Contents
1.1 Introduction
1.2 Artificial Intelligence: The Eternal Dream
1.2.1 Harmful AI
1.3 National AI Strategies
1.3.1 Defining National AI Strategies
1.3.2 No Strategy, No AI?
1.4 To Regulate, Or Not To Regulate?
1.4.1 AI Tensions: Between Innovation and Regulation
1.4.2 Risk Mitigation
1.4.3 Design and Deployment Concerns
1.5 Governance Approaches
1.6 The Limits and Potentials of Ethics in National AI Strategies
1.6.1 AI Ethics Limits: Five Issues
1.6.2 AI Ethics Potentials: Ten Cues
1.7 Conclusion
Notes
References
1.1 Introduction
The development of artificial intelligence (AI) will shape the future of power. The nation with the most resilient and productive economic base will be best positioned to seize the mantle of world leadership. That base increasingly depends on the strength of the innovation economy, which in turn will depend on AI.
(US National Security Commission on Artificial Intelligence, 19 May 2020).
Over the past three to five years, AI technologies and AI research have become a major focus of private and public funding initiatives.1 This heightened attention is paralleled by a growing proliferation of AI technologies across social life. Today, these technologies are embedded into many devices and services that people use on a daily basis, ranging from e-mail spam filters to navigation devices or shopping websites. This development is advancing at a rapid pace, which has led to the competition for (national) leadership in the AI field becoming so fierce that it has been referred to as a “global AI race.”2 In this “race,” AI has become the strategic focus of many global technology companies who commit substantial resources to push AI innovation,3 and the amount of capital invested in AI companies in the US came to a staggering $9.3 billion in 2018.4 In Europe, the investment into tech companies (not only AI companies) reached $23 billion in 2018,5 while Chinese tech giants Baidu, Alibaba, and Tencent equally investing heavily into AI technologies and start-ups, backed by a government plan to build a domestic AI industry worth around $150 billion by 2030 (Mozur, 2017).
Other nations and regions are not lagging behind. Although much attention has been on the heated AI competition between the United States of America and China (Metz, 2018), there is investment and policy activity in other regions and countries as well. For example, the EU Commission pledged investment into AI of €1.5 billion for the period 2018–2020 under the Horizon 2020 research programme, expected to trigger an additional €2.5 billion of funding from existing public–private partnerships and eventually lead to an overall investment of at least €20 billion by 2020 (European Commission, 2018a). National European examples include France announcing a €1.5 billion pure government funding for AI by 2022 (Cerulus, 2018), Germany outlining €3 billion aimed at spending on AI research and development by 2025 (Delcker, 2018), and the United Kingdom forging the AI Sector Deal (part of the Industrial Strategy) worth £1 billion (British Government, 2018). In Asia, China’s government is leading with US$7 billion minimum AI investment by 2030 (Ravi and Nagaraj, 2018), well ahead of South Korea, intending to invest US$2 billion in AI by 2022 (Synched, 2018). Canada has pledged C$125 million (CIFAR, 2017), while Australia announced an AUD$29.9 million investment into AI over four years in its 2018–2019 budget (Pearce, 2018).6 While governments have to foster innovation, they are also tasked with mitigating the potentially adverse effects of AI through regulation and governance.
1.2 Artificial Intelligence: The Eternal Dream
Despite the recent “AI hype” (Spencer, 2019), the idea of an “artificial intelligence” is not new: it could be claimed that it dates back to Homer’s Iliad (Cave and Dihal, 2018; Royal Society, 2018). Between the 1950s and the mid-1970s, as computers became faster and cheaper, AI flourished, which was followed by an “AI winter” in the 1990s and 2000s and a dip in interest and funding in AI, despite the many AI advancements made during that time (Anyoha, 2017). The new AI hype is based on three developments that coincided and that are deeply connected: the availability of large datasets, the rapid advancement of computational machinery and processing power, and the invention of self-learning algorithms7 based on artificial neural networks (“deep learning”).8
The success of new AI technologies has reignited the imaginary of conscious machines or robots that have agency (Royal Society, 2018) and the fear that they may overthrow humanity (Bostrom, 2016). But we are far from that type of “general artificial intelligence” (Knight and Hao, 2019). All of the AI systems in place or under development today are what can be called “narrow artificial intelligence”; basically, statistical models that can (teach themselves to) detect correlation, but not causality.9 This means that AI technology can be very good at very specific tasks, such as identifying the pixels in a photograph to help doctors diagnose a malignant mole.10 But it also means that AI does not possess the capacity to deal with the sheer complexity of social life.11
1.2.1 Harmful AI
AI systems can be riddled with high error rates (especially facial recognition or object detection systems), which can disproportionately affect certain groups, such as people with darker skin tones.12 AI systems can also be very vulnerable to outside influence, for example, to adversarial attacks,13 which can have devastating consequences in high-stake contexts, such as diagnostics, autonomous driving, or combat. These attacks do not need to be digital, “physical world attacks” can also affect deep learning visual classification, such as stickers on stop signs.14
Over the past years, new research has demystified the account that algorithms and AI are de facto neutral and shown that existing power imbalances, inequalities, and cultures of discrimination are mirrored and exacerbated by automated systems. Important works include, but are not limited to: Virginia Eubanks’15 research on how data mining, algorithms, and predictive risk models exacerbate poverty and inequality in the US; Safiya Umoja Noble’s16 work on how search engines discriminate against women of colour; Cathy O’Neil’s17 work on how the large-scale deployment of data science tools can increase inequality; Marie Hicks18 demonstration of how gendered inequalities in computation are not accidental, but derive from a particular cultural landscape and a series of policy decisions; the work of Joy Buolamwini and Timnit Gebru19 on discrimination in image databases and automated ender classification systems; research by Wilson, Hoffman, and Morgenstern20 on higher error rates for pedestrians with darker skin tones in object detection systems; and Bolukbasi et al.’s21 research on gender stereotypes in word embeddings.
The concern for ethics in AI, algorithms, and automated systems is also amplified by scandals that have shaken the tech industry, such as the Cambridge Analytica scandal involving Facebook user data, civilian deaths through driverless cars or the automated replication of the live-streaming of the Christchurch mosque attacks on social media. Meanwhile, the rollout of Europe’s General Data Protection Regulation (GDPR) has brought data protection issues to a broad audience.
1.3 National AI Strategies
Many efforts to address issues around AI and society are now streamlined in and through national AI strategies. Therefore, this chapter provides a qualitative analysis of existing national AI strategies with a specific focus on ethics, and ethics-related concerns. It sets out to examine what work “ethics” do in national AI strategies and identify broad patterns of AI ethics interpretation and representation within these strategy documents.
The empirical material for this study is comprised of national AI strategy documents that were sourced through an online search22 (between February and March 201923). In order to be included in the sample, a nation had to have a formal strategy in place, and the AI strategy documents had to be available in English. After the completion of the data collection, the AI strategy documents were analysed to identify aspects of “ethics” or related concerns and approaches and define core themes that cut across the sample. To account for the AI innovation landscape beyond formalised national AI strategies, additional data was gathered from policy documents, reports and news articles. This chapter should not be read as a comprehensive analysis of all AI strategies that have been proposed globally. It focuses explicitly on how concerns around AI and society, and ethics specifically, are articulated in the national AI strategies that were available at the time this study was conducted. It is therefore limited in its scope.
1.3.1 Defining National AI Strategies
At the most basic level, national AI strategies are frameworks that facilitate the distribution of public funds and incentivise research and innovation, as well as private funding, in certain areas and into certain directions. Bradley, Wingfield, and Metzger24 broadly define a national AI strategy...