Who gets to control the future of AI?

Three of the most influential figures in technology—Geoffrey Hinton, Fei-Fei Li, and Andrew Ng—recently met to discuss whether the public should have open access to advanced AI systems. They are debating two main approaches. The first is closed-source, where a company develops a model behind locked doors. The second is open-weight, where a company releases the finished, trained model to the public. This allows anyone to download and use the AI on their own equipment, bypassing the need for the massive, expensive servers that were required to create it in the first place.

Think of an AI model like a professional chef. Training a model is like sending an apprentice to culinary school for years; it is incredibly expensive, time-consuming, and requires massive amounts of data. Once the chef is fully trained, they possess a set of internal connections—their knowledge, their palate, their techniques—that allow them to cook perfectly. These internal connections are called weights. In a closed system, you can hire the chef to cook for you, but you cannot learn their methods. In an open-weight system, the company effectively publishes the chef’s complete brain. Anyone can copy it, creating a version of the chef that works exactly like the original. This is a massive shortcut for anyone who wants to build new tools without starting from scratch.

The disagreement centers on a difficult trade-off. Safety experts argue that if powerful models are released to the public, bad actors could easily strip away safety guardrails and use them to commit cyberattacks or other harms. On the other side, leaders like Andrew Ng fear that if only a handful of massive companies control the best AI, they will become the gatekeepers of the entire digital economy, limiting innovation. As this high-level debate continues, the reality of AI data collection is also hitting home for regular users. For example, Twitch recently introduced a new setting that lets streamers opt out of having their live content, chats, and personal videos used to train Amazon’s AI models. It is a sign that while experts argue over the future of the technology, the people fueling that technology are starting to take control of their own data.