"Never abandon a theory that explains something until you have a theory that explains more"
About this Quote
A good theory earns its keep by explaining phenomena, unifying cases, and making testable predictions. John McCarthy urges a disciplined patience: keep the model that works until you have one that works better. The point is not loyalty to old ideas for their own sake, but loyalty to explanatory power. A replacement should not be a lateral move; it should cover strictly more ground, resolve anomalies, and preserve the successes of what came before.
This is how mature science advances. Newtonian mechanics was not tossed aside because it had flaws; it remained the backbone of engineering and astronomy until Einstein offered a framework that explained everything Newton did within its domain while also accounting for Mercury’s precession, the equivalence of mass and energy, and the behavior of light and gravity at high speeds and strong fields. Plate tectonics supplanted fixed-continent ideas because it explained earthquakes, mountain formation, seafloor spreading, and fossil distributions in one coherent sweep. Progress measured by expanded explanatory reach keeps inquiry cumulative rather than chaotic.
McCarthy, a founder of artificial intelligence and creator of Lisp, brought this temperament to AI. He valued precise, general theories of reasoning and common sense. Do not discard symbolic representations just because a new statistical method dazzles on benchmarks; adopt the new method when it explains more tasks more robustly, or integrate it until the combination does. The standard is coverage, generality, and predictive success, not novelty or fashion.
Practically, the line is a safeguard against intellectual churn. It counsels skepticism toward hype, patience with imperfect but useful models, and an insistence on real gains before incurring the costs of paradigm change. The ethic is humble and ambitious at once: humble in preserving what works, ambitious in demanding that the next idea truly illuminate more of the world.
This is how mature science advances. Newtonian mechanics was not tossed aside because it had flaws; it remained the backbone of engineering and astronomy until Einstein offered a framework that explained everything Newton did within its domain while also accounting for Mercury’s precession, the equivalence of mass and energy, and the behavior of light and gravity at high speeds and strong fields. Plate tectonics supplanted fixed-continent ideas because it explained earthquakes, mountain formation, seafloor spreading, and fossil distributions in one coherent sweep. Progress measured by expanded explanatory reach keeps inquiry cumulative rather than chaotic.
McCarthy, a founder of artificial intelligence and creator of Lisp, brought this temperament to AI. He valued precise, general theories of reasoning and common sense. Do not discard symbolic representations just because a new statistical method dazzles on benchmarks; adopt the new method when it explains more tasks more robustly, or integrate it until the combination does. The standard is coverage, generality, and predictive success, not novelty or fashion.
Practically, the line is a safeguard against intellectual churn. It counsels skepticism toward hype, patience with imperfect but useful models, and an insistence on real gains before incurring the costs of paradigm change. The ethic is humble and ambitious at once: humble in preserving what works, ambitious in demanding that the next idea truly illuminate more of the world.
Quote Details
| Topic | Reason & Logic |
|---|
More Quotes by John
Add to List
