Microsoft and GitHub accused of violating open-source licenses

This is the story of the week. But it could cost Microsoft dearly. Developer and lawyer Matthew Butterick is suing the publisher as well as its subsidiary GitHub and its partner OpenAI for violation of open-source licenses by GitHub’s AI Copilot!

Launched in June 2022, GitHub’s Copilot service is an Artificial Intelligence programming aid that automatically generates source code from comments. In other words, it transforms natural language into lines of programming in a dozen computer languages. This is assistance that can save developers a lot of time.

This service is based on OpenAI’s AI Codex. The latter was trained from billions of lines of code published on the public repositories of GitHub. And that is the problem. Because in its generation of source code, the AI ​​does not attribute to such and such an author the lines which “inspired” him. According to some specialists, it therefore commits a violation of the terms of open source licenses such as GPL, Apache and MIT which require the retention of copyrights and the recalling of the names of the authors. Some developers have not hesitated to speak of “open source laundering”.

For Joseph Saveri, the firm representing Matthew Butterick in this litigation, “ Microsoft is taking advantage of the work of others by ignoring the terms of the underlying open-source licenses and other legal requirements “.

In September, a professor at the University of Texas Computer Science had already complained on social networks that Copilot could generate large portions of code directly taken from its source code without any attribution.

According to Matthew Butterick, Copilot’s AI violates several rules and licenses:
– GitHub’s own privacy and service policies
– DMCA 1202 which prohibits any removal of copyright information
– The Californian Consumer Privacy Act (slightly equivalent to the GDPR)
– Different laws on copyrights and developer rights.

In addition Matthew Butterick, definitely up against such “developer” AI, worries that such AI will end up preventing the discovery of open source communities since many developers turn to communities when they seek to shorten their time. of development and find ready-to-use code.

And the plaintiff to claim compensation of 9 billion dollars. According to the complaint filed, “ each time Copilot provides output, it three times violates Section 1202 (distribution of licensed materials without: (1) attribution, (2) copyright notice, and (3) license terms). So, if every user receives a single release that violates section 1202 for the entire lifetime of Copilot (up to fifteen months for early adopters), then GitHub and OpenAI have violated the DMCA 3,600,000 times. With minimum statutory damages of $2,500 per violation, that translates to $9,000,000,000. »

Microsoft hasn’t really reacted yet. The dispute also comes at a rather bad time when the publisher talked a lot about the potential of this technology during the last MS Ignite 2022 and intends to extend it to other tools including Visual Studio and its development tools. Low Code “Power”.

It should be noted in passing that the Copilot documentation points out that the generated code “may contain undesirable patterns”, placing the responsibility for copyright violations on the users of the AI ​​rather than on the AI ​​itself.

Also, more often than not, the AI ​​does not generate identical code but rather “similar” code. To what extent the generated code derives directly from the source codes used for training remains a question of interpretation.

One of the parades would be for Microsoft to entirely redo the training of Copilot so that they maintain the associations between blocks of code and their authors and that it automatically adds in comment the name of the authors who “inspired” the lines that she generated. A dantesque job at first sight.

To be continued…

Also read:

Microsoft finally launches its Volterra project, an ARM mini-PC for developers

Microsoft, impacted growth (a little)…

GPT-3, OpenAI’s amazing artificial intelligence that writes as well as humans…

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *