전 세계를 위한 유일한 길, 오픈소스 AI

Proprietary AI is both too expensive and too centralized in control for most countries and companies to rely upon. NYC — Yann LeCun, one of the “Godfathers of AI,” may have long been Meta’s chief AI scientist, but at the United Nations Open Source Week in his keynote speech, he took a very different tack from a rah-rah big AI stance. Instead, he made an aggressive, politically charged case that open‑source AI is not just a nice‑to‑have but the only viable path to global AI sovereignty, cultural diversity, and long‑term safety. Take that, Zuckerberg! LeCun framed AI as an infrastructure‑level platform that will soon mediate “all of our interaction with the digital world with information more generally,” far beyond today’s search engines. He warned, however, that if this mediation is dominated by a few proprietary systems from “big tech companies on the West Coast of the US and China,” the result will be “very dangerous for cultural diversity, linguistic diversity… for democracy, for human rights.” His central political claim is blunt: “In my mind, the only way to get to that point is open-source AI platforms.” He argues that most countries cannot afford to build frontier‑scale models on their own, but can contribute if there is a shared open platform: “Most countries around the world cannot necessarily afford, or maybe don’t have the resources or the talents to actually build their own LLMs… There is a way that the open-source effort that we do collaboratively around the world could actually surpass the proprietary systems.” National delegates from Morocco, Sierra Leone, and Jamaica at the meeting all agreed. Open-source AI is the only way for countries in the Global South to become more than just AI consumers. It’s not just smaller countries, Alberto Gago, Director General, Spanish Agency for the Supervision of Artificial Intelligence (AESIA) , added in a later keynote that we need to “co-design a global ecosystem where we can work alongside each other, so that AI becomes a driver for progress that is transparent, equitable, and human, in which we believe digital sovereignty is the capacity of societies and not of a few techno bros to decide our technological destiny.” LeCun’s sovereignty vision is explicitly federated. “Each country, each region, each academic institution, whatever it is, would digitize its own cultural material and would contribute to training a global AI system that would constitute a kind of repository of all human knowledge, but they would not have to communicate the data. They could contribute to training a global model by exchanging parameter vectors.” LeCun positions his post‑Meta work as a concrete answer to this vision. After leaving Meta, he helped launch the AI Alliance , Advanced Machine Intelligence Labs , and “ Project Tapestry .” The last is “a confederation of partners that can contribute to training a global AI model while preserving sovereignty over data and only exchanging parameter vectors as open as possible.” The mechanics are intentionally bottom‑up and open: “The Tapestry project is very much bottom up. It’s people with expertise in training LLMs and other AI models who decide to collaborate on the GitHub repository . You can just sign up, there’s no… authorizations to get.” He argues that political support could accelerate this dramatically: “Of course, there needs to be political support for it. If governments tell their academics, their companies, and give them an incentive to participate in this project, of course, it will grow faster.” LeCun cited early participation from European countries, Switzerland, the UK, the UAE, India, Kazakhstan, Vietnam, Japan, Korea, and industry players such as IBM, NVIDIA, AMD and Intel as evidence of “a lot of interest for this project.” Tapestry is just taking its first steps. LeCun hopes that by early 2027, it will be in production. LeCun repeatedly situates open‑source AI in a longer history of open platforms displacing proprietary stacks. In the late 1990s, he recalled, launching an internet service meant buying “proprietary hardware from Sun Microsystems, Dell, and HP and then using their proprietary operating system and software on top of it.” That model “was completely wiped out in the early 2000s when people started using commodity hardware with an open-source software stack, and the same thing is going to happen” with AI, he argued. He extends the analogy to mobile networks, noting that “your cell phone regularly actually runs an open-source operating system… and it talks to a cell phone tower that runs an open-source software stack.” The market’s preference is clear, in his view: “The market wants open-source platforms because it’s cheaper, it’s more secure, it’s easier to localize if you need to preserve privacy.” Therefore, he calls the shift to open AI “inevitable.” Governments should embrace it and accelerate its development. LeCun also spent substantial time attacking what he sees as overblown security and existential‑risk arguments being used to justify restricting open models. There is, he said, “another discourse around AI… that essentially claims that AI technology is intrinsically dangerous and should be regulated, its access should be regulated, because bad people will do bad things with it.” He rejects this framing outright: “I think those dangers are very, very widely overstated. I don’t think those dangers are nearly as bad as some people have claimed.” He draws a sharp analogy: “Limiting access to AI technology because of security reasons is akin to, in the 15th century, limiting the use of the printing press, because, of course, we can’t control what information will be disseminated through printing, so I think this is akin to medieval obscurantism.” For LeCun, the bigger danger is using speculative threats to justify narrowing access and weakening sovereignty: “I think the alternative, where if you believe AI is intrinsically dangerous, and we should regulate its access, and therefore open-source AI should be banned, I think it’s extremely dangerous for democracy and human culture in general.” He also disputes specific risk narratives. On bioweapons, for example, he stated that access to information is not the bottleneck: “Having a recipe for a bioweapon is relatively easy. Building a bioweapon is incredibly complicated, particularly if you don’t want it to kill yourself. This danger is hugely overestimated.” On cybersecurity, he points out that offensive capabilities are mirrored by defensive ones: “If you have a system like this that can detect weaknesses, you can use it to solidify your own cybersecurity system.” In his view, open source is not the enemy of safety but a prerequisite for verifiable, controllable systems that reflect diverse values. The real danger, in LeCun’s telling, lies not in open‑source AI itself but in using speculative worst‑case scenarios to justify locking the technology inside a small number of corporate and geopolitical silos. LeCun links his open‑source advocacy directly to concerns about tech concentration, a theme that resonated with the meeting’s international audience. He warns that if AI systems mediating information flows are controlled by a handful of actors, “we cannot afford for all information to be funneled through systems [that] are absolutely necessarily biased. There is no such thing as an unbiased AI system.” By contrast, an open platform would enable “a wide diversity of AI assista