I think I theorize like - a big assumption is AI either leads to fast take off or it moderately helps us but hits a ceiling.
I wonder if a third option is - AI actually slows down progress because -
a thing I observe at work is the hard problems - making hard decisions, influencing and generating consensus. Innovating new ideas. These problems are still hard.
And AI lets you quickly sort of implement a tool or add a feature or do things that are incrementally useful, but lower the cost of doing those things in a way that make solving the hard problems disproportionately _more_ expensive.
So scale that up to society.
In this example, what if AI solves a whole bunch of mathematics at this side of the frontier, but destroys the community and institutions that enable pushing the frontier.
What if it automates AI research in a way that accelerates the asymptote?
I think I theorize like - a big assumption is AI either leads to fast take off or it moderately helps us but hits a ceiling.
I wonder if a third option is - AI actually slows down progress because -
a thing I observe at work is the hard problems - making hard decisions, influencing and generating consensus. Innovating new ideas. These problems are still hard.
And AI lets you quickly sort of implement a tool or add a feature or do things that are incrementally useful, but lower the cost of doing those things in a way that make solving the hard problems disproportionately _more_ expensive.
So scale that up to society.
In this example, what if AI solves a whole bunch of mathematics at this side of the frontier, but destroys the community and institutions that enable pushing the frontier.
What if it automates AI research in a way that accelerates the asymptote?