Joscha Bach: Artificial Consciousness and the Nature of Reality | Lex Fridman Podcast #101

Learning Methods

The conversation begins with Lex Fridman and Joscha Bach discussing reinforcement learning and its potential to achieve higher levels of intelligence. Joscha explains that while reinforcement learning has shown promise, the current models are not yet sparse enough to fully represent potential world states 1. He suggests that building systems capable of dreaming might help overcome these limitations.

I suspect that part of the purpose of dreams is to similar to a generative adversarial network, learn certain constraints and then it produces alternative perspectives on the same set of constraints so you can recognize it under different circumstances.

--- Joscha Bach

The discussion then shifts to meta-learning, where Joscha describes it as a system that discovers algorithms to solve problems, rather than implementing solutions directly 2.

   

Philosophical Computation

Joscha explores the philosophical underpinnings of computational approaches to intelligence. He emphasizes the need for a balance between paradigmatic thinkers and those who work within established paradigms, warning against the stagnation of ideas 3. Joscha also highlights the importance of computation in understanding the world, referencing Wittgenstein's early insights into the relationship between mathematics and philosophy.

Wittgenstein already understood that, for instance, you could ExpressVPN everything with the NAND calculus, that you could reduce the entire logic to NAND gates as we do in our modern computers.

--- Joscha Bach

This computational perspective is crucial for advancing AI and understanding its limitations and potentials 4.