AI used for Education must be driven by teachers not inflicted on teachers

Teachers need to be central agents in deciding how AI is used in education. Perhaps we should talk about EdAI rather than AIEd to make this point. It is teachers who will be the orchestrators of when, and how, to use AIEd tools. In turn, the AIEd tools, and the data driven insights that these tools provide, will empower teachers to decide how best to marshal the various resources at their disposal.

Orchestrator1More than this, though, teachers – alongside learners and parents – should be central to the design of AIEd tools, and the ways in which they are used. This participatory design methodology will ensure that the messiness of real classrooms is taken into account and that the tools deliver the support that educators need – not the support that technologists or designers think they need.

Teachers who take part in these processes will gain increased technological literacy, new design skills, and a greater understanding of what AIEd systems can offer.


The increased introduction of AI-powered tools will serve as a catalyst for the transformation of the role of the teacher. AIEd is well placed to take on some of the tasks that we currently expect teachers to do – marking and record keeping, for example.
Freedom from routine, time-consuming tasks will allow teachers to devote more of their energies to the creative and very human acts that provide the ingenuity and empathy needed to take learning to the next level.

As this transformation takes place, teachers will need to develop new skills (maybe through professional development delivered through an AIEd system). Specifically they will need:
• A sophisticated understanding of what AIEd systems can do to enable them to evaluate new AIEd products and make a sound judgement about its value to them, and their students
• To develop research skills to allow them to interpret the data provided by AIEd technologies, to ask the most useful questions of the data, and to walk students through what the data analysis is telling them (for instance, using Open Learner models)
• New teamworking and management skills as each teacher will have AI assistants in addition to their usual human teaching assistants, and they will be responsible for combining and managing these resource most effectively


Most excitingly, with the evolution of the teacher’s role will also come the evolution of the classroom, as AIEd tools allow us to realise what it is unrealistic to expect any teacher or lecturer to do alone. For example, making the positive impact of one-to-one tutoring available to every child, or realising effective collaborative learning (a difficult activity to keep on track without some form of additional support).


This post is an adapted extract from Intelligence Unleashed published by Pearson.

Will Artificial General Intelligence (AGI) cope with ‘messy’ real world learning?

I was catching up with reading my weekend papers and came across the Observer Tech Monthly profile of Demis Hassabis, the founder of DeepMind. I have never met Demis, but the Observer piece echoes the descriptions I have been given by those who do know him. He is incredibly bright and also extremely modest: a nice ‘ordinary’ North London guy. I feel comforted that someone like this is at the forefront of our efforts to extend the boundaries of Artificial Intelligence (AI) and the achievements of DeepMind are certainly impressive. However I am less convinced about the real potential of Artificial General Intelligence or AGI, especially when it comes to its application to education.

download2I can buy into the vision of a world where smart people work with smart machines to solve wicked problems, such as cancer. And I can see that there is indeed too much information for many of us humans to process, so some artificially intelligent help would be great. I like the idea that AGI will “automatically convert unstructured information into actionable knowledge…. to provide a meta-solution for any problem” But that’s where it falls down for me, I can’t believe that the structured knowledge will be applicable to any problem.

IMG_8934The reason I hold this view is twofold. Firstly, much of the knowledge that helps us negotiate our way through the world is highly contextualized. There is significant evidence that a learner’s context impacts significantly upon their learning process and that in essence each individual person has his or her own individualized learning context. Secondly, teaching and learning in the real world provides extremely messy data. It’s this very messiness in teaching and learning settings that is crucially important. Partly because one never knows if what appears to be mess is actually important for learning. For example, a disagreement between two children will probably upset at least one of them and that in turn will impact on their learning. I would need to see some clear examples of contextualized AGI (is that a contradiction in terms?) and its propensity for messy real world learning settings to be convinced that AGI for education is a way forward.

hero-learning-pathsIt’s not just DeepMind whose remarkable systems are not to my mind suitable to take over education. I was on a panel with Jose Ferreira CEO of Knewton last month and it became clear that Jose believes that Knewton is smart enough to play the role of a teacher. It certainly is impressive technology. However, Knewton relies upon ‘clean’ data and that is not what classrooms are like. To my mind the most likely outcome for AI within Education is not for AGI, but for AI components to provide teachers with a selection of smart tools that teachers can use with learners as and when they think it appropriate. It really is the smart combination of Human and Machine intelligence that will win the day.


