What are the links between neuroscience and Artificial Intelligence when it comes to learning?
The focus of both neuroscience and AI is to understand how the brain works and thus predict behaviour. And the better we understand the brain, the better designs we can create for AI algorithms. When it comes to learning, the neuroscience - AI partnership can be synergistic. A good understanding of a particular learning process by neuroscience can be used to inform the design of that process for AI. Similarly, if AI can find patterns from large data sets and get a learning model, neuroscience can conduct experiments to confirm it.
Secondly, when neuroscience provides learning behaviours to AI, these behaviours can be translated into digital interactions, which in turn are used by AI to look at learning patterns across large numbers of children worldwide. The power of AI is that it can scale this to large numbers. AI can track and search through massive amounts of data to see how that learning happens, and when required, identify when learning is different or goes off track.
A third feature is that of individualized learning. We increasingly also know that learning has a strong individual component. Yet our classrooms are structured to provide common learning to all children. Sometimes these individual differences become crucial to bring out the best in children, which is when we might tailor learning. Neuroscience research on individual differences has shown that detailed information on that individual can reveal a wealth of information about their learning patterns. However, this is extremely cost and labour intensive. Yet, this detailed learning from neuroscience can be provided to AI in order to scale. AI can collect extensive detailed data at the personal level, to design a path to learning for that child. Thus, what neuroscience can study in small groups, AI can implement in large populations. If we are to ensure a world where every child achieves full potential, such personalized learning offers a great promise.
How do we create a structure around AI to ensure learning standards globally?
One thing AI capitalizes on and constantly relies on is large volumes of data. AI algorithms perform better if they are being fed by continuous distributed data. We need to keep in mind that humans are the ones designing these algorithms. This means that the algorithms will only do as well as the data that they have been trained on. Ensuring that we have access to large amounts of data that comes from various situations of learning is crucial. What sometimes becomes an issue for AI algorithms is that most of the training data has been selected from one particular kind of population. This means that the diversity in the forms of learning is missing from the system.
To return to reading and literacy as an example, in neuroscience, a large part of our research and understanding of how the brain learns to read has come from individuals learning to read English and alphabetic languages. However, globally, billions of people speak or read non-alphabetic languages and scripts that are visually complex, which are not really reflected in this research. Our understanding is built on one particular system that does not have enough diversity.
Therefore, it is important that AI algorithms be tested in varied environments around the world where there are differences in culture. This will create more robust learning models that are able to meet diverse learning requirements and cater to every kind of learner from across the world. If we are able to do that, then we can predict what the learning trajectory will look like for children anywhere.
Human beings have similarities in the way they learn, but pedagogies vary across different situations. In addition, those differences must be reflected in the data provided. The results would be much more pertinent if we are able to capture and reflect those differences in the data. This will help us improve the learning of AI, and ultimately understand how the brain works. We would then be better suited to leverage the universal principles of learning that are being used across the world and effects that are cultural in nature. That is also something that we want to hold on to and capitalize on in trying to help children. People designing AI algorithms so far have not given a lot of attention to this, but they are now beginning to consider it in many places across the world.
How do you see AI’s role in inclusive education today, especially in the context of migration?
Societies have become multicultural in nature. If you go to a typical classroom in many countries, you will find children from diverse cultures sitting in the same learning space. Learning has to be able to meet a variety of needs and must become more inclusive and reflect cultural diversity. Innovative pedagogy such as games, interactive sessions and real-life situations are key because they test learning capabilities focused on skills that children should acquire. AI relies on digital interactions to understand learning and that comes from assessing skills and behaviours. We now recognize that what we need to empower our children with are skills and behaviours – not necessarily tons of information.
Digital pedagogies like interactive games are among the ones emerging rapidly to assess children’s skills. They are powerful because they can be used in multicultural environments and can assess different competencies. They are not necessarily tied to a specific language or curricula but are rather performance-based. How do you assess children for collaboration in a classroom? In the context of migration and 21st century skills, these are necessary abilities and digital games provide a medium to assess these in education. When such interactive games are played by children across the world, they provide digital interactions to AI. AI might discover new patterns and ways to collaborate since children have ways of doing things that are often out of the box. A skills-based approach can be applied anywhere, whether it is in a classroom in India, France or Kenya. In contrast, curriculum-based methods are context-specific and show extensive cultural variation.
What are the risks and the challenges?
Data protection and security is of course still a huge issue and is the biggest challenge in this sphere. We have to ensure that children are never at risk of exposure and that the data is not misused in any way. This is something that needs more global attention and backing.
Another crucial point is that learning assessments should not be restricted to just one domain. There are multiple ways, and time and space to learn. Learning is continuous in nature and should be able to be adapted to the child’s needs at that particular point. The assessment should also be continuous in order to get a full picture of the improvement that the child is demonstrating. If there is no improvement, then we can provide interventions to help and find out why learning is not happening. From what we know from neuroscience, the earlier you can provide intervention, the better is the chance of the child to be able to change and adapt. The ability of the brain to learn and change is much easier and faster in childhood compared to adulthood.
Yet, we want to be cautious about the conclusions we draw about how to intervene with children. Poor academic performance might have a social or emotional reason.
Thus, learning today needs to be multi-dimensional. Along with academic competencies, social and emotional skills also need to be assessed. If this information is used wisely, it can provide a lot of insight about the child’s academic and emotional well-being. Based on the combination of the two, the right intervention can be provided. Unless multiple assessments all converge on the same result, the child’s learning abilities should not be labeled. AI gives a great opportunity to conduct multi-skills assessments, rather than just one. And that is something that we should leverage, rather than abandon. The standards for the baselines for the algorithms must be properly taken into consideration for any type of assessment. They must come from a large quantity of distributed data in order to provide more accurate results. That is something that we should not compromise under any condition.
How is the teaching community responding to this new way of learning and assessing?
There are teachers who worry about the future of learning but that is also because they do not necessarily have the full picture. People working and promoting the use of AI in learning must play a crucial role in telling teachers that they will not be obsolete. Teachers will be more empowered and be able to meet the needs of every kind of learner in their classrooms. The ideal world would be to have one teacher per child but that is of course impossible. AI is a tool to guide teachers when it comes to finding the right intervention for a student that might be struggling to learn. That intervention comes from data that has been checked for bias and diversity and does not use ‘a one size fits all ‘approach and therefore teachers can be more certain that it will fit the needs of the child. AI gives the opportunity for the teacher to tailor learning for the child. In addition, we do not really know all the different kinds of learning. Sometimes we have to be prepared to learn from children themselves. Children can give us insights into the different ways that learning actually happens, and teachers should be able apply them back into the classroom. Teachers are extremely powerful individuals who are able to shape the brains of so many children. If they are doing a good job, they are making individuals for life.
UNESCO MGIEP focuses on mainstreaming socio-emotional learning in education systems and innovating digital pedagogies. Based on the neurosciences of learning, the Institute’s flagship Rethinking Learning programme aims to equip young learners with competencies to address current global issues and in turn develop more peaceful and sustainable societies.
Find out more.