LLM needs a grandmother to prevent knowledge collapse
8 hours ago
- #AI and Knowledge Collapse
- #Grandmother Agents
- #Evolutionary Anthropology
- The paper 'AI, Human Cognition and Knowledge Collapse' by Acemoglu, Kong, and Ozdaglar discusses how agentic AI can improve individual decisions while eroding collective knowledge.
- The model assumes homogeneous human learners without considering overlapping generations or preference heterogeneity, missing the role of 'grandmother agents' who ensure the survival and quality of knowledge.
- Evolutionary anthropology highlights the importance of post-reproductive specialists (grandmothers) in maintaining public knowledge, a factor absent in the paper's model.
- The paper proposes 'garbling' AI recommendations to preserve learning incentives, but the grandmother mechanism offers a more natural solution by providing a public-knowledge floor.
- The absence of grandmother agents in AI development mirrors their omission in the paper, suggesting a need for age-structured and expertise-structured infrastructure in AI systems.