Abstract
As artificial intelligence transitions from controlled laboratory environments to high-stakes real-world applications—including medicine, management, and warfare—the practitioners building these systems face increasingly complex moral dilemmas. This paper argues that the current reliance on utilitarian cost-benefit analysis, which mirrors the optimization logic inherent in computer science, is insufficient. By introducing AI researchers to a broader range of ethical frameworks, specifically deontology and virtue ethics, the field can move toward more responsible design and career decision-making. Using the 2015 debate over weaponized AI as a case study, we demonstrate how multi-modal ethical analysis provides a more robust foundation for professional integrity than optimization alone.
Introduction
AI practitioners are often ill-equipped to handle the moral weight of their creations. Most computer science curricula focus on technical proficiency, leaving a gap where a formal toolkit for ethical reasoning should be. In the absence of such a toolkit, researchers often default to acting as "greedy agents," making decisions based on immediate utility—such as securing research funding—or a vague, unexamined desire to "do the most good."
This paper addresses the need for a more nuanced approach to ethics in AI. Rather than viewing ethics as a binary "yes/no" decision-making process, we propose that practitioners engage in a comparative analysis using three major schools of Western ethical thought: utilitarianism, deontology, and virtue ethics.
The Optimization Trap
A primary reason AI researchers gravitate toward utilitarianism is its structural similarity to the field's core methodologies. Utilitarianism, which seeks to maximize happiness or minimize pain for the greatest number, maps directly onto decision-theoretic planning and Markov Decision Processes (MDPs).
When researchers are trained to optimize for an "optimal policy," cost-benefit analysis feels like the only rational way to approach ethics. However, this "optimization trap" can lead to a narrow focus on outcomes while ignoring the inherent duties of the practitioner or the long-term impact of their work on their own professional character.
Ethical Frameworks as Analytical Tools
To move beyond simple optimization, the authors apply three distinct lenses to the historical 2015 open letter petitioning the UN to ban weaponized AI.
1. Utilitarianism
This framework evaluates the consequences of an action. In the context of weaponized AI, a utilitarian might weigh the potential for reduced human casualties against the risk of an automated arms race. The goal is to achieve the highest net utility, though this often requires making difficult trade-offs that can be hard to quantify accurately.
2. Deontology
Deontology evaluates actions based on adherence to universal rules or duties, regardless of the outcome. For a programmer, deontology can be viewed as an axiomatic system. If one adopts the axiom "do not kill," then building lethal autonomous systems is a violation of that duty, even if such systems could theoretically end a war faster. This framework explains why a researcher might refuse military funding on principle, even if that funding could support beneficial, non-combat AI research.
3. Virtue Ethics
Virtue ethics focuses on the "arc of a life" and the development of individual character. It asks: What kind of person does this choice make me? The authors argue that building technology is an internal act as much as an external one. If a practitioner spends a career developing lethal automation, they may eventually adapt their sense of what is "normal," becoming the kind of person who views mechanized warfare as a standard solution to human conflict. Virtue ethics encourages practitioners to consider how their professional choices shape their moral identity over time.
The Funding Dilemma and Professional Responsibility
The paper acknowledges the practical realities of the field, specifically the "selfish" versus "communal" aspects of research funding. Choosing to accept or reject military grants is rarely a simple choice; it involves the ability to support students, maintain research communities, and advance the state of the art.
By applying different ethical frameworks, practitioners can reach a more comprehensive position. For instance, under Just War Theory (a deontological branch), a researcher might justify refusing a ban on weaponized AI if they believe the technology could make strikes more targeted and reduce civilian collateral damage. Conversely, a virtue ethicist might sign the ban not because they believe it will be effective, but because they refuse to be a silent participant in the mechanization of death.
Conclusion and Future Directions
Teaching these frameworks does not provide a single "correct" answer to the dilemmas of the AI age. Instead, it provides practitioners with critical tools to interrogate how their work shapes society and creates new moral crises.
The authors suggest that the next step for the AI community is the integration of these frameworks into standard curricula. One proposed pedagogical tool is the use of science fiction, which allows students to explore these complex ethical landscapes in a low-stakes, imaginative environment before they encounter real-world professional crises. By broadening the ethical education of AI practitioners, we can move toward a future where technical innovation is guided by a sophisticated and multi-faceted moral compass.
Summary updated