As the capabilities of artificial intelligence systems improve, it becomes important to constrain their actions to ensure their behaviour remains beneficial to humanity. A variety of ethical, legal and safety-based frameworks have been proposed as a basis for designing these constraints. Despite their variations, these frameworks share the common characteristic that decision-making must consider multiple potentially conflicting factors. We demonstrate that these alignment frameworks can be represented as utility functions, but that the widely used Maximum Expected Utility paradigm provides insufficient (...) support for such multiobjective decision-making. We show that a Multiobjective Maximum Expected Utility paradigm based on the combination of vector utilities and non-linear action–selection can overcome many of the issues which limit MEU’s effectiveness in implementing aligned AI. We examine existing approaches to multiobjective AI, and identify how these can contribute to the development of human-aligned intelligent agents. (shrink)

There is increasing speculation within military and policy circles that the future of armed conflict is likely to include extensive deployment of robots designed to identify targets and destroy them without the direct oversight of a human operator. My aim in this paper is twofold. First, I will argue that the ethical case for allowing autonomous targeting, at least in specific restricted domains, is stronger than critics have acknowledged. Second, I will attempt to uncover, explicate, and defend the intuition that (...) even in this context there would be something ethically problematic about such targeting. I argue that an account of the non-consequentialist foundations of the principle of distinction suggests that the use of autonomous weapon systems is unethical by virtue of failing to show appropriate respect for the humanity of our enemies. However, the success of the strongest form of this argument depends upon understanding the robot itself as doing the killing. To the extent that we believe that, on the contrary, AWS are only the means whereby those who order them into action kill, the idea that the use of AWS fails to respect the humanity of our enemy will turn upon an account of what is required by respect that is essentially conventional. Thus, while the theoretical foundations of the idea that AWS are weapons that are “evil in themselves” are weaker than critics have sometimes maintained, they are nonetheless sufficient to the task of demanding a prohibition of the development and deployment of such weapons. (shrink)

As engineers propose constructing humanlike automata, the question arises as to whether such machines merit human rights. The issue warrants serious and rigorous examination, although it has not yet cohered into a conversation. To put it into a sure direction, this paper proposes phrasing it in terms of whether humans are morally obligated to extend to maximally humanlike automata full human rights, or those set forth in common international rights documents. This paper’s approach is to consider the ontology of humans (...) and of automata and whether ontological difference between them, that pertains to the very bases of human rights, affects the latter’s claims to full human rights. Considering common bases of human rights, can these bases tell us whether a certain ontological distinction of humans from automata—or a de facto distinction about humans tacitly acknowledged by full-rights-recognizing societies—makes a difference in whether humans are morally obligated to assign these entities full rights? Human rights to security also arise. The conclusion is that humans need not be under any moral obligation to confer full human rights on automata. The paper’s ultimate point is not to close the discussion with this ontological cap but to set a solid moral and legal groundwork for opening it up tout court. (shrink)

Would it be ethical to deploy autonomous weapon systems (AWS) if they were unable to reliably recognize when enemy forces had surrendered? I suggest that an inability to reliably recognize surrender would not prohibit the ethical deployment of AWS where there was a limited window of opportunity for targets to surrender between the launch of the AWS and its impact. However, the operations of AWS with a high degree of autonomy and/or long periods of time between release and impact are (...) likely to remain controversial until they have the capacity to reliably recognize surrender. (shrink)