AI Governance: Rethinking Compute Thresholds
Practical Solutions and Value
As AI systems advance, it is crucial to ensure their safe and ethical deployment. Managing risks associated with powerful AI systems is a pressing issue in AI governance. Policymakers are exploring strategies to mitigate these risks, but accurately predicting and controlling potential harms remains a challenge.
Current governance strategies often rely on defining thresholds for computational power to train AI models, assuming higher compute correlates with greater risk. However, this approach is deemed shortsighted and ineffective in mitigating risks by Cohere for AI researchers.
The proposed approach advocates for a dynamic and comprehensive evaluation of AI systems, considering multiple factors influencing AI’s risk profile. This involves better specifying FLOP as a metric, considering additional dimensions of AI performance and risk, and implementing adaptive thresholds that adjust to the evolving landscape of AI capabilities.
The research highlights that fixed compute thresholds often miss significant risks associated with smaller, highly optimized AI models. Empirical evidence suggests that many current policies need to account for the rapid advancements and optimization techniques that can make smaller models as capable and risky as larger ones.
One noteworthy result from the research is that smaller models, when optimized, can achieve performance levels comparable to much larger models. This underscores the inadequacy of compute thresholds as a standalone governance tool for AI and emphasizes the need for policies that evolve with the technology and accurately reflect the complexities of modern AI development.
If you want to evolve your company with AI, use This AI Paper from Cohere AI to redefine your way of work. Identify Automation Opportunities, Define KPIs, Select an AI Solution, and Implement Gradually. For AI KPI management advice, connect with us at hello@itinai.com. Discover how AI can redefine your sales processes and customer engagement at itinai.com.
