Recursive self-improvement’s history and the case for hardware that improves itself

Jürgen Schmidhuber shares a technical note tracing recursive self-improvement (RSI) through work he says began with his first concrete algorithms in 1987. His overview covers self-modifying policies, neural-network approaches and the self-referential Gödel Machine. He says software-based RSI has become practical, but argues that full RSI will require hardware that improves itself in the physical world, too.