Brain-Inspired AI: Understanding and Engineering Intelligence with Dileep George

Dileep George, a prominent researcher at the intersection of neuroscience and artificial intelligence, argues that a deep theoretical understanding of the human brain's functional principles is indispensable for engineering truly intelligent systems. He posits that the brain serves as an "existence proof" for advanced intelligence, but cautions against purely simulation-based approaches like the Blue Brain Project. George critiques the idea of merely accumulating biophysical details without a guiding computational theory, drawing an analogy to attempting to build a microprocessor by only understanding transistors without the foundational knowledge of Boolean logic and gate design. He advocates for an iterative, symbiotic relationship between neuroscience and AI, where biological insights inform computational models, which in turn provide testable functional hypotheses for further neuroscientific investigation.

A central theme of the discussion is the critical role of feedback connections in the brain, which are far more prevalent than feed-forward connections. George explains that the brain actively constructs and projects an internal model of the world, constantly comparing it with sensory input to form the "best explanation" for observed phenomena, rather than passively interpreting data. This dynamic, iterative inference process, involving both feed-forward and feedback propagation, is crucial for perception and helps explain complex visual phenomena, including illusions like the Kanizsa triangle, where the brain 'hallucinates' edges based on contextual information and internal models, with neurons firing later due to feedback mechanisms.

For the development of advanced AI, George emphasizes the need to move beyond simply mimicking neural dynamics towards understanding the computational underpinnings of brain function. He proposes a hypothesis where cortical columns represent abstract concepts or variables (e.g., the presence of an edge or object), and the connections between these columns encode the relationships between these variables, effectively storing knowledge. Furthermore, the intricate internal structure of cortical columns, in conjunction with the thalamus, is hypothesized to implement the complex computations required for inference, including mechanisms like "explaining away" competing hypotheses (e.g., distinguishing between a burglar and an earthquake as causes for an alarm).

The conversation highlights the profound complexity of the brain and the nascent stage of our understanding, despite impressive detailed neuroscientific experiments. The implications for AI are significant, suggesting that current architectures may be missing crucial elements such as robust feedback loops and sophisticated inference mechanisms that enable efficient world modeling and contextual interpretation. This brain-inspired approach promises not only to advance AI capabilities but also to provide a powerful, functional framework for neuroscientists to integrate their findings into a more coherent theory of the brain, fostering a mutually beneficial relationship between the two fields.

if you want to build the brain we definitely need to understand how it works

unless you understand unless you have a theory about how the system is supposed to work how the pieces are supposed to fit together what they're going to contribute you can't you can't build it at the functional level

getting a single neurons model 99 right does not still tell you how to you know it would be the analog of getting a transistor model right and now trying to build a microprocessor

you have to you have to investigate what are the computational underpinnings pinnings of those findings how do all of them fit together from an information processing perspective

my way of understanding the brain would be to basically say look at the insights neuroscientists have found understand that from a computational angle information processing angle build models using that

our brain is building a model of the world... we are constantly projecting that model back onto the world so what we are seeing is not just a feed forward thing

what the final percept is a combination of what we project onto the world combined with what the actual sensory input is

one hypothesis is that a you can think of a cortical column as encoding a a concept

the connections between these cortical columns are basically encoding the relationship between these random variables and then the the neurons inside this cortical column and in thalamus in combination implement this actual computations needed for inference