Published Oct 20, 2024

Joscha Bach - Why Your Thoughts Aren't Yours.

Joscha Bach delves into the emergent nature of consciousness and agency through a computational lens, examining AI's societal impact and the transformative potential of technologies like liquid neural networks, while advocating for balanced AI regulation that encourages innovation and avoids stifling progress.
Episode Highlights
Machine Learning Street Talk (MLST) logo

Popular Clips

Questions from this episode

Episode Highlights

  • Self-Organization

    explores the concept of self-organizing systems in AI, drawing parallels to human cognition and consciousness. He suggests that consciousness arises from the organization of information, where self-models and world models interact within a virtual environment. This interaction is akin to a game engine, where the self perceives itself as a character solving problems within a constructed reality 1. Bach emphasizes that consciousness and intelligence exploit complex entropy gradients, allowing organisms to navigate and benefit from their environments 2.

    The reason why we are conscious seems to be that we care at some level.

    ---

    This caring translates into an impulse for neurons to model reality, thus facilitating survival and adaptation.

       

    Software & Agency

    Bach delves into second-order software systems, which are self-designing and grow into complex structures, much like a seedling becoming a tree. This concept is crucial for understanding AI's potential to develop coherent, self-organizing systems that behave as if guided by a single goal or function 3. He also discusses agency in AI, noting that while current AI systems like LLMs simulate agency, they lack intrinsic agentic properties. These systems are deterministic, focusing on predicting outcomes rather than possessing true autonomy 4.

    The simulation of a software agent is going to be a software agent.

    ---

    This highlights the distinction between simulated agency and genuine agency in AI.

       

    AI Agency

    The concept of agency in AI is further examined, with Bach arguing that agency is not a physical property but a model for understanding complex systems. He compares it to money, which, while not physical, has causal power and influences reality 5. Bach suggests that AI systems can exhibit agency-like behavior through simulations, but this does not equate to true agency. These systems are designed to predict and simulate outcomes rather than act independently 4.

    Agency is not a physical property, but what is physical?

    ---

    This perspective challenges traditional notions of machine intelligence and highlights the complexity of defining agency in AI.

Related Episodes