Experiments like ARC–AGI have finally made true computer intelligence into a scientifically respectable theory, because AI can now be falsified.
Read More ›
DeepMind, a part of Alphabet (i.e., Google), has made many headlines in the past. The biggest was its development of AlphaGo, which used reinforcement learning to beat the number one Go player at the time (2017). DeepMind generalized this into AlphaZero, which is supposedly able to solve any two-player game of perfect information. DeepMind has come back into headlines recently with the attempt to build an AI which can generate any algorithm. While they are starting with map data, the goal is to generalize this and generate any desired algorithm. The search for such a “universal algorithm” has been essentially equivalent to the search for a perpetual motion machine in physics. The allure of both is obvious. In physics, if you Read More ›