IPHS 300: AI for Humanity
Document Type
Poster
Publication Date
Spring 2026
Abstract
As LLMs are deployed globally, the moral framework they u se matters. This study tested if four frontier LLMs default to Western ethics and how naming Ubuntu, a communal Southern African philosophy, changes their moral reasoning. Five dilemmas were presented to Claude Opus 4.7, GPT-5.4, Gemini 2.5 Flash, and DeepSeek V3, unprompted and with Ubuntu named. Responses were scored on seven Ubuntu dimensions (e.g., relational inclusion, dignity, reciprocity). All models defaulted to Western individualist reasoning when unprompted. With Ubuntu named, DeepSeek (57/70) and Claude (56/70) significantly outperformed GPT and Gemini (32/70 each). Gaps were largest in reciprocity and relational inclusion. Findings suggest current LLM cultural responsiveness is uneven and often superficial.
Recommended Citation
Ampofo-Twumasi, Davelle Akosua, "Whose Ethics? Auditing the Moral Defaults of Large Language Models Against Ubuntu Philosophy" (2026). IPHS 300: AI for Humanity. Paper 60.
https://digital.kenyon.edu/dh_iphs_ai/60
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.