AI4People—An Ethical Framework for a Good AI Society: Opportunities, Risks, Principles, and Recommendations

That AI will have a major impact on society is no longer in question. Current debate turns instead on how far this impact will be positive or negative, for whom, in which ways, in which places, and on what timescale. Put another way, we can safely dispense with the question of whether AI will have an impact; the pertinent questions now are by whom, how, where, and when this positive or negative impact will be felt.

In order to frame these questions in a more substantive and practical way, we introduce here what we consider the four chief opportunities for society that AI offers. They are four because they address the four fundamental points in the understanding of human dignity and flourishing: who we can become (autonomous self-realisation); what we can do (human agency); what we can achieve (individual and societal capabilities); and how we can interact with each other and the world (societal cohesion). In each case, AI can be used to foster human nature and its potentialities, thus creating opportunities; underused, thus creating opportunity costs; or overused and misused, thus creating risks. As the terminology indicates, the assumption is that the use of AI is synonymous with good innovation and positive applications of this technology. However, fear, ignorance, misplaced concerns or excessive reaction may lead a society to underuse AI technologies below their full potential, for what might be broadly described as the wrong reasons. This may cause significant opportunity costs. It might include, for example, heavy-handed or misconceived regulation, under-investment, or a public backlash akin to that faced by genetically modified crops (Imperial College 2017). As a result, the benefits offered by AI technologies may not be fully realised by society. These dangers arise largely from unintended consequences and relate typically to good intentions gone awry. However, we must also consider the risks associated with inadvertent overuse or wilful misuse of AI technologies, grounded, for example, in misaligned incentives, greed, adversarial geopolitics, or malicious intent. Everything from email scams to full-scale cyber-warfare may be accelerated or intensified by the malicious use of AI technologies (Taddeo 2018). And new evils may be made possible (King et al. 2018). The possibility of social progress represented by the aforementioned opportunities above must be weighed against the risk that malicious manipulation will be enabled or enhanced by AI. Yet a broad risk is that AI may be underused out of fear of overuse or misuse. We summarise these risks in Fig. 1 below, and offer a more detailed explanation in the text that follows.

Who We Can Become: Enabling Human Self-Realisation, Without Devaluing Human Abilities

AI may enable self-realisation, by which we mean the ability for people to flourish in terms of their own characteristics, interests, potential abilities or skills, aspirations, and life projects. Much as inventions, such as the washing machine, liberated people—particularly women—from the drudgery of domestic work, the “smart” automation of other mundane aspects of life may free up yet more time for cultural, intellectual and social pursuits, and more interesting and rewarding work. More AI may easily mean more human life spent more intelligently. The risk in this case is not the obsolescence of some old skills and the emergence of new ones per se, but the pace at which this is happening and the unequal distributions of the costs and benefits that result. A very fast devaluation of old skills and hence a quick disruption of the job market and the nature of employment can be seen at the level of both the individual and society. At the level of the individual, jobs are often intimately linked to personal identity, self-esteem, and social role or standing, all factors that may be adversely affected by redundancy, even putting to one side the potential for severe economic harm. Furthermore, at the level of society, the deskilling in sensitive, skill-intensive domains, such as health care diagnosis or aviation, may create dangerous vulnerabilities in the event of AI malfunction or an adversarial attack. Fostering the development of AI in support of new abilities and skills, while anticipating and mitigating its impact on old ones will require both close study and potentially radical ideas, such as the proposal for some form of “universal basic income”, which is growing in popularity and experimental use. In the end, we need some intergenerational solidarity between those disadvantaged today and those advantaged tomorrow, to ensure that the disruptive transition between the present and the future will be as fair as possible, for everyone.

What We Can Do: Enhancing Human Agency, Without Removing Human Responsibility

AI is providing a growing reservoir of “smart agency”. Put at the service of human intelligence, such a resource can hugely enhance human agency. We can do more, better, and faster, thanks to the support provided by AI. In this sense of “Augmented Intelligence”, AI could be compared to the impact that engines have had on our lives. The larger the number of people who will enjoy the opportunities and benefits of such a reservoir of smart agency “on tap”, the better our societies will be. Responsibility is therefore essential, in view of what sort of AI we develop, how we use it, and whether we share with everyone its advantages and benefits. Obviously, the corresponding risk is the absence of such responsibility. This may happen not just because we have the wrong socio-political framework, but also because of a “black box” mentality, according to which AI systems for decision-making are seen as being beyond human understanding, and hence control. These concerns apply not only to high-profile cases, such as deaths caused by autonomous vehicles, but also to more commonplace but still significant uses, such as in automated decisions about parole or creditworthiness.