The Bett(er) the Learning Science, the more intelligent the technology

In my trip to Bett this year I was lucky enough to speak in the arena on the opening day when the place was abuzz. I spoke about the Learning Sciences and in particular the concepts of Metacognitive theory and theory of Deliberate Practice. The first of these concepts is about thinking about our own thinking. It can be thought of as Metacognitive Awareness and Metacognitive Control, and it is important for learning. Metacognition has been shown to be a better predictor of primary school children’s performance than IQ tests, and Metacognitive ability has been shown to have a positive impact on learning across the ages. The great thing about Metacognition is that it can be taught to metacognition-image-e1411763056642learners of all abilities, and in particular it can be taught by intelligent software as well as by human teachers. The second concept, that of Deliberate Practice confirms the adage that I was all too frequently reminded about as a child: Practice makes Perfect. Well, yes it does, provided that it is effort full, that the learner is motivated and that the practice is accompanied by appropriate feedback and the capacity for learners to see their own success. The qualities of practice just listed are within the skill sets of intelligent software as well as being in the skill sets of intelligent humans. Both human and machine are enriched by our knowledge of these concepts.

I was accompanied on the stage at Bett by David Levin the CEO of McGraw Hill . David charted his company’s development path from books, through e-books to adaptive software. This adaptive software enables each learner to take a personalized variable learning path through the educational material about his or her subject. The software tracks each learner’s progress and identifies areas of weakness so that help can be provided. For teachers, this type of software provides detailed monitoring information straight away and for the developers of the educational content used by the system there is also speedy feedback about what is working well and what needs redevelopment.


In addition to David’s presentation, I was really struck by the increased appearance of Adaptive Systems or software at Bett. These are systems that use Artificial Intelligence (AI) techniques to  support student learning by adapting to what the student needs: just like a skilled human tutor. dcqehohll3cy5ytjkdvqThis adaptation might, for example be to change the difficulty of the task the student has been set, or it could be to increase or decrease the extent of the helpful interventions that the software provides while the student is trying to solve a problem. This type of software uses AI to provide such tutorial individualization and I believe that AI may finally have achieved a position that will enable it to provide benefits for Education akin to those Siri has provided for personal assistance.

industry-40-and-its-technological-needs-12-638The adaptive software I saw at Bett was very like the adaptive systems that have been developed and studied by the Artificial Intelligence and Education community for several decades now. Indeed the whole topic of Artificial Intelligence (AI) is very much in the news at present, with much attention being paid to its role in the Fourth Industrial Revolution, in particular the way in which AI is replacing people in an increasing variety of jobs.


brainology-some-amazing-facts-you-didnt-know-about-your-brain_519a2c0a6a4c9_w1500One area that the AIEd community has paid increased attention to is the use of findings from the Learning Sciences to inform the design of AIEd systems. For example, psychologists, such as Carol Dweck, have spent many years developing a theory about motivation that is based upon the idea that some people have a growth mindset, whilst others have a fixed mindset. Those with a growth mindset believe that their abilities can be developed, but those with a fixed mindset believe that intelligence or learning ability is simply a fixed trait. There is a growing body of evidence that shows students’ mindsets play a key role in their success. And again, as with metacognition, students can be taught that the brain is like a muscle that gets stronger and works better the more it is exercised. They can also be taught that every time they stretch themselves, work hard, and learn something new, their brain forms new connections and that, over time, they become smarter. There is growing evidence to demonstrate that changing students’ mind-sets can have a substantial impact on their grades and achievement test score.

Human teachers can of course help learners to develop a growth mindset, and so can AIEd systems. In both instances: the human and the AI, the teaching is better for being informed by the Learning Sciences.



Well designed Artificial Intelligence can bring the ‘Internet of Thingking’ to learners and teachers.

I continue with the theme of Artificial Intelligence and how its tools and techniques can usefully be applied to education. The falling cost of technology combined with the increasing ubiquity and acceptance of Artificial Intelligence bring huge possibilities for building better software and hardware to support people to learn, teach, work and manage more effectively than ever before. The idea of an intelligent assistant for everyone is now a realistic possibility. Imagine how much more satisfying a teacher’s job might become if they had their own personal Artificially Intelligent teaching assistant who could, under their direction, take over teaching a group of students who need help with a particular area of the curriculum: quadratic equations in maths perhaps. Of course, for this to work, it’s essential that the design of such assistants is driven by the needs of teachers and learners and the principles of the Learning Sciences, rather than the capabilities of the technology and its developers. So we need to keep raising our voices for user centred design for EdTech.Girl manager, suitable for use in dialogs with other characters.


But the future of Artificial Intelligence within education can be about so much more than bigger and better systems, there are opportunities to offer something different to teachers and learners. For example, there are some interesting opportunities that arise from the continuing development of wearable computing and the Internet of Things: the network of objects or “things” with embedded computing systems, sensors and network connectivity that can be interconnected with any other network enabled objects or machines. If we think about the Internet of Things as a set of tools through which students can learn about other subjects, such as science or engineering, then we can use them to embrace and enhance the increasingly popular practice-based approach to teaching. Practice-based activities build on the popularity of the ‘Makers Movement’[1] and differ considerably in what they ask students to do and what they are trying to teach. They tend to be open-ended and hands-on, they involve collaborative problem solving processes and can include physical computing and the Internet of Things as tools for learning.


