While I in an earlier post stated that I didn’t want to read too much psychology into the equations in the active inference framework, I can’t resist to try to sneak in some philosophy.
With a little imagination, the expression for expected free energy, the expression that according to the active inference framework guide our behavior
$$\mathcal G[q; \pi] = D_{KL}[q(o \mid \pi) \mid \mid \tilde p(o)] + \mathbb E_{q(s \mid \pi)} \mathbb H[p(o \mid s)]$$
can be seen as representing two opposing forces in Buddhism.
The first term \(D_{KL}[q(o \mid \pi) \mid \mid \tilde p(o)]\) quantifies the discrepancy between what we desire to observe represented as a probability distribution, \(\tilde p(o)\), and what we think we will actually observe if we put in the effort \(\pi\), \(q(o \mid \pi)\). It thus reflects craving (tanha) – the drive to align with desires, goals, or preferences. It’s about pursuing what you want, often tied to attachment or striving toward specific outcomes.
The second term \(\mathbb E_{q(s \mid \pi)}[\mathbb H[p(o \mid s)]]\), quantifies the average clarity and specificity of our observations if we do \(\pi\). It reflects mindfulness (sati) – the practice of staying present without judging and seeing things as they are. It encourages clarity and awareness of the current moment, rather than being overly fixated on specific goals.
This balance mirrors the Buddhist idea of walking the middle way, which is also the Swedish lagom way: neither clinging too tightly to desires nor detaching completely from the world to become a passive observer.
The minimization of expected free energy thus neatly represents the pursuit of a harmonious state, free from the extremes of overattachment and unawareness.
There is deep wisdom popping out from Jensen’s inequality.