Friday, January 17, 2020

My 3 takeaways from Karl Friston's "active inference" framework

Seeing ‘What is real?’ is the central aspect of mindfulness and perceptual clarity is an important dimension of it. Hence, it is not surprising that I got attracted to Karl Friston’s “active inference” framework which sheds a fresh light on perceptual processing. Karl Friston, currently at University College London, is one of the world’s leading neuroscientists. This body of work – championed by Friston known with various names such as active inference, free energy principle, predictive processing, Bayesian brain hypothesis – has turned the dominant neuroscience view upside down. It proposes, to borrow Andy Clark’s words, that “We are not cognitive couch potatoes idly awaiting the next ‘input’, so much as proactive predictavores – nature’s own guessing machines forever trying to stay one step ahead by surfing the incoming waves of sensory stimulation.”1

I must admit that I haven’t looked at the mathematical formulation of “active inference” in detail yet. And I hope to do so in the future. However, there are sufficient non-mathematical articles by Friston & others available for us to get a gist of the concept. Please see the endnotes for further reading.

Here are my 3 takeaways from the “active inference” framework.

1. Perception is primarily a prediction: According to active-inference framework, the brain infers the causes of the sensory input. This predictive information flows top-down and meets the sensory data coming from bottom-up. What flows up are prediction errors. Based on the perceived reliability of the prediction errors, the prediction gets refined and/or the prediction errors get suppressed. In short, perceptual processing involves precision weighted prediction error minimization. Friston says, “Our percepts are ‘fantasies’ generated by the brain to explain our sensations. Quite literally, the brain is a ‘fantastic’ organ.”2

2. Perception-action is a unitary process: Is my action, say to lift my foot while walking, a response to the incoming sensory data? No, says Active-inference framework. It says, brain’s top level goal of walking is translated into predicting the position where my foot needs to be to touch the ground. My action of putting the foot forward is a way of testing this prediction. Thus action is just another way of minimizing prediction errors. Friston says, “Action and perception are facets of the same underlying imperative – namely, to minimize hierarchical prediction errors through selective sampling of our sensor inputs.”3 Another way Friston puts it is, “Perception is enslaved by action”.4

3. There is no “one” home: Is there an independent entity or agent or self, doing the thinking, making decisions and taking actions? No, says the active-inference framework. According to Friston, each individual is a model or a hypothesis of what might ‘work’ in its ecological niche.5 The model is unique in that ‘You can only see your own red’6 and is constantly being updated to adapt to the changing environment. However, the model is not independent of its environment. It is engaged in minimizing the surprise by increasing the accuracy of its prediction by selective sampling through appropriate action and also by reducing complexity by simplifying its model.

Then what about the sense of self I carry? According to Friston, self-consciousness is an emergent statistical property just like temperature and pressure.7 Friston calls this view – dual-aspect monism as opposed to Cartesian dualism. Consciousness is an inference process with material properties like mass, position, speed and emergent statistical properties like beliefs and self.8

Friston brings out the essence of active-inference through the following interesting example. He says, “In one sense, these ideas are also your ideas (however latent), because you have to know what you are going to see next before you can confirm it by reading these words – this is the essence of active inference  and how we sample the world to minimize surprise.”9

1.      Andy Clark, “Surfing uncertainty: Prediction, action and the embodied self”, Oxford University Press, 2016, pp. 52.
3.      Same as above (Picard and Friston, Neurology 2014).
7.      Same as above (Hobson and Friston, 2014).
8.      Same as above (Hobson and Friston, 2014).
9.      Same as above (Hobson and Friston, 2014).
Image source:

No comments:

Post a Comment