At Media Party Buenos Aires 2024, Andrés Cotton explored the moral system behind Artificial Intelligence (AI) and how we define the “greater good” in its development. Through experiments based on the Oxford Utilitarianism Scale, the speaker analyzed whether these models possess their own ethical vision, demonstrating that they do not reflect a specific, pre-determined philosophy but rather align with the morality of the general population thanks to the Reinforcement Learning with Human Feedback (RLHF) training process.
Based on this, Cotton concluded that AI actually functions as a mirror of what we want it to be, projecting our own priorities and desires. To guide this technology toward a true collective benefit, he proposed fostering cognitive diversity within technical teams, ensuring a heterogeneity of thought that enriches algorithmic design.
The author explained that diverse groups systematically outperform homogeneous ones when solving complex problems, provided they are managed under conditions of a common language and a combinatory approach. In conclusion, the talk emphasized that the best way to face the massive ethical challenges of AI is to build pluralistic teams that approach these tools from a broad, generous, and collaborative perspective.
Don’t miss the full presentation! Watch Andrés Cotton’s talk at Media Party Buenos Aires 2024 (English subtitles available).
Speaker Profile
Andrés Cotton
Researcher, Torcuato Di Tella University
