Regulating Artificial Intelligence

Binary Ethics and the Law

Dominika Harasimiuk, Tomasz Braun

• Verlag: Routledge
• Jahr: 2021
• ISBN:
  9781000320398
• Auflage: 1
• Thema: Law
• Thema: Law Theory & Practice

Exploring potential scenarios of artificial intelligence regulation which prevent automated reality harming individual human rights or social values, this book reviews current debates surrounding AI regulation in the context of the emerging risks and accountabilities. Considering varying regulatory methodologies, it focuses mostly on EU's regulation in light of the comprehensive policy making process taking place at the supranational level.

Taking an ethics and humancentric approach towards artificial intelligence as the bedrock of future laws in this field, it analyses the relations between fundamental rights impacted by the development of artificial intelligence and ethical standards governing it. It contains a detailed and critical analysis of the EU's Ethic Guidelines for Trustworthy AI, pointing at its practical applicability by the interested parties. Attempting to identify the most transparent and efficient regulatory tools that can assure social trust towards AI technologies, the book provides an overview of horizontal and sectoral regulatory approaches, as well as legally binding measures stemming from industries' self-regulations and internal policies.

1 Instead of Introduction – Algorithmic Society, Artificial Intelligence and Ethics

1.1 Topic’s relevance

The development of digital technologies, which entails fast and more efficient growth of capacity of Artificial Intelligence, brings significant changes to all aspects of everyday lives of individuals and societies. Whatever the world is experiencing right now is compared to the times when electricity transformed the economy, culture and politics.¹ Digital technologies based on algorithms are widespread and present in the surrounding reality of an average person who uses mobile phones, personal computers, home appliances, TV sets, cars, and many other electronic devices. AI-based technologies may have hugely beneficial impact on human lives. Technology enables the improvement of diagnoses and development therapies for diseases by reducing energy consumption and lessening the need for pesticides, contributing to a cleaner environment, optimising resources, anticipating disasters by improving weather prediction, establishing faster and safer transportation by increasing general road safety. It also drives economic productivity growth and contributes to sustainability, improving financial risk management and detecting fraud and cybersecurity threats. Aside from those listed, it also enables law enforcement and helps prevent crime more efficiently.² Benefits are counterbalanced by serious concerns, which relate to growing automation leading to possible increase in unemployment rates, biased decision-making, excessive access to privacy by authorities, overcomplicated technological solutions increasing the imbalance in the access to knowledge and extraordinary power concentration over in the hands of few corporations of the worldwide reach, like Google, Facebook or Amazon. Finally, development of AI industry causes fierce competition between major global actors, namely USA, China and EU.³

1 More on Industrial, through Digital, towards AI Revolution, see Spyros Makridakis, ‘The Forthcoming Artificial Intelligence Revolution’ (2017) 1 Neapolis University of Paphos (NUP), Working Papers Series https://www.researchgate.net/publication/312471523_The_Forthcoming_Artificial_Intelligence_AI_Revolution_Its_Impact_on_Society_and_Firms accessed 20 July 2020.

2 The High Level Strategy Group on Industrial Technologies has recommended including AI as one of the key enabling technologies due to its cross cutting enabling potential crucial for European industry, see High Level Group on industrial technologies, Report on ‘Re-Defining Industry. Defining Innovation’ (Publication Office of the EU Luxembourg 2018). See also, Commission, ‘Artificial Intelligence for Europe’ (Communication) COM(2018) 237 final, 1. See also, Paula Boddington, Towards a Code of Ethics for Artificial Intelligence, Artificial Intelligence: Foundations, Theory, and Algorithms (Springer Int. Publishing 2017) 2.

3 Commission, ‘Coordinated Plan on AI’ (Communication) COM(2018) 795 final, 1; See also, Paul Nemitz, ‘Constitutional Democracy and Technology in the Age of Artificial Intelligence’ (2018) 376 Philosophical Transactions A, The Royal Society, 3 https://ssrn.com/abstract=3234336 accessed 20 July 2020.

