Hands-On Reinforcement Learning for Games

Generalizing 3D vision

As previously mentioned in Chapter 11, Exploiting ML-Agents, we saw how the team at Unity is one of the leaders in training agents for 3D worlds. After all, they do have a strong vested interest in providing an AI platform that developers can just plug into and build intelligent agents. Except, the very agents that fit this broad type of application are now considered the first step to AGI because if Unity can successfully build a universal agent to play any game, it will have effectively built a first-level AGI.

The problem with defining AGI is trying to understand how broad or general an intelligence has to be as well as how we quantify the agent's understanding of that environment and possible ability to transfer ...