Internet of Things technologies will make it much easier for Artificially Intelligent systems to model the physical body and its movements. These models will then be available for use in educational systems that can be designed to help with tasks such as designing and constructing an artefact in engineering, setting up an experiment in science and simpler individual tasks, such as drawing or writing. The combination of  principles and techniques from the Learning Sciences, the Internet of Things technologies as learning tools and the application of Artificial Intelligence to support the learning process can ensure that the Internet of Things becomes the Internet of Thingking for learners and teachers.

[ File # csp2108410, License # 1307856 ] Licensed through in accordance with the End User License Agreement ( (c) Can Stock Photo Inc. / lenm

[ File # csp2108410, License # 1307856 ]
Licensed through in accordance with the End User License Agreement (
(c) Can Stock Photo Inc. / lenm


[1] Halverson, E. R., & Sheridan, K. M. (2014). The Maker Movement in Education. Harvard Educational Review, 84(4), 495-506.

More Intelligence Unleashed: Artificial Intelligence and Natural Man, we need people driven not data driven intelligence.

1980469Last week we held the 2nd in the series of events jointly hosted by London Knowledge Lab and Pearson. This event made me think back to when I first studied AI, some years ago, I loved the fact that it is an interdisciplinary subject that combines psychology, computer science, linguistics, philosophy …. I also loved the fact that even then it was full of promise. Coming from an Education background I could see straightaway that I wanted to know more about how AI and education could be combined. I joined the AIED community and enjoyed my research but there seemed little public appetite for AI with respect to Education. But now AI is everywhere and there is real interest in how AI can improve and support teaching and learning. And here tonight we are going to take a very practical approach and come up with ideas about how AI can address some of the important challenges within our education system. So how might AI enrich education?

Picture2For me the important thing is to look for ways in which artificial intelligence and human intelligence can be combined in complimentary unions.

For example, how about a system that collects data about classroom interactions between teachers and learners, learners and learners and learners and technology – a learner could carry this with them as as part of their persthinking caponal computing device as they attend different classes about maths, geography or English etc. The data collected by the system would be subjected to a series of AI analysis methods that would generate outputs that reflect individual learners’ metacognitive development and suggestions about how this could be improved. Teachers would concern themselves with teaching the specific subjects and the AI technology would link these together and provide invaluable information about one of those all important 21st Century skills.

Or how about taking advantage of the fact that there is now strong evidence to demonstrate the effectiveness of well designed one-to-one tutoring systems that use AI techniques to provide individualized tutoring to a learner. This kind of tutoring could be used to great effect with learners who are struggling and whose parents can’t afford to pay for a home tutor to make sure the child gets into a good school (or annex). It could be funded using pupil premium funds and might help to ‘level the playing field’ a little.


(Artificial)Intelligence Unleashed on Education: reasons to be cheerful, part 1

Tomorrow evening I am going to an event being hosted jointly between Pearson and the London Knowledge Lab. It will be the first of three seminars that explore the relationships between Artificial Intelligence and Education.


This first seminar poses several questions and I am discussing each of these over a few blog posts:
“Education is a key area in which AI is increasingly present in tools such as adaptive curricula, online personalized tutors, and teachable agents. So should we be worried? What do we know about how smart technologies work, and what might be realistically possible in the near and distant future? And how can artificial intelligence be best, and most responsibly, leveraged to support teachers in their work to improve outcomes for learners?”
Here are some initial thoughts about each of these:

So should we be worried?

It’s perfectly understandable to be worried about things we don’t really understand and since I suspect that most people don’t understand AI, then that suggests that most people may well feel at least a little apprehensive about AI in the classroom.The truth of the matter is that AI as applied to education is mainly done through building computer models of a particular curriculum, a way of teaching and of the learners who use the AI software. The computer models allow information about learners’ interactions with the AI software to be captured, analysed and used to predict what educational interactions would best suit each particular learner who is using the AI software. A good example of AI based educational software can be found at Carnegie Learning. Their Cognitive Tutors use a ‘model tracing’ approach whereby a subject expert is asked to provide a detailed account of the possible ways in which a student might successfully and unsuccessfully tackle the problems contained within a specified curriculum. The expert’s account is then used as the basis for a computer model of the possible solutions and errors a student might make. As the student progresses through each problem, their path is traced over the model in order to predict what their next steps might be and therefore how the tutor can offer appropriate support. cognitiveTutor_big

These adaptable or personalised software based tutors are not something that we should worry about. They work well for well defined areas of the curriculum, but they do not replace teachers, rather they are complimentary to teachers, because they can free up teacher time to spend on teaching and learning interactions that are not readily replaced by technology.

However there are worries associated with the large-scale uptake of such AI software IF it is seen by managers and administrators as a way of making efficiency savings, rather than a way of maximising the variety of teaching tools that are being used to support learners.

For anyone who wants the research on AI and Education their are several decades of work to be found in the AIED community in their journal and conferences.