Free Software Foundation Says 'Responsible AI' Licenses Which Restrict Harmful Uses are Unethical and Nonfree

Source: Slashdot

Background

The Free Software Foundation’s Licensing and Compliance Manager published a blog post stating that “Responsible AI” licenses (RAIL) are nonfree and unethical. The licenses aim to restrict AI and machine learning software from being used in a specific list of harmful applications—e.g., surveillance and crime prediction—according to the license’s website¹. The steering committee for the licenses is composed of volunteers from multiple academic institutions.

FSF’s Position on Responsible AI Licenses

The FSF argues that, despite being marketed as a solution to ethical challenges, RAIL do not require any of the essential freedoms needed for users to control their machine‑learning computing. Specifically, they omit:

• Access to complete training inputs
• Access to training configuration settings
• Access to the trained model
• Access to the source code of the software used for training, testing, and running ML tools

Because these core freedoms are missing, the FSF concludes that RAIL‑restricted machine learning is likely to be unethical. Use‑case restrictions do not stop the licenses from being employed as a means of exercising power over users.

Why Use Restrictions Are Insufficient

RAIL licenses contribute to the unethical marketing of machine learning by disguising morally‑loaded restrictions as ethical safeguards. Limiting how software can be used does not address the root causes of social injustice. Instead, it:

• Prevents users from employing the software for the common good.
• Undermines the collaborative development of tools that could mitigate moral dangers.
• Reduces the pool of truly free software, limiting real choices for users seeking freedom‑respecting ethical programs.

Recommendations for Promoting Software Freedom

To advance software that helps decrease social injustice, the FSF recommends focusing on:

1. Government and community support for freedom‑respecting tools and services.
2. Releasing programs under strong copyleft licenses that guarantee the four essential software freedoms.
3. Entrusting copyrights to organizations with the resources to enforce copyleft provisions.

Increasing the amount of free software makes it more likely that collaborative tools and services will emerge without moral hazards, and that users will have genuine options for ethical machine‑learning applications.

Footnotes

1. FSF blog post: “Responsible AI Licenses (RAIL) are nonfree and unethical.”
   License website ↩