Artificial Intelligence is challenging open source because it ruptures the boundary between software and data. As a result, it's not clear what it actually means for an AI to be "open source". Existing open source licenses cannot cope because it's not even clear what the "source code" of an AI model actually is. Additionally, applying copyleft to AI models results in a number of unanswered questions such as "should the training data be open too?" and "does the license of training data affect the resulting models?" To make matters worse, large players such as Meta are muddying the waters by calling their AI "open source", even though they restrict commercial use and do not provide source code.
This talk takes a look at why AI creates such a seismic shift in open source and what makes it so different from regular software. It analyses what it means for AI to be transparent and open and explains the OSI efforts to define "Open Source AI". Finally, this talk showcases current research efforts to enhance AI transparency and supply-chain security.
Dr. Merlijn Sebrechts is a senior researcher at imec and teaches at Ghent University. He develops deployment platforms and helps companies streamline how they deliver software in the cloud and on devices. He teaches topics such as Systems Design, Open Source ecosystems, and Computer Security. Merlijn has extensive experience building open source communities and is currently serving on the Ubuntu Community Council. He's also worked on a number of open source projects and is currently standardizing WebAssembly System Interfaces for IoT devices together with the W3C and the Bytecode Alliance.
Livestream: