Data, learning, and students on the autistic spectrum

Learning in data.

There is an undeniable push towards personalised learning in education (Gasevic et al 2019). Few factors influence this. The enthusiasts of personalisation claim that it can make education more inclusive. Students will surely enjoy it and learn more (Bulger 2016). The other camp warns the change is driven by commercialisation of education (Biesta 2017). They claim the trend will negatively affect the social dimensions of learning (Bulger 2016). For students on the Autistic spectrum both the perceived advantages and disadvantages of datafication of education might look somewhat different.


Personalised data-driven learning is to open the door for exploring one’s own interests (Bulger 2016). This could be beneficial for ASD student, who fixate on topics of interest. Furthermore, they could also be spared studying subjects they find no interest in, or difficult to comprehending. Also, the move from summative assessment to competency (ibid.) – based education seems better for those students.

Another plus is that in a lot of data-driven learning the student sits in front of a machine, limiting their social interactions. ASD students struggle with social situations. In particular, they find looking at faces harder. This means that interacting with the teacher, e.g. to setup tasks or receive feedback, via computer is better for them.


The main problem with datafication of education is that it often perpetuates existing inequalities (Tsai et al 2020). Broadly speaking, neurodiverse students are already a marginalised group, so they constitute the primary target here. It’s especially true for women with ASD, who are often undiagnoses so are neither receiving support nor showing on the radar of schools’ disability services.

There is also a concern with how much the initial input models actually accurately describe the learners from varied backgrounds (ibid.). For the algorithm to work, ‘signature behavioural patterns must be identify’ (Wilson et al 2017 p. 997). ASD behavioural patterns vary significantly from those of neurotypical students. Worse yet, they vary among individuals with autism, because ASD is not a linear scale, but a wide range spectrum (Baron-Cohen 2016). If the machine is taught to respond based on neutorypical students, it will fail when faced with atypical ones. This could lead to all sorts of problems with feedback loops (O’Neil 2016). For example, the unusual response could trigger an intervention (Bulger 2016), which would not only be unnecessary, but also stressful to someone with ASD. Or an intelligent tutor (Bulger 2016) might misinterpret their body language.

Bulger (2016) points out that one problem with personalised learning is the disregard for the social aspect of education. While autistic people are loners, many, especially high functioning ones, express the wish to socialise and the need for friends (Baron-Cohen 2016). Therefore, a complete removal of this aspect of learning isn’t necessarily optimal for them. Some students with ASD in the Tsai et al (2020) study confirmed this sentiment.


All in all, data-driven learning might offer students with ASD some benefits, such as reducing social interactions or offering personalised paths of learning to harness their ability to hyperfocus. However, the risks associated with the trend are greater for this group than neurotypical students.


Baron-Cohen, Simon. (2016) Autism: An evolutionary perspective. 1st Symposium of EPSIG.

Biesta, Gert. 2017. ”The Beautiful Risk of Education” :

Bulger, M. 2016. Personalized Learning: The Conversations We’re Not Having. Data & Society working paper. 

Gasevic, D., Tsai, Y.-S., Dawson, S. and Pardo, A. (2019), “How do we start? An approach to learning analytics adoption in higher education”, International Journal of Information and Learning Technology, Vol. 36 No. 4, pp. 342-353.

O’Neil, Cathy (2016) Weapons of Math Destruction. Random House Audio (Audible release date 09-06-2016)

Tsai, Yi-Shan. Perrotta, C. & Gašević, D. (2020) Empowering learners with personalised learning approaches? Agency, equity and transparency in the context of learning analytics, Assessment & Evaluation in Higher Education, 45:4, 554-567, DOI: 10.1080/02602938.2019.1676396

Wilson. A. Watson, C. Thompson, T. L. Drew, V. & Doyle, S. (2017) Learning analytics: challenges and limitations, Teaching in Higher Education, 22:8, 991-1007, DOI: 10.1080/13562517.2017.1332026

Leave a Reply

Your email address will not be published. Required fields are marked *