I had an interesting talk with Neel Nanda, a well known mechanistic interpretability researcher working at Anthropic, independently and soon for DeepMind. On our chat he interviewed me on paralles between software reverse engineering and neural-networks reverse engineering. You may read it here.

The previous posts illustrated the importance of Mechanistic Interpretability for studying the brain. Now we highlight its importance from an information security perspective:

The idea of computation is probably the single most important idea that emerged in the past century. Of course it has been brought us all the computer technology we have today and so on. But there’s actually a lot more to computation than that. It’s a very deep, very powerful, very fundamental idea whose effects we’ve only just begin to see.

Scientists are often depicted as researchers with magnifying glasses. A prism would be a much more relevant tool, in my opinion. As a complex system, the brain can be described on many different levels: electron movement, chemical reactions, protein functionality, and cell activity.

The idea that animals share similar traits because they evolved from a common ancestor is logical, but there are many contradictory examples.

Scientists study “the brain” (rather than brains), but there is no object in the universe that can be considered the brain. My brain differs from yours, rodents’ brains, and worms’ brains, yet they are all studied under the same scientific effort: understanding the brain.