TagsRoadmap to Artificial General Intelligence - Herbert Roitblat
Abstract
Artificial general intelligence continues to be elusive in part because of a focus on a narrow range of problems that can all be described as path problems. A researcher structures the problem, the goals, the optimization methods, the parameters of the problem, and thus the range of solutions to it. A generally intelligent agent will have to be able to do all of this structuring itself, but at present we have practically no idea of how to do that. The computational tools that solve today’s machine learning problems are substantially different from those needed to structure the solutions to those problems. Computers succeeded at playing games like go and chess once it was realized that the game could be “solved” using a tree structure. Human generated heuristics guided the search of that tree, but the invention of those heuristics cannot be explained by the search of a similar tree. In contrast, human inventions depend on developing new representations, like the tree and like the heuristics. Richard Feynman won the Nobel prize in physics in part for his invention of Feynman diagrams, that made possible to solve previously unthinkable problems in physics. Artificial general intelligence will require similar skills to create its own representations and its own inventions. As the ancient Greek poet, Archilochus observed, “a fox knows many things, but a hedgehog one important thing.” Artificial intelligence researchers have been able to build very sophisticated hedgehogs, but foxes remain elusive. And foxes know how to solve insight problems.
About Data Application Lab:
https://www.dataapplab.com/about-us/