The chapter provides a brief overview of the structure and logic of the book by indicating the content of the cases covered in each section. It concludes by identifying the concept of ethics used in this book and how it is located in the broader discussion of ethics, human rights and regulation of AI.
One of the big problems that AI ethics and ethicists might face is the opaqueness of what is actually happening in AI, given that a good grasp of an activity itself is very helpful in determining its ethical issues. There is thus an expectation that AI ethicists are familiar with the technology, yet no one really knows how the most advanced algorithms do what they do, including AI developers themselves.
Despite this opacity of AI in its current forms, it is important to reflect on and discuss which ethical issues can arise due to its development and use. The approach to AI ethics we have chosen here is to use case studies. This approach will enable us to illustrate the main ethical challenges of AI, often with reference to human rights.
Case studies are a proven method for increasing insights into theoretical concepts by illustrating them through real-world situations. They also increase student participation and enhance the learning experience and are therefore well-suited to teaching. We have therefore chosen the case study method for this book. We selected the most significant or pertinent ethical issues that are currently discussed in the context of AI and dedicated one chapter to each of them.
The structure of each chapter is as follows. First, we introduce short real-life case vignettes to give an overview of a particular ethical issue. Second, we present a narrative assessment of the vignettes and the broader context. Third, we suggest ways in which these ethical issues could be addressed. This often takes the form of an overview of the tools available to reduce the ethical risks of the particular case.
Below we present a short overview of the cases:
Unfair and Illegal Discrimination (Chap. 2)
The first vignette deals with the automated shortlisting of job candidates by an AI tool trained with résumés from the previous ten years. Notwithstanding efforts to address early difficulties with gender bias, the company eventually abandoned the approach as it was not compatible with their commitment to workplace diversity and equality.
The second vignette describes how parole was denied to a prisoner with a model rehabilitation record based on the risk-to-society predictions of an AI system. It became clear that subjective personal views given by prison guards, who may have been influenced by racial prejudices, led to an unreasonably high risk score.
The third vignette tells the story of an engineering student of Asian descent whose passport photo was rejected by New Zealand government systems because his eyes were allegedly closed. This was an ethnicity-based error in passport photo recognition.
Privacy (Chap. 3)
The first vignette is about the Chinese social credit scoring system, which uses a large number of data points to calculate a score of citizens’ trustworthiness. High scores lead to the allocation of benefits, whereas low scores can result in the withdrawal of services.
The second vignette covers the Saudi Human Genome Program, with predicted benefits in the form of medical breakthroughs versus genetic privacy concerns.
Surveillance Capitalism (Chap. 4)
The first vignette deals with photo harvesting from services such as Instagram, LinkedIn and YouTube in contravention of what users of these services were likely to expect or have agreed to. The relevant AI software company, which specialises in facial recognition software, reportedly holds ten billion facial images from around the world.
The second vignette is about a data leak from a provider of health tracking services, which made the health data of 61 million people publicly available.
The third vignette summarises Italian legal proceedings against Facebook for misleading its users by not explaining to them in a timely and adequate manner, during the activation of their account, that data would be collected with commercial intent.
Manipulation (Chap. 5)
The first vignette covers the Facebook and Cambridge Analytica scandal, which allowed Cambridge Analytica to harvest 50 million Facebook profiles, enabling the delivery of personalised messages to the profile holders and a wider analysis of voter behaviour in the run-up to the 2016 US presidential election and the Brexit referendum in the same year.
The second vignette shows how research is used to push commercial products to potential buyers at specifically determined vulnerable moments, e.g. beauty products being promoted at times when recipients of online commercials are likely to feel least attractive.
Right to Life, Liberty and Security of Person (Chap. 6)
The first vignette is about the well-known crash of a Tesla self-driving car, killing the person inside.
The second vignette summarises the security vulnerabilities of smart home hubs, which can lead to man-in-the-middle attacks, a type of cyberattack in which the security of a system is compromised, allowing an attacker to eavesdrop on confidential information.
The third vignette deals with adversarial attacks in medical diagnosis, in which an AI-trained system could be fooled to the extent of almost 70% with fake images.
Dignity (Chap. 7)
The first vignette describes the case of an employee who was wrongly dismissed and escorted off his company’s premises by security guards, with implications for his dignity. The dismissal decision was based on opaque decision-making by an AI tool, communicated by an automatic system.
The second vignette covers sex robots, in particular whether they are an affront to the dignity of women and female children.
Similarly, the third vignette asks whether care robots are an affront to the dignity of elderly people.
AI for Good and the UN’s Sustainable Development Goals (Chap. 8)
The first vignette shows how seasonal climate forecasting (SCF) in resource-limited settings has led to the denial of credits for poor farmers in Zimbabwe and Brazil and the accelerated the layoff of workers in the fishing industry in Peru.
The second vignette deals with a research team from a high-income country requesting vast amounts of mobile phone data from users in Sierra Leone, Guinea and Liberia to track population movements during the Ebola crisis. Commentators argued that the time spent negotiating the request with seriously under-resourced governance structures should have been used to handle the escalating Ebola crisis.
And the concluding chapter (Chap. 9) should not be missed.
A predominant approach to AI ethics is the development of guidelines, most of which are based on mid-level ethical principles typically developed from the principles of biomedical ethics. This is also the approach adopted by the European Union’s High-Level Expert Group (HLEG) on AI. The AI HLEG’s intervention has been influential, as it has had a great impact on the discussion in Europe. However, there has been significant criticism of the approach to AI ethics based on ethical principles and guidelines. One key concern is that it remains far from the application and does not explain how AI ethics can be put into practice. With the case-study-based approach presented in this book, we aim to overcome this point of criticism, enhance ethical reflection and demonstrate possible practical interventions.
Overall, AI is an example of a current and dynamically developing technology. An important question is therefore whether we can keep reflecting and learn anything from the discussion of AI ethics that can be applied to future generations of technologies to ensure that humanity benefits from technological progress and development and has ways to deal with the downsides of technology.