General reading impressions - 2020 Week 4

In this week, most of my readings were directed towards two goals: finishing a grant application (on a theme not covered by the recommendations below) and becoming acquainted with some technical literature that is relevant for my research interests: while I don’t work directly with human-computer interaction, competition law, or the economics of AI, understanding what is going on in those subjects is relevant for discussing AI regulation. I also managed to squeeze in some leisure reads, which could be of interest.


Michael Fairhurst. 2018. Biometrics: A Very Short Introduction. Oxford: Oxford University Press. Also available from Amazon BR.

An accessible, non-technical, introduction to the technical aspects of biometrics. It not only presents the current state of the art, but also the architectures used to build biometrical solutions and the problems those technologies are meant to solve (identification, validation, prediction). Given the space and technicality constraints of a Very Short Introduction, the author makes two choices, one of them less fortunate than the other.

The “good” choice was to focus on a few examples along the book — mainly fingerprints, irises, faces, and signatures —, which are rich enough to provide examples and narrow enough to be well developed. The “bad” choice was to reduce the space dedicated to the social consequences of biometrics, which are mostly sidelined (except for a discussion on ageing). Still, the technical framework presented in the book will provide good foundations who want to know a bit more about the technological possibilities and requirements in order to develop critiques about the social impact of biometrics in its many forms.

Victor Kaptelinin. 2014. Affordances and Design. Interaction Design Foundation.

An introduction to affordances and their central theoretical and practical questions. Roughly speaking, the term affordances refers to the possibilities of action that an environment offers us — for example, one can infer from a doorknob’s shape and other aspects that it is meant to be grabbed and turned, resulting in an open door.

The book presents an even-handed comparison of the main theories regarding affordances, and finishes by highlighting the problems that plague existing accounts of the concept. Even if affordance as a notion cannot be salvaged from those critiques, it has still contributed to progress in our understanding of human-computer interaction, and, since affordances are gaining traction on contexts outside HCI — see, as an example, Ryan Calo and Mireille Hildebrandt discussing law as an affordance —, it is important to understand the concept and its development.

Mary Robinette Kowal. 2013. Shades of Milk and Honey. Corsair. Also available from Amazon BR.

A run of the mill Regency romance, but with some low-key magic (“glamour”). It was a light read, good to distract me from the mostly dull workload of this week, but it made me think that perhaps I should try again to read Jane Austen, after some disastrous attempts in my early 20s.


Reuben Binns and Elettra Bietti. 2019. Dissolving privacy, one merger at a time: Competition, data and third party tracking. Computer Law & Security Review, available online.

Has some valuable contributions: a clear introduction to third party tracking technologies, an overview of how competition law struggles to deal with data — especially with the consequences of its non-priced uses —, and an empirical investigation of the combination of data from third-party trackers resulting from corporate mergers.

Joanna J. Bryson. 2020 forthcoming. The Artificial Intelligence of the Ethics of Artificial Intelligence: An Introductory Overview for Law and Regulation. In M. Dubber, F. Pasquale, & S. Das (Eds.), The Oxford Handbook of Ethics of Artificial Intelligence. Oxford University Press.

As a CS graduate, one of many things that get me exasperated is that so much of the AI Law debate relies on third-hand accounts of what AI is or can do. Many renowned scholars try to come into the debate with little more preparation than unironically using the terms from Erwan Cario’s AI bullshit bingo. That is not to say, of course, that there is no serious legal scholarship on AI — in fact, there are many serious legal thinkers of all origins and theoretical currents presenting relevant questions and frameworks, and AI technologists would do well to listen to them —, but serious engagement with the consequences of artificial intelligence requires that one at least know what is talking about.

Bryson’s paper, therefore, fills a much-needed gap, as the author highlights some key facts and issues regarding the nature of AI: intelligence as an informational process that happens in the physical world, AI as a human-designed artefact, and the possibility and relevance of logging, among many others. At the same time, she dispels many myths, such as the reification of algorithms and the possibility of replicating human beings through AI. In doing so, Bryson highlights the relevant questions for the governance of AI systems, which rightfully put human welfare and rights in the central place, and provides a solid first step for lawyers and legal scholars who want to engage with technological themes.

Jack Clark and Gillian K. Hadfield. Regulatory Markets for AI Safety. Economics of Artificial Intelligence, Toronto.

The paper provides a good overview of existing regulatory approaches — both those conducted by state actors and those led by the private sector — and their limits. Their alternative model involves private actors buying regulatory services from private regulators, which would have freedom to operate, within the space defined by direct regulation by national or international public institutions. Regulators would not compete in outcomes, as those are established by governments, but in terms of how efficiently their regulatory requirements can ensure the overall standards and targets set to the regulatory market.

As the authors are well aware, the private regulators themselves might be subject to conflicts of interest, capture, free-riding problems and insufficient competition because of lack of scale, among other issues. Still, their proposal highlights that any feasible solution for AI regulation will need to find a healthy balance between strict enforcement of socially desirable outcomes and the flexibility in means that is necessary for technological development.

Inge Graef, Damian Clifford, Peggy Valcke. 2018. Fairness and enforcement: bridging competition, data protection, and consumer law International Data Privacy Law 8(3), pp. 200–223.

