A System samples its Environment to build, update and use Models of it.
If it does this to change the model, that’s Exploration of this environment (or epistemic action), if it does so to change the environment, it’s Exploitation of the environment (or pragmatic action).
The overall goal that determines the trade-off between the two is to
avoid surprise in the future, i.e. minimising expected free
energy[^expected-energy] or free energy of the expected futureMillidge et al. (2021)
The key [in model building] is figuring out how to identify which properties are important, which itself is a dynamic property of the system.Bar-Yam (2017), 1
No matter how we construct the model, it will be flawed, and what is more, we do not know in which way it is flawed.Cilliers (2001), 137
This means that when the environment is complex or changing, the system needs to do more exploration and less exploitation, i.e. more epistemic than pragmatic action.
This means when building individual, Explicit Models, we should Launch safe-to-fail probes and adapt models continuously.
The alternative is to Switch from individual model building to collective pattern recognition altogether and thereby Support the evolution of new implicit models.
- Bar-Yam (2017): “Why Complexity is Different”
- Cilliers (2001): “Boundaries, Hierarchies and Networks in Complex Systems”
- Friston et al. (2015): “Active inference and epistemic value”
- Millidge et al. (2021): “Whence the Expected Free Energy?”