In the age of algorithms and artificial intelligence (AI), the number of adaptive and personalised learning education products vying for educators’ attention and funds is immense. Using the latest advances in machine learning, big data and learning analytics, adaptive and personalised learning products can process a range of data, (educational, psychometric and social media), to map individuals quantitatively, compare them globally and generate recommendations for their learning potential and pathways. These techniques are being translated into impressive learning gains in the context of standardised curricula.
For example, LEAP Innovations, a Chicago-based research and piloting company found a 13 per cent increase in literacy using an AI-based reading product versus the control group. Laudatory testimonials from schools featured on product websites means it is easy to slip into a ‘black box’ mentality, focusing on the application of these technologies, without further unpacking the mechanics of how they produce the outcomes. However, by understanding the mechanics behind the production of profound learning insights and results that these products offer, educators can develop personalised learning strategies that make the most of the strengths and limitations of AI.
Opening the black-box: Artificial, adaptive and personalised learning
Knowing which questions to ask relies on understanding how these products are designed. AI, adaptive and personalised learning are the latest education technology buzzwords, but what distinguishes them? While there are many types of AI, they all involve some form of learning and applying knowledge to achieve a purpose. Learning and problem-solving is something that humans do very well, but sophisticated AI systems, along with huge data sets and computing power, can process and compute complex problems at a scale unfeasible for humans. AI systems can learn autonomously, which means we do not have to invest time and resources into writing code ourselves. A classic example of AI in education technology products is adaptive learning, which refers to the data analytics strategies that algorithms use to continuously learn about how individual students learn, and thus optimise learning sequences.1
Adaptive Learning
As students use adaptive learning software, they generate countless data points, which are recorded and fed to algorithms. These algorithms have been designed to learn from this data to better perform against designed criteria. An advance in a type of machine learning, using neural networks, has accelerated the functionality and potential of machine learning. Artificial neural networks are statistical models that are based on the way that the brain computes and processes information in non-linear, continuous and adaptive ways. Deep learning occurs when these artificial neural networks and learning algorithms observe and analyse raw data, learn, identify patterns and make predictions to determine optimum outputs.
Algorithm-based personalised learning
In the case of the adaptive learning education technology products, such as ALEKS and Dreambox, these algorithms have been designed to identify the ideal learning pathways for an individual student based on the criteria derived from neuroscience and cognitive-based learning and memory theories. These theories emphasise the importance of reinforcement and retrieval in short and long-term memory, and how this can be improved through interleaved and spaced learning and assessment. Every student’s ideal pattern of interleaved and spaced learning, forgetting, retrieval and reinforcement will be different. The algorithm’s or ‘machine’s’ capability of operating autonomously and writing its own code means it can learn how the student learns, what they have mastered and which areas they require improvement in. These insights are used to diagnose gaps in knowledge and skills, make predictions of the ideal learning pathways that will fill these gaps, and translate these into personalised recommendations. With the current computing power available, such inferences can be made in nanoseconds. The algorithms learn from data, and continuously teach themselves to get better. Thus, the more a student uses such software, the more data there is generated, the more the machine learns and the better it gets in making predictions and determining the ideal sequence of interleaved learning and assessment, i.e. personalising learning. Put simply, machine learning provides personalised learning for mastery.
More than personalised learning pathways
It is not just pace that can be personalised—more sophisticated products personalise according to learning preferences, complexity of tasks, students’ interests, and special needs. A few products enable educators to customise systems through gamification, using simulations and apps. Many products provide smart content services, using AI to process textbooks and resources into learning pathways and assessments for students, effectively designing digital courses for educators. For leaders and those desiring a macro perspective, some products offer aggregated analytics at the class, cohort and national level.
Critical Considerations for Adaptive Learning
There are many benefits these products provide that enable educators to justify the financial, time and human investment in adaptive and personalised learning technologies. The automation of administration, learning, homework and assessment can help to alleviate teacher workloads, saving time and money. Arguably, outsourcing the ‘boring’ ‘drill and kill’ parts of teaching to machines allows educators to spend more time improving the quality of instruction in other areas.
