I predicted that people would change their search behaviour with the adoption of AI tools, and this behaviour change would create pressure on us to change how our systems are designed. There are plenty of examples of that happening now - often seen in the search query logs where searchers are moving from keyword based to natural language based. 

One thing I didn't think about is how our infrastructure itself can introduce change. 

This piece from Wiley is fascinating- https://www.wiley.com/en-gb/publish/ai-insights/ai-research-discovery-scholar-gateway-insights/?erg. It's not unbiased as Wiley are betting on their scholar gateway (a system to allow interoperability of scholarly content with GenAI agents), but there are few people with direct access to how these systems work, so it's interesting nonetheless.

- AI re-writes the original queer from the user, and can generate broader search patterns. 
- Usage is no longer linear as a result. 
- There is a tradeoff - the scholar gateway is now one step removed from the user, but it does allow users in different contexts to access content in new ways. 

It's early days, but it's an interesting writeup. 

This paper from James Evans garnered a lot of attention over the last few days - https://www.science.org/content/article/ai-has-supercharged-scientists-may-have-shrunk-science. 

One potential way to help with the convergence trap that Evan's describes is to make the broad body of literature more accessible to AI tools.