Symbolic AI and expert systems of the 1980s, including Japan's Fifth Generation Computer Project, were ultimately unsuccessful because they could not scale, leading to an 'AI winter' until neural networks became feasible through GPU acceleration in the 2000s.