Open Access Open Access  Restricted Access Subscription Access

The Ethical Spectrum of Artificial Intelligence: A Review of Divergent Discourses and a Unified Framework for Risk

Anshul Desai, Gajanan M Naik

Abstract


It’s often said that debates about the ethics and safety of artificial intelligence (AI) are split into two hostile camps. On one side, you have a community focused on the harms we can see right now: discrimination in automated decision-making, the rapid growth of surveillance infrastructures, and the sheer opacity of these algorithmic systems. On the other side, you have researchers worrying about the long game—specifically, the catastrophic or even existential risks that might come from highly capable AI. This paper suggests, however, that treating these as separate conversations hides an important continuity. Drawing on work in algorithmic fairness, critical data studies, and AI alignment, we develop what we call an Ethical Spectrum of AI that links local harms, systemic effects, and catastrophic risks through a small set of recurring mechanisms. We reread near-term case studies in criminal justice, healthcare, and commercial AI as clear instances of misaligned optimisation around imperfect proxy objectives, operating inside particular institutional and economic constraints [1]–[3], [5]. Then, we interpret long-term safety research as an extension of these concerns to more capable systems that might exhibit specification gaming, reward hacking, or deceptive alignment [9]–[13]. We also take a look at how contemporary governance instruments—including the OECD Recommendation on AI, the EU Artificial Intelligence Act, UNESCO’s Recommendation on the Ethics of AI, and the U.S. Executive Order 14110—map onto this spectrum and where they leave gaps [17]–[20]. Thus, the paper proposes a set of joint design goals for policy and system design, aimed at remaining robust across the Ethical Spectrum. These include treating data, models, and compute as a linked governance space; adopting full-spectrum evaluation and red-teaming; building institutional channels for contestation and oversight; incorporating long-term safety concerns into current deployments; and strengthening international coordination and capacity-building.


Full Text:

PDF

References


J. Angwin, J. Larson, S. Mattu, and L. Kirchner, “Machine bias: There’s software used across the country to predict future criminals. And it’s biased against blacks,” ProPublica, May 2016.

V. Eubanks, Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor. New York, NY, USA: St. Martin’s Press, 2018.

J. Buolamwini and T. Gebru, “Gender shades: Intersectional accuracy disparities in commercial gender classification,” in Proc. 1st Conf. Fairness, Accountability and Transparency (FAT)*, 2018, pp. 77–91.

F. Pasquale, The Black Box Society: The Secret Algorithms That Control Money and Information. Cambridge, MA, USA: Harvard Univ. Press, 2015.

Z. Obermeyer, B. Powers, C. Vogeli, and S. Mullainathan, “Dissecting racial bias in an algorithm used to manage the health of populations,” Science, vol. 366, no. 6464, pp. 447–453, 2019.

S. Zuboff, The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. New York, NY, USA: PublicAffairs, 2019.

S. Barocas, M. Hardt, and A. Narayanan, Fairness and Machine Learning: Limitations and Opportunities. Cambridge, MA, USA: MIT Press, 2023.

A. D. Selbst, D. Boyd, S. A. Friedler, S. Venkatasubramanian, and J. Vertesi, “Fairness and abstraction in sociotechnical systems,” in Proc. Conf. Fairness, Accountability, and Transparency (FAT)*, 2019, pp. 59–68.

N. Bostrom, “Existential risk prevention as global priority,” Global Policy, vol. 4, no. 1, pp. 15–31, 2013.

D. Amodei et al., “Concrete problems in AI safety,” arXiv preprint arXiv:1606.06565, 2016.


Refbacks

  • There are currently no refbacks.