CAN AI SYSTEMS develop MUSCLE MEMORY?
I played basketball throughout my childhood and teen years. I spent hours in the gym, practising dribbling and shooting, working on my post-up techniques and playing scrimmages with the team. At one point, I could make a layup or nail a jump shot without even thinking about the proper form - my muscles just knew what to do (don’t ask me about my left-hand lay-up though).
But it’s been over a decade since I regularly played basketball. My “balling” days ended when I started travelling and took on other hobbies. I didn’t give basketball much thought until last weekend when I found myself back on a court for the first time in a while.
I was surprised at how quickly my body snapped back into basketball mode. My legs effortlessly remembered the footwork for pivoting around defenders. My hands knew just how to grip the ball for a proper wrist shot. It was like my muscles stored a mental blueprint for executing all these basketball skills, even after years of not using them.
This got me wondering if AI could retain knowledge as our muscles do. I asked Claude2.0 if it has muscle memory and it outlined a few reasons why not:
Narrow focus - AI models learn specific singular tasks versus general intelligence. Without retraining, those narrow skills quickly deteriorate.
No true understanding - AI relies on patterns, not abstract reasoning like humans.
Data dependence - AI needs constant new data as the world changes. Human learning generalises more readily.
No common sense - Humans accumulate vast general knowledge that reinforces learning. AI does not.
However, when I looked further into it, UCLA research suggests that AI may be capable of its own mechanical muscle memory.
UCLA engineers were able to create structural material with tunable beams that can alter its shape and properties when exposed. The material learns behaviours over time, adapting to dynamic conditions like an aeroplane wing and optimising its shape during flight.*
However, while this UCLA research shows promising steps towards artificial muscle memory, a lot of AI systems still fall short of human learning capabilities. Most AI systems today rely on limited statistical associations from narrow datasets. So for AI to develop true lifelong learning, it needs broader training beyond isolated tasks. Researchers must explore going beyond correlations.
General artificial intelligence still lacks our brains' integrated knowledge networks. Muscle memory persists decades later due to consolidated neural pathways encoding complex real-world wisdom. This deep comprehension remains AI’s next frontier on the journey towards human-like learning.
Resources:
*https://www.techgoing.com/ai-materials-can-learn-themselves-and-develop-muscle-memory/
*https://samueli.ucla.edu/ucla-engineers-design-ai-material-that-learns-behaviors-and-adapts-to-changing-conditions/#:~:text=Just%20like%20a%20pianist,muscle%20memory%E2%80%9D%20of%20its%20own%2C