Paper reveiw: The free-energy principal: a rough guide to the brain

Review  by Mathew Richardson

The free- energy principal: a rough guide to the brain, Karl Friston (2009)

Recently I started a masters level 8 week course on the problem of pain with Prof Mick Thacker and Laura Rathbone. This course was going to challenge everything that I think I know. I like to challenge my knowledge, beliefs and translating this new knowledge into the clinic. The paper I am reviewing today is The free energy principal: a rough guide to the brain by Karl Friston (2009). My first thought was WTF is free energy becuase I had no  idea. I love how the paper starts with “the free energy principal is a simple postulate”, while I am sitting there scratching my head thinking I don’t get it, am I dumb. I think my biggest hurdle with starting to understand free energy was, I was thinking of energy in its traditional sense, an example mechanical, heat etc and I had to shift this thinking………

The first big shift for me was looking at free energy from “an information theory quantity that bounds the evidence for a model” (Friston 2009), this shifted me to more a mathematical way of looking at free energy. “ When free-energy is greater that the negative log evidence  or ‘surprise’ in sensory data, given a model of how they were generated” (Friston 2009). So, if the sensory data inputs don’t match the predictions coming down, there is a discrepancy in the data and the larger that decrepency the greater the free energy. Another term for this is prediction error.

Entropy: the average surprise of outcomes sampled from a probability distribution or density. A density with low entropy means, on average, the outcomes is relatively predicable (Friston 2009)

Surprise: or self-information is the negative log-probability of an outcome. An improbable outcome is therefore surprising (Friston 2009)

A biological organism will try to resist disorder or entropy and it can do this by minimising free-energy and decreasing surprise. It can do this by either updating its models or predictions or by changing the way it samples and interacts with its environment which is more congruent with the current predictions, all with the goal of minimising free energy and decreasing the discrepancy of sensory inputs with predictions. Thus, minimising prediction error to make sure the organism is well adapted to its environment (Julian Kiverstein, 2021).

“If we change the environment or our relationship to it, sensory input changes. Therefore, action can reduce free energy (ie prediction errors) by changing sensory input, whereas perception reduces free- energy by changing predictions” (Friston 2009)

Before reading this paper I was confused and just wasn’t getting it, then the light bulb moment of changing the way I was thinking about energy, from more traditional views to a more mathematical view.  This paper has been another piece in this complex puzzle of trying to understand and make sense of free-energy theory, predicative process and Bayesian models.

References

Karl Friston, The free energy principal: a rough guide to the brain, Elsevier, 2009

Julican Kiverstein, Le Pub Scientifique, 2021