Generative AI systems are more than just open and closed source, and the latest leaks about Google and OpenAI’s plans for a powerful AI just prove that discussions within the AI community about how to properly share these systems with researchers and the public are still ongoing. From the fully closed systems to the fully open ones, there is a wide gradient of options when it comes to releasing AI technology, and each one comes with its own unique set of benefits and potential harms.
The most closed systems are almost all unfamiliar to the public. This makes sense considering the risk of misuse and potential harm associated with having access to such powerful AI technology. However, there are more and more publicly announced closed systems, like Meta’s Make-a-Video model, that are built with the idea of gradual access to researchers.
Perhaps the most familiar systems to casual users are those that are publicly accessible and hosted by organizations like OpenAI and Midjourney. While these systems give users a convenient user interface, they can also be susceptible to misuse. If these systems are not hosted in a responsible manner, then the data they contain may not be appropriately protected or well understood.
Finally, there are the completely open systems. These systems, pioneered by organizations like Hugging Face and EleutherAI, are accessible in all of their components – models, code, and training data. Open development decentralized power so that many people can collectively work on AI systems to make sure they reflect their needs and values, making these systems ideal for research and development.
Organizations like OpenAI and Hugging Face are just two examples in the AI community. OpenAI, a leader in artificial intelligence research, and Hugging Face, an AI company focused on democratization and transparency, have each emerged as pioneers while developing and releasing powerful AI systems. Meta, Google, and Midjourney are some of the other organizations that are pushing forward the research and discussions on responsibly deploying increasingly powerful generative AI systems.
Though there is no consensus on the best practices for releasing systems, there are important considerations when it comes to open source and closed source systems. These considerations range from safety, security, and ethical openness to the need for collaboration between researchers, developers, and the public. Ultimately, the debate ultimately comes down to striking the right balance to ensure AI systems are responsibly and safely utilized.
