It's a 2 ours interview and the bit I got fascinated the most was about the Theory of Change that Dario Amodei explained in the podcast and how it well applies in the real world.
Theory of Change is a framework for planning and evaluating the steps necessary to achieve a desired long-term goal. It is wildly used in the social impact space. Like all the well intentioned social impact endevours, it rearely see a practical use. The framework outlines the causal pathways, identifies the required preconditions, and specifies the interventions needed to facilitate and measure successful change.
Goal: Improve access to clean drinking water in rural communities.
Theory of Change:
This is a chatGPT project example. I have very little trust that this could ever happen and be succesful, for a variety of reasons. The main reason is, people are not incentivized in the long term to follow this plan.
Anthropic's mission is to deploy safe AI worldwide. Their social impact aims to enhance safety not only for their AI systems but for all AI globally. How? They offer a powerful AI equipped with safety features, using these features as a "competitive advantage." In this manner, when Anthropic and OpenAI engage with a new client, Anthropic can promote their safe AI, compelling OpenAI to develop their own safety features to remain appealing to the client.
The interesting aspect of this Theory of Change is the use of economic incentives (such as increased sales) to implement their changes in the world.
As Dario says, this creates a race to the top of AI safety, and once their competitive advantage gap is closed by other players, they will need to develop new safety features. Smart!
A similat ToC was delivered by Tesla and their mission to move from fossil fuel to electricity and green energy. They created electric cars so good compared to the petrol cars, that will move the market towards electrivity, 2024-11-12