First of all, it's a talking point of the industry that there's some kind of tension between having safety or reliability or trustworthiness in our AI systems and having them roll out well into the economy and lead to high productivity. I think these are actually the same thing. The primary barrier to better adoption of AI in the economy and boosts of productivity is that people simply don't trust them. They don't feel they can rely on the AI systems. They don't understand how they're operating.
This is a product of how these systems have been developed. They create one system to try to do everything and to be a full human replacement rather than build specific tools for specific purposes. I believe if we take a different tack of building purpose-driven AI tools for things like scientific research, for things like mathematics, or even for things like helping you keep your calendar, and they're actually tested for being able to do those things well and reliably and in a trustworthy and safe way, that will be a huge economic boost relative to the current direction in which things are going, which is more focused on the wholesale replacement of human labour.
The crucial thing is that there is an external evaluation system that can test, by an independent agency or authority, that an AI system is safe and reliable before it goes to market, just as Max was discussing.