Greg Kamradt, President of the ARC Prize Foundation, defines intelligence as the ability to learn and learning as the ability to anticipate future outcomes across domains, calling world modeling essential for AGI
Dimitris Papailiopoulos reposted the statement.
#1182Greg Kamradt@GREGKAMRADTReality is composed of many sub environments
Work, video game, basketball, relationships, love, school
They all require internal modeling of a mini “world”
Our brain is very good at making extracting multiple rules at various abstraction levels about a world works
Take a basketball game for example. Your brain can not only model objectness, but also how the ball and ground interact (gravity), the ball acts on its own (bouncing characteristics), friendly social modeling (your team), adversarial social modeling (other team). It’s able to do that all while separating that environment from other aspects of life.
Where I’m stuck now is how those physical reality’s get translated into electrical signal (via your senses) and then processed by your brain which never interacts with the world directly.
These aren’t new questions, people have been asking them for a while.
Evolution clearly gave us core priors so we aren’t starting from scratch.
The result is I feel awe via two questions:
1. What optimization or value function did evolution build us against? for everyone who says, “that’s easy, it’s…”, then why can’t we recreate it?
2. How are programs represented in our head that result in a world model?
Greg Kamradt@GregKamradtIntelligence is ability to learn Learning is ability to anticipate the future of different domains You do t have that without a functional understanding of how your environment works “World modeling” is the way forward