Teachers may then focus on developing students’ higher order thinking skills, or explore opportunities to enhance their own and students’ creativity, providing students with a variety of learning experiences beyond rote memorisation. Students’ outcomes can be boosted by the personalised and shorter feedback loops, leading to better retention and recall, efficiency and quality of learning. Thus, some schools are using AI products as part of a larger social justice and equality agenda. Adaptive learning platforms are used for lunch-time or after school homework clubs, targeting and assisting disadvantaged students who traditionally do not make the same rate of progress as their more affluent peers.
Datafication
There are several critical areas that have been discussed at length by practitioners and in academia, but what does this mean for educators and how they critically use adaptive learning products? One area requiring critical attention is the way in which such adaptive and personalised learning systems reinforce and promote the datafication of education, and thus a quantified relation to students.
Datafication pertains to how phenomena are converted into quantitative forms that are digestible by algorithms, producing certain ways of knowing and understanding education.2 There have been staggering advances in the range of data these products collect, crunch and convert into meaning for educators and students. These data points include what a student reads, how long for, the number of attempts per question, as well as emotional and biophysiological data.
Although we are aware of the biased nature of any data-collection and analysis, it is difficult to challenge the apparent, self-evident and objective story these metrics and visualisations tell us. Thus, it is important that educators continue to contextualise these metrics with their tacit and unquantifiable understanding and knowledge of their students and their learning processes. One-on-one conversations, long-term relationships and intuition are as immediately powerful, if not more so, than elaborate pie charts triangulating thousands of data points, when telling the story of our students and our practice. Hopefully these products will continue to evolve and incorporate softer and interpersonal forms of feedback.
Ethical Considerations
The capturing and analysing of data poses ethical issues for the education field as a whole. To what extent are educators willing to make decisions, or let machines make decisions, about students and their learning based on theories of evolutionary psychology, biological determinism-based genetics and correlation-based neuroscience? Although many initiatives to use biodata in education are research-based (Neuroscape and BrainLENS at UCSF), rather than commercial, this an area requiring educators’ input.
These products promote a quantified relation to the self with implications for how students know themselves as learners and their potential. Regular engagement with a quantified narrative of their learning and the subtle ways that these programmes narrate students’ activities can profoundly shape students’ academic self-concept, reinforcing intrinsic and extrinsic motivation in positive and negative ways.3 This calls into question how the outputs and user interface are modifiable by educators, and the interpretation of data within a broader view of what is important when reflecting on learning.
For example, in feedback and conversations with students, educators can emphasise the number of attempts and efforts a student made, or the amount of progress they made over a term. Educators can frame how students view the platform’s story of their learning, expanding the students’ perspective to include reflections on learning through different mediums and alternative ways of valuing learning. Many products include meta-cognitive reflection on learning habits, based on a plethora of data inputs. However, it is up to educators to foreground the importance of the nature of learning rather than summative outcomes achieved.
Ultimately, educators will determine the entire story students tell themselves, how much students invest in the story being told to them by these platforms and where necessary, challenge this with other knowledge. This can begin with a simple question, such as whether they and their students agree with what the adaptive learning platforms tell them about their understanding and what is missing in this narrative.
Gamification
The perils of gamification are another concern. Regular external reinforcement of self-efficacy may over-expose students to extrinsic sources of motivation. This may promote learning outcomes in the short-term, but undermine long-term aims of education to help students develop intrinsic capacity to motivate their learning.4
Narrow-Mindedness
A recurring critique of the UK education system is the emphasis given to narrow quantitative measures of education that promote certain subjects, such as the EBacc subjects, and pedagogies, such as teaching to the test, at the expense of balanced and holistic valuations and measures of education.5
Ofsted has now woken up to the distortive practices incentivised by narrow measures of success and are now prioritising the breadth and balance of the curriculum offered by schools. Given that many of these products serve mainly core subjects, educators must be careful that investments in these products, financially and time-wise, are part of an overall strategy to provide a broad and varied range of subjects and learning experiences for students.
