Rethinking AI Governance: Beyond the Model

In the fast-paced world of AI development, the focus has traditionally been on model-level governance. This approach assumes that the capabilities of an AI are primarily determined by the compute and data used during its training. But what if that's only part of the equation? The landscape is shifting, and it's time to consider additional factors driving AI advancement.

Beyond the Model

AI capabilities are increasingly being shaped by what experts term 'non-model gains'. These gains, distinct from the improvements in the base model, are becoming significant players. They're categorized into three vectors: inference gain, systems gain, and asset gain.

Inference gain refers to the scaling of compute during the AI's operation rather than just at the training stage. Systems gain involves post-training enhancements that make the system more reliable, while asset gain focuses on improving a model using limited, specialized assets. Together, these vectors suggest that our traditional risk management strategies might be falling behind.

Why Governance Must Evolve

If AI systems are evolving through these non-model avenues, shouldn't our governance strategies evolve too? The data shows that reliance solely on pre-deployment evaluation could leave us vulnerable to unforeseen risks. It's a call to action for adopting a governance approach that includes system, entity, agent, and cloud governance. The market map tells the story: AI's impact is no longer confined to the model itself.

This broader governance framework isn't just a theoretical exercise. It's important for managing the potential impacts from emerging trends like AI embodiment, continual learning, and diffusion. Ignoring these could mean we're unprepared for future challenges that AI might pose.

Societal Resilience: A Key Pillar

In addition to these governance strategies, societal resilience must be a key focus. As AI becomes more integrated into various sectors, building societal resilience will be critical to mitigating any adverse effects. Comparing revenue multiples across the cohort, it becomes evident that the competitive landscape shifted this quarter, driven in part by these non-model innovations.

Here's a pointed question: Are we ready to adapt our governance frameworks to address this broader spectrum of AI advancements? The time to act is now, as waiting could leave us playing catch-up in a rapidly advancing field.