Annual Research Review: Educational neuroscience: progress and prospects

There is a long history of interaction between psychology and education (Bruer, 1997). In response to the emergence of educational neuroscience, it has been argued that the field of psychology is sufficient to inform education with a scientific understanding of learning processes. For example, Bowers claimed “psychology is the relevant discipline to improve educational outcomes for all children” (Bowers, 2016b, p. 633) and Bishop (2014) claimed that “the people who can be useful to teachers are psychologists.” This makes sense, because education concerns the effect of instruction on changes in behaviour. Psychology studies behaviour, while neuroscience is the study of brain mechanisms underlying behaviour.

Yet there are risks in a pure psychology approach. Psychological theory infers hidden causal mechanisms to explain and predict observed behaviour. For example, it infers mechanisms such as ‘working memory’ and ‘attention’, which are then imbued with certain properties: working memory has a ‘limited capacity’; attention works like ‘a spotlight’ or enforces a ‘bottleneck’ on perception. What is the inspiration for these inferred mechanisms? Several heuristics are used, among them task analysis, reverse engineering and introspection. Historically, technological metaphors have also been influential, including hydraulic systems, steam engines and telephone exchanges. The symbolic/desktop computer has been particularly influential in modern cognitive psychology and is the origin of the concept of working memory. The symbolic computer employs general purpose processing components: a central processing unit, working memory, the hard drive and functions by abstracting information into a common format so that these general devices can process information of all different types. Research in computer science, artificial intelligence and machine learning has shown that such an engineering solution can create extremely powerful computational devices. However, this solution of domain-general computation may not be one that is deliverable by the brain. The risk of a pure psychology approach, then, is that it can lead to erroneous theories based on inferred causal mechanisms that cannot be realised in real time by the brain (Mareschal et al., 2007).

Let us make this more concrete. Here are some education-relevant questions that point to peculiar properties of learning: Why am I liable to forget what the capital of Hungary is, but not to forget that I'm afraid of spiders? Why do I find I have learned things better after a good night's sleep? Why does my mind go blank when I'm stressed in an exam? I get 7/10 in a test – why am I delighted if I was expecting to get 5, demoralised if I was expecting to get 9? When I became a teenager, why did I sometimes do stupid, occasionally dangerous things just to impress my friends? Why can I learn a new language so much more easily when I'm 5 years of age than when I'm 50? From a computer science perspective, one could imagine designing a machine learning system that did not have any of these peculiar properties. The machine learning system could store Capitals-of-European-Countries and Animals-I'm-Scared-Of as similar types of memories. It need not forget either. One could build the machine learning system without emotions like ‘stress’ or ‘anxiety’, emotions which on the face of it seem to detract from human learning performance. The system could be driven by absolute performance metrics so that a 7/10 is just a 7/10. Moreover, battery life permitting, one could build a system that did not need to sleep to achieve efficient learning. The answers to the above education-relevant questions lie not in psychology, but in the particular way the human brain works, because of its particular biological and evolutionary origins. That is, there are aspects of learning that simply do not make sense except with respect to the brain.

There are several reasons to suspect that the set of inferred mechanisms that psychology currently uses is flawed because psychology has not paid sufficient attention to the constraints of neuroscience. First, as with the symbolic computer, proposed cognitive mechanisms tend to be general purpose: mechanisms such as working memory, long-term memory, attention and cognitive control. When people train on a given task, their performance usually improves. Let's say a person trains on a task taken to involve one of these general mechanisms, such as working memory. One would then expect the benefits of training also to be observed across a wide range of other abilities assumed to utilise the same general mechanism. However, as a rule, such ‘far transfer’ is rarely observed; people tend to show improvements only on abilities similar to those on which they train (Sala & Gobet, 2017; Thorndike & Woodworth, 1901). This is not how one would expect a system to behave if it were based on domain-general mechanisms. The implication is that the actual mechanisms used by the brain are less general than current cognitive theory supposes, and that it uses specific circuits for specific skills.

Second, there are behavioural observations that cannot be predicted from psychological theories, such as why the ability to learn should alter with age, or why forgetting should occur in particular ways for particular memories. Third, there are phenomena that seem surprising given current cognitive theory and which lead to ad hoc theoretical additions, such as the role of sleep in learning, or the effect of meditation on behavioural regulation (see later). Fourth, current psychological theories show a very poor match with the activation of neural structures, exhibiting numerous many-to-one relationships (where many cognitive processes are associated with the same brain region) and one-to-many relationships (where a single cognitive process is associated with multiple brain regions; Price & Friston, 2005). While the way that cognitive processes are implemented in the brain is of little relevance to educators per se, it becomes relevant if the cognitive primitives being proposed are wrong, because then teachers' expectations will be wrong. For example, if teachers believe attention is a single cognitive process of ‘focusing on one thing’, they may expect certain behaviours to develop together and be trainable together. However, the multiple brain areas involved in attention suggest that it is not one thing but a cluster of overlapping mechanisms, which involve, respectively orienting for action, top–down contributions to stimulus detection and sustained maintenance of task set (e.g. Petersen & Posner, 2012). These mechanisms may not develop in harness or be equally trainable.

These reservations about current cognitive theory are not principled limitations for psychology. They merely suggest psychology needs better theories. Theoretical advances will occur more quickly with closer ties to neuroscience, so that less time is spent considering possible ways the cognitive system could have worked, and more time considering how it actually works in the brain. In the meantime, for educators, psychological constructs such as working memory remain a double-edged sword. They can be a useful summary of behavioural observations. A child with problems in ‘working memory’ will have problems keeping in mind a list of instructions to follow, a set of moves to make, or a sequence of numbers, and teachers can adjust classroom learning environments to mitigate these difficulties (Alloway & Alloway, 2014). Yet if teachers expect training working memory to yield general improvement across a diverse range of abilities, they would be wrong (Sala & Gobet, 2017; Simons et al., 2016).

This model of interaction between fields – of neuroscience with psychology, and then psychology with education – has been advanced by a number of researchers, from Bruer (1997) onwards (e.g. Howard-Jones et al., 2016; Mareschal, Butterworth, & Tolmie, 2013; Varma et al., 2008). Crucially, it requires that evidence from neuroscience is used to modify psychological theory, rather than merely demonstrating how the brain ‘implements’ current cognitive theory – that is, using brain imaging to produce a list of brain regions that are activated during the operation of each putative cognitive process. Such a list offers little added value to educators.

However, this indirect, two-step route between neuroscience and education is not the only possibility. By virtue of the fact that the brain is a biological organ and therefore subject to metabolic constraints, there may be direct links. Factors such as energy supply, nutrition, response to stress hormones and environmental pollution can potentially influence brain function, including learning. Thus, while educational neuroscience predominantly places psychology at its centre, research on the impact of non-psychological factors on educational outcomes, such as aerobic fitness, diet and air quality, also falls within its remit. These two pathways are shown in Figure 1.