A year ago, the Biden administration launched the Artificial Intelligence Safety Institute. This initiative aims to evaluate the potential risks associated with foundational artificial intelligence models. In other words, its goal is to assess the potential impact of models trained using vast amounts of data, such as GPT, Claude, Gemini, and LLaMA.
Industry collaboration is essential for this initiative because access to these models is necessary for comprehensive study. On Thursday, the U.S. government announced the first agreement with two leading companies in the sector: OpenAI (GPT) and Anthropic (Claude). As a result of this collaboration, these companies will share their models with the government for study before launching them.
Formal collaboration. The AI Safety Institute, which reports to the National Institute of Standards and Technology, under the Department of Commerce, has announced a “formal collaboration on AI safety research, testing and evaluation with both Anthropic and OpenAI.”
What will they do? As part of this collaboration, OpenAI and Anthropic will provide access to their major new models “prior to and following their public release.” This will enable “collaborative research on how to evaluate capabilities and safety risks, as well as methods to mitigate those risks.”
In return, the AI Safety Institute “plans to provide feedback to Anthropic and OpenAI on potential safety improvements to their models.” The organization will give this feedback “in close collaboration with its partners at the UK AI Safety Institute.” In essence, the primary goal is to audit the models, identify potential risks, and find solutions, all in the pursuit of responsible AI development.
How to regulate AI. This agreement is a small but significant step forward. Regulating artificial intelligence is a challenging task due to its numerous complexities. Moreover, there’s an associated risk that comes with it: Excessive regulation could stifle innovation and disproportionately impact smaller developers.
In addition, the U.S. is home to leading AI companies like OpenAI, Anthropic, and Meta, but China is rapidly gaining ground. The Asian country has filed a significantly higher number of AI-related patents than the U.S. between 2013 and 2023: 38,000 compared to “just” 6,300.
According to Elizabeth Kelly, director of the AI Safety Institute, “These agreements are just the start, but they are an important milestone as we work to help responsibly steward the future of AI.” For his part, Jason Kwon, OpenAI’s chief strategy officer, mentioned that the company believes “the institute has a critical role to play in defining U.S. leadership in responsibly developing artificial intelligence and hope that our work together offers a framework that the rest of the world can build on.”
Self-regulation. On July 21, 2023, seven companies (Amazon, Google, OpenAI, Microsoft, Meta, Anthropic, and Inflection) committed to follow a set of eight voluntary rules regarding aspects such as transparency, testing, risk management, and investment in cybersecurity.
Although this agreement isn’t legally binding, the collaboration between OpenAI and Anthropic is part of the first commitment. This involves companies agreeing to “internal and external security testing of their AI systems before their release. This testing, which will be carried out in part by independent experts, guards against some of the most significant sources of AI risks, such as biosecurity and cybersecurity, as well as its broader societal effects.”
Image | Focal Foto (via Flickr) / iStrfry (via Unsplash)
View 0 comments