No AGI without mastery of the real world!
From IDSIA Technical Report 22-22: "The only AI that works well today is AI in the virtual world behind the screen... There is no AI-controlled robot that can do what a plumber can do, or what a capuchin monkey can do."
Task Domain Matrix (Toggle tasks to measure domain divergence)
4 Tasks ActiveVirtual World Behind-Screen Tasks (DLH Sec. 20)
Physical World & Robotic Tasks (Real World Demands)
Schmidhuber IDSIA-22-22 Core Arguments
1. Software vs Hardware Self-Improvement
"No true self-improvement without self-improving hardware, as opposed to the already existing, self-improving, meta-learning software [DLH]." Current LLMs modify weights within fixed data-center architectures.
2. The Plumber & Capuchin Monkey Threshold
"Passing the 'Turing Test' is much easier than True AI in the physical world. There is no AI robot that can do what a plumber can do, or what a capuchin monkey can do."
3. Embodied Physics & Sensorimotor Demand
Physical environments feature continuous temporal friction, non-repeatable tactile dynamics, unmodeled sensor noise, and mechanical wear that text token predictors do not encounter.
Physical Constraint Simulator Real-world Penalty Engine
Unstructured physical debris, slippery pipes, dynamic contact mechanics.
Occluded pipes, latency, sensor jitter vs crystal-clear textual tokens.
"Any modern AI would run out of context if you operated it like that for a week."