Four Analogies for AI Ethics: Tools, Animals, Humans, and Gods

On April 23, 2025, Fangtang Institute held the first lecture in its Wenqu series. Stephen Cave of the Leverhulme Centre for the Future of Intelligence at the University of Cambridge spoke on “Four Analogies for AI Ethics: Tools, Animals, Humans, and Gods.” Beginning with the ethical puzzle in Plato’s Euthyphro, he placed dialogue between humans and AI within a longer history of civilisation.

AI’s challenge to rule systems

Starting with Robbie the Robot in Forbidden Planet, Cave described an early AI imagined as a “precision tool,” with ethics closely tied to rule systems. Asimov’s Three Laws of Robotics appear to offer a technical safety net, but they also treat intelligent agents as fully programmable machines and overlook potential autonomy. Once machines act autonomously, simple rules cannot resolve the moral ambiguity of complex social settings.

After deep learning’s 2012 breakthrough, machine learning unsettled this rule-based framework. Transparency and explainability became central concerns. Cave compared black-box systems to domesticated animals: their many neural-network parameters make their decision-making difficult to understand. Training replaces programming, blurring the conventional chain of responsibility between designer and user. He recalled AlphaGo’s 2016 victory over Lee Sedol, when AI seemed to show an intuition that exceeded human capacities.

Large language models further alter the ethical relationship. People expect not only rule-following but honesty, fairness, and compassion. Drawing on Aristotle and Confucian ethics, Cave suggested that cultivating AI virtue may become a central question, while cautioning that genuine moral reflection can be displaced when ethical responses merely accommodate user preference.

The double bind of superintelligence

If AI exceeds human understanding, should people accept decisions they cannot explain, or insist on a human right to understanding? If human explainability is made absolute, AI may be confined to the limits of human cognition; if people blindly submit, they may become “moral infants” and lose independent judgement. Cave used AlphaGo’s celebrated “divine move” to show how human experts can shift from judges to interpreters, a transfer of authority that becomes especially consequential in medicine and law.

“We will gain immense power through AI, but we may also lose control by giving up understanding and criticism. It is a difficult choice: either limit AI to the range of human understanding, or embrace its power at the risk of losing moral autonomy.”

Questions from the room

An audience member asked whether people must surrender their own standpoint even when AI appears to make better decisions. Cave framed this as a question of technological determinism and human choice: accelerationist claims that AI must advance without ethical constraints are dangerous. AI should serve all humanity, and its risks require collective governance; imperfect regulation of nuclear technologies remains preferable to unrestrained development.

Other questions addressed those who do not trust AI in an AI-saturated future, and the engineers and companies that stand behind ostensibly private human-AI relations. Cave noted that AI may also be compared with corporations or bureaucracies: both can be opaque, difficult to control, and capable of amplifying small errors or biases at scale. Each analogy is limited, so AI ethics requires multiple conceptual frames.

A more plural global narrative

Cave was also asked why tools, animals, humans, and gods should be the only analogies. He described the first three as a rough history of AI ethics in the English-speaking West, while the divine analogy draws on the Platonic tradition. The Global AI Narratives project has sought different cultural metaphors with scholars in Egypt, Russia, China, and elsewhere. The Western hierarchy of god, human, animal, and plant differs from traditions such as Confucianism, Daoism, and Buddhism, which give greater weight to relations than to rank. Other communities have invoked metaphors such as octopuses and weaving looms.

Cave welcomed further ideas, including approaches rooted in ren in Confucianism or ziran in Daoism. Such perspectives can enrich global AI ethics and move beyond a Western-centred framework.