What are examples of problems where neural networks have been used and have achieved human-level or higher performance?

Each answer can contain one or more examples. Please, provide links to research papers or reliable articles that validate your claims.


2 Answers 2


Here is an initial list of AI systems that used neural networks and have achieved human-level or superhuman performance. All of these systems are reinforcement learning systems that play videogames.

  • AlphaGo and AlphaGo Zero (an improved version of AlphaGo that does not use human knowledge but learns by playing against itself) have achieved superhuman performance in the game of go and, in the case of AlphaZero (a generalized version of AlphaGo Zero), also in the games of chess and shogi.
  • DQN has achieved human-level or superhuman performance in many Atari games
  • DeepStack has achieved human-level performance in poker
  • AlphaStar defeated a top professional player in the real-time strategy game StarCraft 2
  • OpenAI Five defeated world champions in the game of Dota
  • $\begingroup$ It's worth understanding that neither alphaStar nor OpenAI Five were able to actually visually understand the games they are playing. They are fed numeric information such as coordinates and unit type. This is significant as visual processing of the game is as important as anything else. $\endgroup$ Aug 30, 2020 at 22:14
  • $\begingroup$ @FourierFlux Thanks for noticing this. To be honest, I am only familiar with the details of DQN and some details of AlphaGo, but not the details of the other two. $\endgroup$
    – nbro
    Aug 30, 2020 at 22:19

AlphaDogfight - from Defense Advanced Research Projects Agency (DARPA) a programme that pitted computers using F-16 flight simulators against one another and later went on to defeat Air Force’s top F-16 fighter pilots.

Check out this and this news and events by DARPA.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .