---
title: "Boycott Routledge over AI training? It's probably in the contract..."
layout: post
image: 
    feature: header_ai.png
---
There's a movement at the moment on social media where angry academic authors are gathering with the intent to boycott Routledge, who are apparently distributing academic works for training in AI.

I say: read your publishing contracts. I believe that Routledge's default agreement certainly allows this. An example contract that I signed with them reads:

"Copyright in the Work will remain with the author, who licenses the Publisher to make a copy available in any format worldwide on a non-exclusive basis. This permits the Publisher to distribute the Work in any format at any time."

This is actually a more liberal clause than many publication agreements, since "the Work may be licensed or made available by the Contributor(s) to another party at any time" whereas many publication agreements are exclusive and do not permit this.

But I digress. The point is: the license here permits the Publisher (Routledge) to distribute a copy _in any format_ to anybody. It doesn't say "distribute a copy in any format worldwide for the purpose of human reading only" but only, paraphrased, "to make a copy available \[to anybody they choose\] in any format worldwide". This will include making copies available in digital indexes, search engines, and other digital systems, from which AI is no different. The license says nothing about restricting downstream re-uses. Let's say a new AI Audiobook system came out that required the text to be imported; would that be allowed? It seems to be part of the "in any format" clause. Indeed, we could argue that ingesting books into AI systems is just another "format"; a linear regression model based on the textual vectors of the work in question. A mathematical "format" of the book, combined with many others.

The other challenge is that we don't have any regulation yet that says that a special license is required to train AI on texts. It's possible that the EU will introduce one (I bet that the US courts will say that it is transformative use), but they will also be wary of existing contractual wordings and what this might introduce. I mean, if you can't use a non-exclusive perpetual worldwide license to publish in any format to do AI, then what other restrictions are there on what has typically been considered an unrestricted licensing covenant?

I don't know where I stand on all this and it may be that authors have just cause to feel morally outraged about their books being used as fodder for text generation (the most prominent, but by far the only, use of the latest AI systems). I just don't think that they have contractual grounds to dispute it with Routledge.