If you’ve read Thinking, Fast and Slow by Daniel Kahneman, you would know about the two modes of thinking: System 1 (fast, intuitive, and natural) and System 2 (deeper and more deliberate thought that’s often done when learning or working on something new).
As we approach general intelligence, it is difficult to think about where significant value will get created in the next 10-15 years if any digital task can be done by an AI and any physical task can be done by humanoid robots. We can answer this question by realizing that AI will most likely only provide us with a System 2 mode of work in the near term. This means artificial systems will allow us to complete most tasks, but they will be done in a slower and more deliberate manner due to design choices (ensuring safety) and fundamental limitations (humanoid robots are unlikely to be as efficient as a specialized robot at a task, e.g., compared to manufacturing robots in factories).
The way to accelerate progress and achieve post-scarcity faster is not to directly scale up the use of System 2 work done by AI, it’s to figure out the most efficient way to convert System 2 work into System 1, essentially scaling up the “muscle memory” of the economy. There could be novel ways to convert System 2 work into System 1 processes, such as by using AI to produce more high-quality software libraries and packages and improving the reuse of such libraries to speed up software creation instead of building everything from scratch. In the physical world, an obvious example would be to use humanoid robots to create factories that produce specialized, narrow-domain robots. An added benefit of relying on System 1 for our productivity is that it’s likely to be a more interpretable, limited, and deterministic system, which could speed up adoption as it requires less trust relative to System 2 workers, which could potentially go off the rails more often.
Build System 1 Companies
If you’ve read Thinking, Fast and Slow by Daniel Kahneman, you would know about the two modes of thinking: System 1 (fast, intuitive, and natural) and System 2 (deeper and more deliberate thought that’s often done when learning or working on something new).
As we approach general intelligence, it is difficult to think about where significant value will get created in the next 10-15 years if any digital task can be done by an AI and any physical task can be done by humanoid robots. We can answer this question by realizing that AI will most likely only provide us with a System 2 mode of work in the near term. This means artificial systems will allow us to complete most tasks, but they will be done in a slower and more deliberate manner due to design choices (ensuring safety) and fundamental limitations (humanoid robots are unlikely to be as efficient as a specialized robot at a task, e.g., compared to manufacturing robots in factories).
The way to accelerate progress and achieve post-scarcity faster is not to directly scale up the use of System 2 work done by AI, it’s to figure out the most efficient way to convert System 2 work into System 1, essentially scaling up the “muscle memory” of the economy. There could be novel ways to convert System 2 work into System 1 processes, such as by using AI to produce more high-quality software libraries and packages and improving the reuse of such libraries to speed up software creation instead of building everything from scratch. In the physical world, an obvious example would be to use humanoid robots to create factories that produce specialized, narrow-domain robots. An added benefit of relying on System 1 for our productivity is that it’s likely to be a more interpretable, limited, and deterministic system, which could speed up adoption as it requires less trust relative to System 2 workers, which could potentially go off the rails more often.