Thinking about Artificial General Intelligence

Anyone who thinks about AI and its potential consciousness stumbles over the Pain Question sooner or later. In this article, I offer an interpretation of pain rooted in functionalism and Buddhism that nicely outmaneuvres the uncertainty of phenomenal consciousness.

Your brain processes vastly more information than ever reaches conscious awareness. Right now, your visual system analyzes the entire visual field, your proprioceptive system tracks every joint and muscle, your autonomic system regulates heartbeat and digestion. Millions of signals, constant processing.

We've now assembled three architectural components that seem necessary for phenomenal consciousness: attention mechanisms that monitor processing, pre-structured topologies that create distinct phenomenal qualities, and hierarchical objective functions that assign valence and generate drive. We've also touched on something remarkable: through...

We now have two pieces of the puzzle: attention mechanisms that allow monitoring of processing, and pre-structured topologies that create distinct phenomenal qualities. But something crucial is still missing. Pain isn't just information about tissue damage. It isn't just a distinct phenomenal quality processed through specialized architecture....

The attention mechanism explains how phenomenal consciousness manifests - through monitoring our own processing. But it doesn't yet explain why different experiences have such radically different phenomenal qualities. Why does red look so different from how C-sharp sounds, even though both are ultimately just neural activation patterns? Why does...

If qualia were actually without function as those philosophers suggest, we would never be able to deduce whether ChatGPT actually feels something in the background of all its processing. We however doubt this conclusion. The key lies in understanding a specific functional difference between biological and artificial neural networks - one that might...

Consider a modern GPU running a neural network trained to classify whether movie scenes are sad or funny. Millions of transistors switch states, electrical signals propagate through layers of artificial neurons, performing mathematical operations on matrices of numbers. The network successfully identifies sad scenes - detecting minor key music,...

The modern high point of the discussion around phenomenal consciousness began in the second half of the 20th century. In the 1970s Thomas Nagel sparked a new wave with his essay What is it like to be a bat?. Stating that Consciousness has an essentially subjective character that cannot be captured by objective, third-person scientific descriptions....

As long as the idea of AI exists the question if it has or can have consciousness arises. If you've tried to discuss whether AI is conscious, you've probably felt the conversation going in circles. Arguments seem to slide past each other, people talk at cross purposes, and you never quite feel like you've gotten a solid grip on the...