I gave the opening keynote at the ALPSP University Press Redux conference in Liverpool earlier today, discussing how generative AI is affecting scholarly publishing. My professional background is in trade rather than scholarly publishing, so instead of focusing on the university press business model or specific AI tools, I tried to step back and look at a macro question: what happens when AI changes both how knowledge is produced and how it is discovered.

That framing implies two axes of uncertainty. The first is where discovery happens: do researchers continue to work primarily within traditional academic platforms, or does discovery migrate to general AI assistants and agents? The second is whether AI amplifies signal through better synthesis and analysis, or overwhelms existing filtering mechanisms with sheer noise. Combining those uncertainties produces four plausible scenarios. Of course, the real world is never as neat as a strategist’s Powerpoint slide, but the exercise creates a framework for stress-testing strategy under uncertainty that admits a range of possibilities, rather than the false precision of a forecast. Many of the same questions recur across the scenarios: where does discovery begin, what gets monetised, and what capabilities actually matter for publishers?

My argument is that the enduring role of a publisher is not the production of content but the exercise of judgement: deciding what deserves trust, attention, distribution and preservation. The challenge I suggested this morning is making sure that judgement remains visible and usable as research workflows become increasingly mediated by AI models.

My slides from the day are available here.