AI Knowledge Collapse: Diversity under threat
Race and gender bias in AI gets a lot of attention but Knowledge Collapse also threatens diversity. Board rooms take note:
- Why you should care
- How it impacts your knowledge of the world
- Three ways to avoid it
Based on the publication ‘AI and the Problem of Knowledge Collapse’ by Andrew Peterson, Assistant Professor, University of Poitiers. Overview by Andrew Tollinton CEO, SIRV.
Visualisation content
Large language models (LLM), a form of AI, can help us learn. For example:
- It suggests search terms
- Interprets government data
- Aids journalists
The more we ‘go through’ and use LLMs to learn, the more it mediates our knowledge. Here’s the problem:
AI will recall facts that occur frequently within training data and struggle with long-tail knowledge. This leads to ‘streetlight effect’ where most search is done under the lighted area. Not because it’s more likely to be the answer but it’s easier to look there.
Therefore, this threatens:
- Fairness
- Diversity
- Innovation
- Heritage of human culture
To avoid Knowledge Collapse and promote diversity;
- Seek niche, specialised and eccentric perspectives.
- Avoid AI that is recursively interdependent; that uses other AI generated content as inputs.
- Ensure AI generated content is as representative as possible of the full distribution of knowledge.
This is an overview by Andrew Tollinton of the publication:
‘AI and the Problem of Knowledge Collapse’ by Andrew Peterson, Assistant Professor, University of Poitiers
About
SIRV uses AI to help risk managers anticipate, manage and recover from business interruption.
Andrew Tollinton is SIRV CEO and Chair, Artificial Intelligence in Risk Management, Institute of Strategic Risk Management.