Yet the relationship between the degree and quality of agency that people enjoy and how much agency we delegate to autonomous systems is not zero-sum, either pragmatically or ethically. In fact, if developed thoughtfully, AI offers the opportunity of improving and multiplying the possibilities for human agency. Consider examples of “distributed morality” in human-to-human systems such as peer-to-peer lending (Floridi 2013). Human agency may be ultimately supported, refined and expanded by the embedding of “facilitating frameworks”, designed to improve the likelihood of morally good outcomes, in the set of functions that we delegate to AI systems. AI systems could, if designed effectively, amplify and strengthen shared moral systems.

What We Can Achieve: Increasing Societal Capabilities, Without Reducing Human Control

Artificial intelligence offers myriad opportunities for improving and augmenting the capabilities of individuals and society at large. Whether by preventing and curing diseases or optimising transportation and logistics, the use of AI technologies presents countless possibilities for reinventing society by radically enhancing what humans are collectively capable of. More AI may support better coordination, and hence more ambitious goals. Human intelligence augmented by AI could find new solutions to old and new problems, from a fairer or more efficient distribution of resources to a more sustainable approach to consumption. Precisely because such technologies have the potential to be so powerful and disruptive, they also introduce proportionate risks. Increasingly, we may not need to be either ‘in or on the loop’ (that is, as part of the process or at least in control of it), if we can delegate our tasks to AI. However, if we rely on the use of AI technologies to augment our own abilities in the wrong way, we may delegate important tasks and above all decisions to autonomous systems that should remain at least partly subject to human supervision and choice. This in turn may reduce our ability to monitor the performance of these systems (by no longer being ‘on the loop’ either) or preventing or redressing errors or harms that arise (‘post loop’). It is also possible that these potential harms may accumulate and become entrenched, as more and more functions are delegated to artificial systems. It is therefore imperative to strike a balance between pursuing the ambitious opportunities offered by AI to improve human life and what we can achieve, on the one hand, and, on the other hand, ensuring that we remain in control of these major developments and their effects.

How We Can Interact: Cultivating Societal Cohesion, Without Eroding Human Self-Determination

From climate change and antimicrobial resistance to nuclear proliferation and fundamentalism, global problems increasingly have high degrees of coordination complexity, meaning that they can be tackled successfully only if all stakeholders co-design and co-own the solutions and cooperate to bring them about. AI, with its data-intensive, algorithmic-driven solutions, can hugely help to deal with such coordination complexity, supporting more societal cohesion and collaboration. For example, efforts to tackle climate change have exposed the challenge of creating a cohesive response, both within societies and between them. The scale of this challenge is such that we may soon need to decide between engineering the climate directly and designing societal frameworks to encourage a drastic cut in harmful emissions. This latter option might be undergirded by an algorithmic system to cultivate societal cohesion. Such a system would not be imposed from the outside; it would be the result of a self-imposed choice, not unlike our choice of not buying chocolate if we had earlier chosen to be on a diet, or setting up an alarm clock to wake up. “Self-nudging” to behave in socially preferable ways is the best form of nudging, and the only one that preserves autonomy. It is the outcome of human decisions and choices, but it can rely on AI solutions to be implemented and facilitated. Yet the risk is that AI systems may erode human self-determination, as they may lead to unplanned and unwelcome changes in human behaviours to accommodate the routines that make automation work and people’s lives easier. AI’s predictive power and relentless nudging, even if unintentional, should be at the service of human self-determination and foster societal cohesion, not undermining human dignity or human flourishing.

Taken together, these four opportunities, and their corresponding challenges, paint a mixed picture about the impact of AI on society and the people in it. Accepting the presence of trade-offs, seizing the opportunities while working to anticipate, avoid, or minimise the risks head-on will improve the prospect for AI technologies to promote human dignity and flourishing. Having outlined the potential benefits to individuals and society at large of an ethically engaged approach to AI, in the next section we highlight the “dual advantage” to organisations of taking such an approach.