Machine Learning Knows No Boundaries?
Edward P. K. Tsang · 2023
This chapter starts with a description of machine learning’s success which ignited the world’s enthusiasm for AI. It then reviews the potential of machine learning. By using machine learning, Google DeepMind’s AlphaGo beat the top human players in the boardgame Go. DeepMind went on to prove that it can learn without relying on humans’ input of game-playing strategies. This encouraged Google to develop “General AI” – to let machines learn everything from scratch without human inputs. This chapter examines where we are in this ambitious journey. It further explains the limitations of computation in general. Computer programs help us tremendously in many areas. But fundamentally, it is limited by the “combinatorial explosion problem” – the amount of computation that prevents computers from finding solutions to many problems within a reasonable amount of time. However, this limitation has been ignored by classical economics. Important assumptions in classical economics such as perfect rationality (that humans are fully rational) and homogeneity assumption (that every trader has the same knowledge and computation power) do not hold when this limitation is taken into consideration. This chapter explains the impact of computation on these assumptions.