible. But AI is coming, so it’s important to start looking at a few options for how it might add value and allow teachers to do what they do best: to inspire, enthuse, and pass on a deep love of learning.
In their 2013 paper, Woolf et al list five ‘grand challenges’ to education and suggest how AI can provide a solution. I want to look at three of these, as I think these hold the key to how we might see AI impacting on education over the coming years.
1.Virtual mentors for every learner
We know through Hattie that quality feedback has the largest impact on learning, potentially adding more than two grades progress over a single year. However, it is challenging for teachers to give each learner the time and attention they need to offer tailored feedback, and we also know that students learn at different rates and respond in different ways to feedback.
But what if every student had their own AI mentor? We are already seeing this in business: Cogito are a company that help to improve customer service through AI bots that analyse thousands of hours of phone calls with customers, learn which interactions are most effective at achieving customer satisfaction, and use this learning to mentor newer or less effective employees.
It is not too great a leap to imagine every student having a similar bot to analyse their work and offer timely suggestions as to how it might be improved. The timely aspect is important here: feedback given a week or more after the work has been completed is far less effective than feedback that moves the learner on at the moment they need it. The more the bot learns the learner’s profile, the greater the impact the intervention should have. Cogito have seen customer satisfaction and sales rise by as much as 50% after AI intervention. The impact on learning could easily be as great.
2. Mining data to to better understand learners’ needs
Feedback doesn’t only flow in one direction. It’s also what the student, or indeed class, gives the teacher to indicate how well they are learning. A disruptive class is a form of feedback, indicating to the teacher that the lesson is dull, pitched to the wrong level, or both. For the teacher this is deeply valuable, creating opportunities to change direction even mid lesson when they realise they are losing the class. We’ve all been there.
AI can support teachers and institutions in honing their learning materials so that they are aimed at the sweet spot, neither too easy nor so challenging that learners give up. Woolf et al highlight two research communities that have begun using AI to address these challenges: learner analytics (LA) and educational data mining (EDM). Both use data in a number of ways, supporting learners in reflecting on their achievements, predicting which students need extra support, helping teachers to plan interventions, and improving courses and curricula (Woolf et al 2013).
Two examples of the latter are recently to market. Hubert.ai (still in Beta) chats with students through an instant messenger as they complete assignments, compiling feedback that can help teachers to improve their course materials. Zoomi is one of the first LAs to use AI to offer rich, real time feedback to course creators, analysing every interaction between student and programme and determining which materials are the most useful and impactful. Whilst this is more focused on business elearning at present, it surely can’t be long before we see this in schools.
3. Universal access to global classrooms
Woolf et al define this as ‘providing learning that is universal, inclusive, available anytime/anywhere, and free at the point of use’ (Woolf et al 2013). The rise of MOOCs in recent years is suggestive of a growing democratisation of access to learning materials from some of the world’s top learning institutions.
However, the dropout rate from these courses is high: on average, only 15% of enrollees complete them. If we are to create universal access to learning that truly adds value, we need to personalise this learning. These massive courses can have upwards of 20,000 participants: FutureLearn delivered a MOOC on IELTS preparation for some 440,000 students. It is no wonder that dropout rates are so high when students have no connection with either the course deliverer or other students on the course.
But what if the course was administered through AI? We have already seen how AI can already take masses of data and create unique experiences for individuals: there is no reason why this cannot be the case for online learning.
To go back to our AI bot that can offer bespoke feedback: it could also offer tailored learning resources, curated to perfectly match the learner’s needs, and could even suggest when in the day is best to learn certain subjects or topics (or indeed when in the learner’s career when they head into the workplace). The course would still rely on the human element, to set the right materials, ensure learning is well-paced, to offer accreditation for an additional fee. Humans are still better at the bigger picture.
That said, surely the time will come when even this can be curated by AI: a teacher could input the course objectives and expected outcomes, and the AI bot could sift through tens of thousands of documents and hours of video to assemble the ideal course to achieve those objectives, and the ideal way to check outcomes. However, this will have little effect if the student drops out anyway: AI’s ability to mine vast amounts of data and learn the reasons for this dropout might help course designers to create more ‘sticky’ courses that maintain higher retention rates.
This is only a start. Over the next few years we will see greater innovation in the education space, with entrepreneurs providing solutions to the growing need to provide low-cost, effective learning resources and innovative approaches to delivery. We may be a distance from the singularity, but that doesn’t mean we should ignore both the benefits, and potential dangers, of artificial intelligence.