The Free Software Foundation finally has AI/machine learning apps on its radar

The Free Software Foundation announced Tuesday that it has begun work on “Freedom in Machine Learning Applications.” Or in particular a “statement” to be published on free machine learning applications for software and the associated scripts and training data.

The Free Software Foundation announced that they are striving for freedom in machine learning, limiting themselves not only to the software but also to the training data. They have spent the last few months debating the matter and are “close to concluding” their testimony.

As part of Tuesday’s FSF.org press release:

“All software included in a free ML application must provide every user with the four freedoms that define free software. This applies to both the software that processes training data and the software that interprets model parameters as context for prompts to produce human-usable results.” This is necessary but not sufficient. Additionally, based on our current understanding of ML applications, we cannot say that an ML application is “free” unless all of its training data and the associated scripts for processing it take all users into account. In addition, granting the four freedoms result in the release of the ML application containing the model parameters representing its training and allowing users to use and distribute the parameters and modified versions thereof.

ML applications that do not provide the four freedoms to all users are by definition non-free, even if their software components are free.”

The FSF efforts around AI/machine learning may well be another of their initiatives that comes too little, too late. Focuses largely on their praise/guidance, but without committing technical resources to the further development of free AI/ML software. And the FSF is preparing to make a statement now, around the end of 2024… How long has the AI/ML trend been going on? Their delay remains, just as their FSF holiday shopping guide last year recommended 802.11n Wi-Fi devices and USB-to-parallel printer cables as recommended hardware in recent years. The FSF has noble and great intentions, but often comes on the scene very late and has no significant significance for most end users. Just like their list of high-priority software projects often stagnates without action, while they largely focus on making bold public statements.

We’ll see if the upcoming FSF work on ensuring freedom in machine learning applications leads to improvements, but I wouldn’t hold my breath except for the very dedicated free software enthusiasts.