Abstract:
This paper argues that the evidence-led decision making focus of contemporary education policy is converging with recent developments in Artificial Intelligence (AI) to create the conditions for an emerging 'computational rationality' in education policy, particularly in reference to what Williamson (2016) has termed 'educational data science', a hybrid of approaches from statistics, computer science, and psychology and neuroscience from 'learning sciences'. 'Computational rationality' describes work overlapping the fields of artificial intelligence, cognitive science and neuroscience, premised on the idea that there is a 'computational' basis for intelligence (Gershman et al., 2015). The argument of this paper is that this 'computational rationality' overlaps with the ways in which contemporary education policy making and analysis is focusing on predictive capacity. This capacity aims to utilise AI, such as machine learning, to provide ways of mining, sorting, classifying and identifying new patterns from existing data 'warehouses' of performance and administrative data. This will allow for processes of resource allocation, decision making around student potential and teacher performance, including the coupling of data dashboards with real time analytics.
To make this argument, the paper first outlines some key aspects of computational rationality including: the rational agent; machine learning; and neural networks. This will trace out how AI has moved from a focus on rules to a focus on data in decision making. Second, the paper looks at overlaps with AI and education policy, with a focus on: 1. The role of personalised data and personalised learning, where interoperability and standardisation of data are key to the introduction of AI in education; and, 2. Examining how decisions are made in AI, including in machine learning. This will focus on the 'black-box' of AI, which covers the issue of transparency in AI. This issue connects to know when to trust predictions made by machines. AI, including machine learning, works on probabilistic ideas where predictions are based on approximations, and, as such, 'rendering the production of prediction visible is a central challenge in data mining and machine learning itself' (Mackenzie, 2015, p. 435). The paper will examine the move to big data analytics that are claimed to provide predictive capacities that promise new degrees of control over the future. We conclude the paper by mapping out the characteristics of what we term an emerging 'computational education policy'.
To make this argument, the paper first outlines some key aspects of computational rationality including: the rational agent; machine learning; and neural networks. This will trace out how AI has moved from a focus on rules to a focus on data in decision making. Second, the paper looks at overlaps with AI and education policy, with a focus on: 1. The role of personalised data and personalised learning, where interoperability and standardisation of data are key to the introduction of AI in education; and, 2. Examining how decisions are made in AI, including in machine learning. This will focus on the 'black-box' of AI, which covers the issue of transparency in AI. This issue connects to know when to trust predictions made by machines. AI, including machine learning, works on probabilistic ideas where predictions are based on approximations, and, as such, 'rendering the production of prediction visible is a central challenge in data mining and machine learning itself' (Mackenzie, 2015, p. 435). The paper will examine the move to big data analytics that are claimed to provide predictive capacities that promise new degrees of control over the future. We conclude the paper by mapping out the characteristics of what we term an emerging 'computational education policy'.