All AI-generated code should be Open Source

Generating code using artificial intelligence should be made available as open source, I firmly believe. There are multiple reasons behind this stance. Firstly, AI models were trained on a substantial amount of open source code scraped from the internet. This training data was obtained without obtaining permission or giving proper credit.

Additionally, the models were also trained on Stack Overflow and other sister projects under Creative Commons licenses that legally require attribution. Utilizing this collective knowledge to create proprietary, closed-source code appears to be an unethical practice. The second reason revolves around legality. Code generated solely by AI is not written by a human, which means it should not be eligible for copyright protection.

This principle in the US traces back to the Monkey Selfie case, which established that non-humans cannot hold copyrights. A similar argument was later cited by the U.S. Copyright Office in its decision regarding AI. Following this logic, Wikimedia Commons requires all purely AI-generated images to be tagged as public domain. However, copyright laws vary across different countries.

For example, the United Kingdom grants limited copyright protection to computer-generated works. To ensure fairness across all jurisdictions and to respect the open ecosystem that enabled the development of AI, developers and companies should explicitly release AI-generated code under open source licenses or place it in the public domain using the Creative Commons Zero (CC0) license.

It is important to note that code is not always entirely manual or entirely automated. However, developers are increasingly relying on AI without even reviewing the generated output or modifying it to suit their specific needs. Code produced through this process, where humans put minimal effort except writing prompts, should unquestionably be open source. If you find this piece intriguing, you can follow me on Twitter @jcubic and on LinkedIn.