AI technologies are not only impacting industries and economy, but also political structures and democratic mechanisms. It is well established that the market for AI includes both business-to-consumer (B2C) and business-to business (B2B) markets and platforms. What goes beyond these traditional spheres are public-to-citizens services (P2C), with new tools of civic participation, e-democracy and e-government.⁴ In all these areas there is a need for a comprehensive regulatory approach towards AI, embedded in ethics and trust. These are two paradigms around which all the legislative and regulatory measures are being adopted nowadays at the European level. The European approach towards AI technologies, which is the topic of this book, consists in putting various European values at the heart of policymaking processes. The European Union perceives itself nowadays as a major global stakeholder in the field of AI regulation. Such a position is part of the broader European Digital Single Market policy, in which AI is becoming a strategic area for European economic development. The EU’s approach towards AI regulation intends to cover socio-economic, legal and ethical issues. In longer term, such a holistic vision is about creating the European Single Market for Trustworthy AI, where the EU could benefit from the value of AI, while minimising and preventing its risks.⁵

4 See, Deloitte Insights, ‘How Artificial Intelligence Could Transform Government’ (2017) https://www2.deloitte.com/insights/us/en/focus/artificial-intelligence-in-government.html accessed 20 July 2020.

5 High-Level Expert Group on AI (HLEG AI), ‘Policy and Investment Recommendations for Trustworthy AI’ (Brussels 2019) 6–7.

The relevance of the research topic stems from its up-to-date character and its future impact. The regulatory approach towards AI adopted at present, will shape our reality in the following years and decades. The thorough analysis of current EU policies, regulatory and legislative processes and proposals touching upon AI technologies, will give us the possible insight on how the development of our economies and societies will look like in the long-term perspective. It could also bring some reflections on the most appropriate approach towards discussed topic.

1.2 Goal of the book

As stated above, digital technology breakthrough and artificial intelligence in particular can help to address many of the world’s biggest challenges. The pace of technological progress that is now being developed across the world is incredibly rapid. At the same time AI itself brings new challenges and raises serious legal and ethical questions.⁶ This phenomenon is not new. Usually, legal concepts and norms had to adjust to the novelty challenges posed by the progress in the sphere of science, culture, politics, economy. It is not different with the current technological changes.⁷

6 Commission, ‘Building Trust in Human-Centric Artificial Intelligence’ (Communication) COM(2019)168 final, 1.

7 Expert Group on Liability and New Technologies, ‘Liability for Artificial Intelligence and Other Emerging Digital Technologies’, (Report from New Technologies Formation) (Publication Office of the EU Luxembourg 2019) 11.

This phenomenon is being seen from various angles. Eventually it is about locating the phenomenon of algorithmic changes within the societies in the governance and regulatory environment. The regulation can be seen as polycentric social system with six elements, creating its dynamics: goals and values, knowledge and understanding, tools and techniques, behaviours of individuals, behaviours of organisations and trust and legitimacy.⁸ Regulatory environment could be defined as organised attempts to manage risks or behaviour to achieve a publicly stated objective or set of objectives.⁹ According to this theory there are two main forms of regulation: command and control regulation and design-based regulation. The first form of regulation refers to the use of legal or regulatory rules that dictate behaviour. They come with punishment and incentive mechanism. In reaction to them, on the side of the addressee of these norms is an arbitrage to comply for the reward or to ignore and risk punitive consequences.¹⁰ The second form of regulation, which is design-based, is to create regulatory standards adjusted to the design of the entire regulated system. In other words, it is based on constructing an architecture adapted to human behaviour that matches the preferred behaviours.¹¹

8 Julia Black, Andrew D. Murray, ‘Regulating AI and Machine Learning: Setting the Regulatory Agenda’ (2019) 10 European Journal of Law and Technology http://eprints.lse.ac.uk/102953/ accessed 22 July 2020.

