Can AI make us “more human”? The paradox of artificial moral enhancement by Veronica Dolcini
The hypothesis of moral enhancement mediated by external artificial technologies might sound like a disturbing paradox: why should we entrust technology with the task of making us better, or more “human”? At first sight, this scenario might even seem to involve a certain circularity: as increasingly powerful AI systems require adequate moral capacities if they are to be used and governed responsibly, AI itself contributes to making more urgent the very moral enhancement required to govern its use, while at the same time being proposed as one of the means through which such enhancement might be pursued. Yet it is precisely within the tension contained in this apparent contradiction that the most audacious — and perhaps also the most effective — solution for genuine moral enhancement may lie.
An artificial moral advisor would take the form of an intelligent system designed to foster forms of ethical deliberation that are more lucid, informed, and capable of taking into accountcomplex and morally relevant variables that may escape our immediate intuition. It is now widely recognized, indeed, that human beings are often unable to make choices fully consistent with their own moral aims, particularly when called upon to decide under conditions of urgency or severe cognitive constraints, such as information overload, high levels of stress, reduced attention, and the influence of cognitive biases. As a result, we very frequently find ourselves — as Giubilini and Savulescu observe — to be suboptimal information processors, moral judges, and decision-makers. We are limited epistemic subjects insofar as we do not systematically consider all the information relevant to the decisions we are called upon to make. We may prove to be imperfect moral judges because we fail to bring our judgments into conformity with normative principles that we ourselves recognize as valid. Finally, we are suboptimal decision-makers whenever, despite having the possibility of implementing what we regard as the best choice, we fall prey to weakness of will or to cognitive and emotional factors that hinder the translation of judgment into action.
To this must be added the problem of adaptive preferences, highlighted by Jon Elster and Amartya Sen: what individuals prefer is inevitably shaped by the social and environmental conditions in which they are situated — hence, their choices may often fail to coincide with those they would make if they were more fully aware of their own situation.
It is precisely at this point that the possibility arises of a form of artificial intelligence capable of gathering information from the environment, processing it according to specific operational criteria that we provide — for example, values, goals, and moral principles — and advising us on the best option, namely, the one that best satisfies our moral criteria. Such data would include information concerning the agent and their environment, their moral principles and values, as well as common cognitive biases liable to influence the moral decision-making process. Moral AI could thus alert the agent to potential influences and distortions, suggest strategies aimed at mitigating their effects and, where requested, advise them on particular courses of action.
The use of artificial systems in deliberative contexts is, moreover, by no means such a remote possibility. As Giubilini and Savulescu observe in The Artificial Moral Advisor. The “Ideal Observer” Meets Artificial Intelligence, such systems can often be understood as forms of ambient intelligence: a type of artificial intelligence that operates through “context awareness” and that, in other words, is capable of recognizing and interpreting elements concerning the user and the surrounding environment, modelling them through digital representations and continuously processing the information acquired, before interacting with the human being in accordance with specific criteria and requirements.
To illustrate concretely how such a technology might operate, the authors provide two examples, differing in degree of complexity but sharing the same functional structure.
First, they identify a relatively rudimentary form of ambient intelligence in applications already used in everyday life, focusing on a device designed to find a restaurant that simultaneously satisfies a plurality of user preferences: geographical proximity, price range, and food quality. In this case, the system acquires information concerning the environment, organizes and models it on the basis of the criteria provided, performs the required task, and ultimately returns a set of options compatible with those parameters. The same functional structure characterizes, at a much higher level of complexity, decision-support systems such as IBM Watson, conceived to interpret natural language, integrate information from different sources, formulate and compare hypotheses, and assist human beings in making complex decisions. In both of the scenarios described, the process retains an analogous structure: the system acquires information, models it in light of the criteria provided, performs certain operations upon it, and returns an output consistent with those criteria.
Although the aforementioned cases do not involve any kind of moral input, there is nothing to prevent the same type of technology and the same functions from being employed to assist us in making decisions based on moral requirements and instructions. The underlying operations would remain essentially the same, with the crucial difference that both the criteria taken into account by the software and the results it is expected to provide would be moral in nature.
The paradox outlined at the beginning of this article can therefore be at least partly resolved. Technology could make us, in a specifically moral sense, “more human” not by assuming responsibility for the choice itself, nor by rendering us incapable of error, but by helping to strengthen precisely those capacities — to understand, question, compare, justify, and revise — through which we can recognize ourselves as the authors of our own judgments and whose full exercise is often hindered by the finitude of our cognitive and temporal resources, as well as by the influence of biases and other psychological constraints.
Understood in these terms, an AI moral advisor could invite us to reconceptualize artificial intelligence as an interlocutor that restores to individuals, with renewed clarity, both the honor and the duty of ethical reasoning.