Competition law, data protection law, and consumer law have various intersections: as an example, one might think of the data consequences of corporate mergers and acquisitions, or how consumer data may be used to shape individual behaviour. In this paper, the authors highlight that, despite some tensions on how each of these fields of law operate, there is ample room for positive effects from one to another.

That can happen when consumer or competition law are used to address concerns that are not well covered by the individualistic focus of data protection law, or, as Graef et al. emphasise, by interpreting competition concepts such as market definition and market power in light of data protection requirements. The authors propose that the notion of fairness, present in those three fields of law, may be used to harmonise the actions of consumer, data protection, and consumer law, and that seems to have some potential, especially when combined with Palka (2020)’s proposals for data management regulation that goes beyond the individual focus of data protection.

Juan Mateos-Garcia. 2017. To Err is Algorithm: Algorithmic fallibility and economic organisation. Nesta.

The author proposes a mathematical model for understanding the impacts of algorithmic errors and the possible impact of human intervention. This model is concerned with decisions such as content filtering, in which a decision is intended to accept good content and filter bad contempt, and so it does not really consider situations in which good input may still lead to inadequate decisions, either by the machine itself or by the human intervenor, which are the situations most immediately relevant for my own research. However, the proposed model is a good start for treating automated decisions, and already shows some insights, such as when the use of human supervisors might become prohibitively expensive.

Yong Suk Lee et al. 2019. How Would AI Regulation Change Firms’ Behavior? Evidence from Thousands of Managers. Economics of Artificial Intelligence, Toronto.

This paper uses a survey to understand how managers would change their behaviour in response to various approaches to AI regulation. Based on the responses, regulation leads to slower adoption of AI systems, especially when it comes to small firms, who become less likely to innovate or adopt new systems. However, it leads to greater awareness of safety and transparency issues regarding AI systems.

The impact of regulation is more noticeable when it is performed through general norms, with industry- or agency-based norms having a smaller impact, and different industries have different reactions to AI. Finally, AI regulation contributes to top-heavy companies, as more managers are hired to cope with the new demands, with a corresponding reduction in the hires of technical or low-skilled workers. Understanding those tradeoffs is relevant for designing regulation that achieves its desired goals without generating unintended side-effects in the economy.

Online readings

Anita Gurumurthy and Nandini Chami. 2020. The intelligent corporation: Data and the digital economy.

An essay about how dataification and artificial intelligence have transformed corporate power, with special attention to the consequences of that to the Global South. The descriptive part of this text is very interesting, but I do not buy into the solutions proposed by the authors. First, an economic rights framework for data might end up legitimising practices that should not be acceptable, even if one can adequately price the relevant externalities. Furthermore, and this is what most alarms me, a locally-centred model of digital governance would contribute to the reinforcement of local inequalities and structures of powers, and running critical infrastructure as public utility might have the same effects. (see, e. g., the Russian plans for a separate Internet). While I agree that decentralised digital governance is desirable, there must be more effective ways to perform that breakdown. Despite my reservations, the text itself still provides a very interesting perspective.

Juan Mateos-Garcia. 2020. The Economics of AI Today. The Gradient, January 18, 2020.

This text presents an overview of a recent conference on the economic consequences of AI and how economics has studied them. The blogpost itself provides a brief introduction to the most popular points in research, as well as summaries of the papers that were presented. In doing so, it provides a good introduction to the field.


During my vacations, I try to complete at least one MOOC, so as to diversify my horizons a bit. After the academic year begins, I usually fail to keep up with the courses, so I occasionally take some time to do intensive sessions for each course. That, of course, contributes to my falling behind on my reading backlog, but it is usually a nice trade-off.

Since I am not a big fan of video lectures, I usually prefer to read the transcripts. That is why I will allow myself to cheat a bit and list those courses in my reading recommendations.

Ancient Philosophy: Plato and his predecessors and Ancient Philosophy: Aristotle and his successors. Coursera.

These two courses, designed as a pair, provide a swift introduction to Classical Greek philosophy, with instructive classes and good selections of excerpts. The first course dedicates one section to the Milesian naturalists, one section each to Heraclitus and Parmenides, and the remainder to Plato: Eutyphro, Meno, Republic (Books 1–4 and 5–7), and Timaeus. The second

For each course, the grade is composed by a quiz for each section, and a final essay. The Plato course lasts 4 weeks, while the Aristotle one lasts 5, in both cases demanding about 4 hours per week between lecture videos, reading the texts, and completing coursework.

Social Norms, Social Change: Part I. Coursera.

First part of a course taught by Christina Bicchieri on social norms. This part covers the various types of social norms, their influence on individual and group behaviour, and how to detect and measure them. It does a great job of showing how philosophical discussion and social science may be joined for real-world impact.

The course lasts 4 weeks, demanding approximately 3 hours per week. Grades are based exclusively on quizzes.

European Business Law: Understanding the Fundamentals. Coursera.

Introduces the core concepts of European Union law, with special attention to those related to business law, such as the Four Freedoms. Besides this conceptual introduction, Week 2 provides a practical introduction to European law, as it teaches much of the terminology and of the databases where one can find legislation and case law.

Grades are based on quizzes, with an optional essay for those who want to finish the course with honours. The course is divided in 5 weeks, and it should take about 10 hours between lecture videos and coursework.

Researcher, Law and Artificial Intelligence

Currently researching the regulation of artificial intelligence at the European University Institute.