Educators need to be mindful of what is and is not in the products’ curriculum. Although these products represent innovation in learning, they are not serving long-term goals of systemic education reform at the curriculum level. As well as under-represented subjects such as the arts, educators need to consider how these products promote or undermine valuable goals of developing the personal interests of students, their sense of curiosity, and their social and emotional development, physical and mental health.
As powerful actors such as Pearson develop international databases of trillions of data points and machine learning products, they increasingly determine the nature of education through data-driven algorithmic governance practices. A product’s tracking, anticipation and recommendations for students’ learning pathways are based on ‘norms algorithmically inferred from a global database’.6
Educators must constantly question what these ‘norms’ are, who decides them and why. The theories of learning and human behaviour that inform the design of these products should not be taken for granted as absolute positives. For example, currently these products provide powerful gains in personalised learning. However, personalising learning may not be such a worthy educational goal. Participative, networked learning is viewed as more relevant across personal and professional lives, across a range of industries.7
Many products promote individual, acquisition-based learning through isolating activities. These have their place in education, as many subjects require mastery in fundamental content and knowledge in order to progress to higher order thinking and application of knowledge in creative ways. However, if educators overemphasise the use of such products, they may be foregoing opportunities for students to develop interpersonal, collaboration and distributed cognition skills.
Key questions
There is much for educators to celebrate and embrace in the opportunities afforded by emerging AI and personalised learning technologies. Crucially, the desirability of the education visions offered by these must remain in question. These critical issues pertain to the use of current packages and those to come, that will increasingly recommend and determine education practices. Navigating these issues requires teachers and students to question all aspects of these technologies, and collaborative dialogue between stakeholders working in schools, researchers and the education technology industry.
Important questions educators can ask now, as they consider and implement these products are:
- What parts of learning and education do these products capture and convey well, and which do they not? Where do I as an educator need to provide the other parts of the story of my students?
- As we promote adaptive learning in the classroom or at home, are our students experiencing different mediums of learning, away from the virtual and screen?
- How are these platforms empowering my students and their capacity to learn? Are they developing independent learning skills that enable them to learn in all contexts with different online and offline resources?
- How is the extra time that these ‘effective’ and ‘efficient’ products save for educators, being re-invested in other areas of curriculum development to deliver a broad and balanced curriculum?
Harbir Kaur is currently a PhD student at the University of Warwick. She is a former science teacher, and policy and technology researcher.
Notes
1. Luckin, R., & Lakhani, P., (2018). The ‘No BS’ guide to AI. https://www.ascl.org.uk/download.5DA35907-F1E7-49F8-8960A31362F709C8.html
2. Williamson, B. (2016). Digital education governance: data visualization, predictive analytics, and ‘real-time’ policy instruments. Journal of Education Policy 31: 123-141.
3. Eynon, R. (2015) The quantified self for learning: Critical questions for education. Learning, Media and Technology 40: 407-411.
4. Hanus, M.D., & Fox, J. (2015). Assessing the effects of gamification in the classroom: A longitudinal study on intrinsic motivation, social comparison, satisfaction, effort, and academic performance. Computers & Education 80, 152-161.
5. OFSTED (2018). Curriculum research: Assessing intent, implementation and impact. https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/766252/How_to_assess_intent_and_implementation_of_curriculum_191218.pdf
6. Williamson, B. (2016). Digital education governance: Data visualization, predictive analytics, and ‘real-time’ policy instruments. Journal of Education Policy 31, 123-141.
7. OECD (2018). OECD The Future of Education and Skills 2030. http://www.oecd.org/education/2030/E2030%20Position%20Paper%20(05.04.2018).pdf