Logic engines might allow the first AGIs to play reality like a game of chess

If chained-together neural networks convert sensory data into variables and pass them into a probabilistic logic engine before a neural network uses the logic engine to maximise the probability a goal is achieved by processing variable relationships and adding output variables, then the output variables are turned into motor, light and electronic outputs, as well as focus points, in theory, the neural network could use the logic engine like AlphaZero uses tree search to play chess; by creating a branching tree of the best possible moves and picking the move sequence that results in the best outcome.

For example, with 10 million events processed per second, for each event span, a neural network could use a logic engine to look 8 sub-second events ahead (5 ^ 8) for immediate actions and 10 events ahead (5 ^ 10) for minutes-long events, hours-long events, days-long events and etc., all the way up to astronomically-long events. The steps for astronomically-long events would be used as the goal for the next events down, which would be used as the goal for the next events down and so on.