9 Julia Black, ‘Learning from Regulatory Disasters’, (2014) 24 LSE Law, Society and Economy Working Papers, 3 http://dx.doi.org/10.2139/ssrn.2519934 accessed 22 July 2020.

10 Julia Black, ‘Decentring Regulation: Understanding the Role of Regulations and Self-Regulation in a <Post-Regulatory> World’ (2001) 54 Current Legal Problems, 105–106.

11 Robert van den Hoven van Genderen, ‘Legal Personhood in the Age of Artificially Intelligent Robots’ in Woodraw Barfield, Ugo Pagallo (eds.), Research Handbook on the Law of Artificial Intelligence (Edward Elgar 2018) 224 ff.

According to this concept much of the present algorithmic governance and regulatory framework constitutes a design-based sort of regulation.¹² In line with this theory, a design-based regulation and algorithmic decision support system is a type of nudging. Nudging is a regulatory philosophy that has its origins in behavioural economics based on the assumption deriving from cognitive psychology claiming that people are less rational than universally believed. They display biases and psychologically preferred stereotypes.¹³ That sometimes deviates them from the expectations of rational choices theory and causes damages to their own long-term well-being. The best example of this is shown in the general tendency to over prioritise the short-term future. People discount the value of future events too much that is instead of doing it according to an exponential according to a hyperbolic curve. They favour sooner rewards even though they are smaller rather than larger ones if they ought to come later.¹⁴

12 Karen Yeung, ‘Hypernudge: Big Data as a Mode of Regulation by Design’ (2016) 1,19 TLI Think! Paper Information, Communication and Society, 4; John Danaher, ‘Algocracy as Hypernudging: A New Way to Understand the Threat of Algocracy’ (2017) https://ieet.org/index.php/IEET2/more/Danaher20170117?fbclid=IwAR3gm6lIWN8Twb8bE6lTIdtintwhYSWF2FTDkRGzMs1xa8XTD4bGgoQJiXw accessed 22 July 2020.

13 Jonathan Beever, Rudy McDaniel, Nancy A. Stamlick, Understanding Digital Ethics. Cases and Contexts (Routledge 2020) 82.

14 Danaher, ‘Algocracy as Hypernudging’ (n 12).

The same concept can also apply to regulatory domains. Legislators and regulators may create a kind of decision-making situation building so called choice architectures that benefits from the nature of human psychology and nudge them into preferred behavioural patterns.¹⁵ This approach has gained a lot of popularity only recently. The public authorities have started to set up behavioural analysis units to implement nudge-based policy settings in multiple areas.¹⁶

15 Antje von Ungern-Sternberg, Autonomous Driving: Regulatory Challenges Raised by Artificial Decision-Making and Tragic Choices in Woodrow Barfield, Ugo Pagallo (eds.), Research Handbook on the Law of Artificial Intelligence (Edward Elgar 2018) 257.

16 Yeung (n 12) 4–6.

Therefore, nudging is a type of design-based regulation because it is not about enforcing the created rules and regulations but about handwriting policy preferences into behavioural architectures. The algorithmic governance systems work like nudges especially within decision-support systems. These are forms of algorithmic governance which use data-mining techniques to present choice options. People typically do not question the defaults provided by our algorithmic systems they use on a daily basis. This same mechanism might be used as a support regulatory framework where algorithmic decision support systems are used in many policy domains.¹⁷

17 Ibid.

Against this theoretical background, we would like to draw the goal of our book, which is to outline the general regulatory approach of the EU towards algorithmic reality. Ethics is the central notion around which all regulatory steps are revolving. And regulation of what is supposed to be an ethical AI takes various forms. It is both commands based, and design based. It combines proper centralised legislative measures with decentred regulation resting in hands of interested stakeholders. Thus, there is a complex network of top-down measures and bottom-up initiatives, binding and non-binding rules, hard and soft laws, horizontal and sectoral rules, supranational, international, national and industry-based regulations. Such a complex, intertwined regulatory environme...