A recent paper titled "Computing Power Then Artificial Intelligence Governance," authored by researchers from OpenAI, Cambridge, Oxford, and several other academic institutions, presents a striking conclusion regarding the management of artificial intelligence (AI). The researchers suggest that the most effective strategy to combat the potential misuse of AI lies in the continual advancement of more potent AI technology, under the auspices of governmental oversight.
The paper underscores the pivotal role of computing power in both the development and regulation of AI systems. According to the researchers, controlling access to the hardware necessary for training and operating advanced AI models is paramount in determining who wields the most powerful AI tools in the future. They argue that policymakers can leverage computing resources to bolster regulatory oversight, allocate resources for beneficial outcomes, and impose constraints on irresponsible or malicious AI endeavors.
In essence, the researchers advocate for preemptive measures aimed at curbing malevolent AI applications by exerting control over the hardware infrastructure pivotal to AI development. This necessitates the establishment of systems for monitoring the production, sale, and utilization of hardware essential for advancing AI capabilities. Moreover, the researchers propose the integration of "kill switches" into hardware components, empowering governments to enforce regulatory measures remotely, such as shutting down illicit AI training facilities.
However, the paper acknowledges the potential pitfalls of implementing naive or poorly conceived computational governance approaches. Concerns encompassing privacy infringements, economic ramifications, and the consolidation of power underscore the complexity of navigating AI regulation. Notably, efforts to monitor hardware usage may clash with established guidelines like the "Artificial Intelligence Bill of Rights Blueprint," which underscores citizens' rights to data protection.
Furthermore, the advent of communication-efficient training techniques could facilitate decentralized computing practices, complicating governmental oversight endeavors. The decentralization of AI development processes might render it increasingly challenging for authorities to identify, monitor, and intervene in illicit AI-related activities, such as unauthorized model training. Consequently, the researchers foresee a potential escalation toward an arms race dynamic as governments strive to counter the illicit use of AI technologies. They emphasize the imperative for society to judiciously leverage advanced computing capabilities to mitigate emerging risks associated with unchecked AI proliferation.
