Assume you have 2 objects orbiting each other; you have all the information you need to predict the motion of those objects (current position, mass, distance, radius and the vector describing the motion of the objects); with this information you can perfectly predict the positions of the object into the future for any time length.

I can write a formula that will give me the precise positions of the bodies; for any time length. 10 minutes or a billion years, it doesn't matter.

Now add a third object into the mix; again we have perfect information about the initial state. In this 3 body system, we cannot know where the objects are going to be in the future there is no general formula for calculating this because it is a chaotic system.

What we can do is solve it "numerically"; basically we solve the motion of the system for very small time jumps (depends on how fast the bodies are moving), over and over to get to the future positions. If I want to know what the system will look like in a billion years, I need to solve the equation 100 billion times.

The more complex the system is the harder it is to even write equations to solve for small time jumps. If you have 100's or 1000's of interacting inputs, it is completely possible to look back in time and see what changes eventually led to the current system, but that gives you no predictive power going forward.

My issue with the model put forward by Robert Sapolsky; is that it is post-hoc a rationalization of a chaotic system. It provides no predictive power, and thus the free will part I see is that even if you knew perfectly all of the inputs you still cannot predict the output beyond providing a probability space (range of possibilities) of decisions that may be taken.