I have a PhD in Computational Neuroscience from UCSD (Bachelor’s was in Biomedical Engineering with Math and Computer Science minors). Ever since junior high, I’ve been trying to figure out how to engineer artificial minds, and I’ve been coding up artificial neural networks ever since I first learned to program. Obviously, all my early designs were almost completely wrong/unworkable/poorly defined, but I think my experiences did prime my brain with inductive biases that are well suited for working on AGI.
Although I now work as a data scientist in R&D at a large medical device company, I continue to spend my free time studying the latest developments in AI/ML/DL/RL and neuroscience and trying to come up with models for how to bring it all together into systems that could actually be implemented. Unfortnately, I don’t seem to have much time to develop my ideas into publishable models, but I would love to have the opportunity to share ideas with those who do.
Of course, I’m also very interested in AI Alignment (hence the account here). My ideas on that front mostly fall into the “learn (invertible) generative models of human needs/goals and hook those up to the AI’s own reward signal” camp. I think methods of achieving alignment that depend on restricting the AI’s intelligence or behavior are about as destined to failure in the long term as Prohibition or the War on Drugs in the USA. We need a better theory of what reward signals are for in general (probably something to do with maximizing (minimizing) the attainable (dis)utility with respect to the survival needs of a system) before we can hope to model human values usefully. This could even extend to modeling the “values” of the ecological/socioeconomic/political supersystems in which humans are embedded or of the biological subsystems that are embedded within humans, both of which would be crucial for creating a better future.
If you can get access to the book, try reading The Intelligent Movement Machine. Basically, motor cortex is not so much about stimulating the contraction of certain muscles, but it’s instead encoding the end-configuration to move the body towards (e.g., motor neurons in monkey motor cortex that encode the act of bringing the hand to the mouth, not matter the starting position of the arm). How the muscles actually achieve this is then more a matter of model-based control theory than RL-trained action policy.
It’s closely related to end-effector control, where the position, orientation, force, speed, etc. of the movement of the end of a robotic appendage are the focus of optimization, as opposed to joint control, which focuses only on the raw motor outputs along the joints of the appendage that cause the movement.
You can also try diving deeper into the active inference literature if you want to build an intuition for how “predictive” circuits can actually drive motor commands. Just remember that Friston comes at this from the perspective of trying to find unifying mathematical formalisms for everything the brain does, both perception and action, which leads him to use terminology for the action side of things that is unintuitive.
Active inference is not saying that the brain “predicts” that the body will achieve a certain configuration and then the universe grants its wish. Instead, just like perception is about predicting what things out in the world are causing your senses to receive the signals that they do, action is about predicting what low-level movements of your body would cause your desired high-level behavior and then using those predictions to actually drive the low-level movements. Or rather, the motor cortex is finding the low-level movements (proprioceptive trajectories) that the agent’s intended behavior would cause and then carrying out those movements. Again, don’t get too hung up on the “prediction” nomenclature; the system does what it does regardless of what you call it.