Many AI Safety Orgs Have Tried to Criminalize Currently-Existing Open-
The article argues against the common misconception that AI safety organizations do not oppose open-source AI, clarifying that several well-funded groups have actively proposed regulations designed to ban or severely restrict models currently available, such as Llama 2. By setting technical thresholds based on computational power and performance benchmarks, these proposals would effectively criminalize the public release of open weights for existing state-of-the-art models. This effort aims to cap open-source capabilities below those of proprietary systems, thereby preventing the community from sharing or building upon current advancements. These regulatory pushers, including entities like the Center for AI Safety and The Future Society, advocate for strict oversight mechanisms that are incompatible with the decentralized nature of open source, such as mandatory liability and continuous monitoring. The author contends that this constitutes a deliberate strategy to stifle open development and consolidate market power among a few large corporations under the guise of safety. Despite the harmful implications for innovation and security research, these groups remain influential in policy circles, leveraging opaque communication strategies to advance agendas that lack broad public support. This dynamic is critically relevant to the open-data community because it frames open-source AI as a primary target for restrictive legislation rather than a tool for safety. The article suggests that the open-source movement is currently unprepared to counter organized efforts that seek to treat open access as a security risk rather than a benefit. Consequently, the future of open data in AI depends on the community’s ability to unite and advocate against policies that would effectively privatize intelligence capabilities and halt the collaborative research essential for both safety and progress.
Source: 1a3orn.comPublished on 2024-01-17