DeepMind and OpenAI are building AI that can play various games. Starting with chess, recently Go, and soon poker and StarCraft. We can measure AGI progress by how many games current best AI can play generally. (Not by how many games have a narrow AI play.)
Current approach seems to be working for a lot of games. It seems reasonable to expect it to continue to work for more games, but it's not obvious if it will work for all games. If it does work for all games, it's not clear if the resulting AI will be an AGI. It seems that one of these must be true:
- Using the current approach, we'll get an AGI as soon as we have an AI that can play any game.
- We won't be able to create an AI that can play any game using the current approach.
- An AI that can play any games doesn't result in an AGI.
Trading stocks. If AGI is significantly better than everyone, it's possible it can make a lot of money. It's possible it can make enoromous of money really fast if it learns how to model other automated traders.