Long Friston Interview

This Karl Friston interview is long and wandering (almost 4 hours!) and a bit distorted by the somewhat affected (?) naïve “Theory of Everything” – everything including the kitchen sink – agenda of interviewer Curt Jaimungal, but anyway … some very rough mental notes:

Deflationary view – hooray. The very point. Using / understanding general intent of simplifying view (of FEP-based systems view) and not having to worry about complexity in possible detail. (See overfitting and internal<>external independence – later.)

Mathematics and computation origins – and philosophy – Andy Clark, Richard Feynman, Kolmogorov, Helmholtz, Gibbs free-energies.

Chaos and meta-stable attractors.

Levels of Sentience – good.
Sensing external world, holding a “representation” of the external world, having self-awareness of the self and the model held of the world and self, etc. (Sentience does NOT mean feeling pleasure and pain.)

Thing – recognising one generally, starting from thing <> no-thing / not-thing – Markov chains / boundaries / blankets. Circular causality implied by that dynamic definition of a thing – maintaining its steady state.

Statistical “surprisal” theory, Bayesian probability densities. Minimising free-energy or prediction error.

Influence as causation?
Sean Carroll says causation is illusion – yawn.
(This whole thesis is about explaining causation – only kind presumed is elementary in simple control-theoretic sense – in the if-this-then-than logic of the maths …) Is the axis always time? (Yawn again – time & causation fundamentally pretty weird ab-initio – so only dependencies here are to intuitive sense other than independent variable in the maths, again. Underlying Langevin equation of state flows > Path Integrals, many “equivalent” representations. (Sabine also used the “confusion” over causation to diss alternative descriptions of “free” conscious will. Fact is no-one has “definitive” model of causation beyond the maths – no scientist, no politician.)

Flows up or down / parallel to probability / concentration gradients and flows across / along iso-probability contours. Many special cases of the general. Human scale is intermediate between hot / random-fluctuation / quantum scale and cold / cosmic / stable scale.

Schrödinger ? Yes.

Warnings about Free-Energy theories – Emperor’s clothes / impenetrable difficulty – oh yes. Hidden simplicity, lost at typical human levels – quite funny.

Back to ToE question? … interesting that Friston does address the question … but falls back to basic Markov blanket point in statistical mechanics, thingness, partitioning “things”. Deflationary again. His “Deflationary Theory of Everythingness” – very good!

Algorithmic complexity. Jamesian / Bayesian belief structures. Prediction errors / Surprise minimisation / Uncertainty resolution. [Post note – Meta-surprise. The explore<>exploit distinction sometimes intentional exploration of the uncertain IS the goal, not the surprise.]

Existential “goodness” – processing new information thermodynamically efficiently. Yay! Battery powered miniaturisation – versus super-computed complexity. Quite analogous to Occam. Avoiding “overfitting” data, locally, today. (This right scale – complexity / simplicity / frequency / time-base – very much John Doyle?) (There’s no prediction in Roulette – dummy! – other behaviours / incentives too.) But yes – leaving latitude in play. Game theory involves multiple “test” moves as well as optimisation moves – eg if there really is some skewed chance in the casino’s wheel.)

Newcombs paradox. Yawn again! Question is always about genuine “population” stats or a single psychological choice. See games again.

Zipping … is a good analogy.

AGI (re-)generative models currently “a long way from these principles”.

Jordan Peterson !?!?! The “Tao” between order and chaos – again the right middle scale is the point. Sure. Self-organised criticality … Stuart Kauffman … towards the “edge of chaos”. Repertoire of dynamics needed to give you that “latitude”.

Architectural decisions in model structure, which must itself “learn” – structural learning. (Where I started mid-80’s)

McGilchrist question.  (Ross-Ashby reference in his answer.) Iso-morphism between the world and our internal model (Bernado Kastrup too – hmmm) “sufficient” iso-morphism. (No acknowledgement of McG in the answer / discussion?) Architectural iso-morphism – the gist – Yay! (It’s Solms that takes Friston into actual brain / mind processes.) And nothing about hemispheric hypothesis. Curt asks a long question about L-R asymmetries … Friston doesn’t really have a view on McGilchrist – architectural connectivity not physical geometry / shape. (Need Solms here. But McGilchrist doesn’t have a view on Friston-Solms yet either.)

Topology of flat sensors vs long-thin processors (Axons). Computation architectures. Yay! Split of whatness / thingness and whereness / relational-properties. Good point.

NOTHING so far on the advertised “What Is Life, Consciousness, the meta-Hard Problem”? (Sound getting out of sync with video)

Speed and granularity (and density / sparseness of connectivity) of processing on multiple levels. 90% of our sensorium generated by ourselves and fellow humans – so that model clearly very important to our world view. Shared narratives, language, etc but also at many levels.

Hierarchies? Yes – layers of granularity.

[Pause at 2h13m] [Continuing ...]

Shared narrative to get along? Religion?

Shared narrative to be able to communicate (but again, see games) No strong view on the religion question, beyond societal norms on beliefs implied by the shared narratives – simpler is easier to share of course (memetics – Yay!)

Deity as the simulator???

Does he believe in free-will? Yes – autonomous dynamics is emergent. (Better developed by Solms.) And Yes, our own “sensorium” – awareness of our experience. My perceptions, my predictions, my needs, my surprises.

Self-fulfilling prophecies? Sure the world we perceive is filled-in (partially constructed) my our predictions. Fulfil and realise. But false inferences and delusions too. Schizophrenia and Autism?

We’re getting well into things Friston wouldn’t claim current expertise, but being asked “what he believes” (although he has psychiatric experience).

Interviewer’s experience of a panic attack? Schizophrenia? Hearing a voice and worrying about it. Even talked to his doctor about it! – where is this going for this interview – afraid of his own mind? Sleeping drugs? Huh? Now giving psychiatric advice! Does have colleagues that have had deep psychotic “episodes”. Group talking therapies. Discord group suggested by interviewer.

[Two interviews stitched together at 2:45]

Our own existence? Actually similar to the (circular) self-fulfilling prophecy.

At last the question about “generative” model. Model whose consequences are observable by the model.

Donald Hoffman on consciousness? (Gerry Edelman previous colleague) No opinion.

Penrose-Hameroff (Orch-OR)? Entertaining, like other QM theories of consciousness – but sceptical of explanatory value. AND REMINDER Friston himself is not proposing any theory of consciousness. (Solms is.)

Sentience – sentient behaviour – is found at our intermediate scales – important. Life as self organisation to some non-equilibrium states – at neither quantum nor cosmic scales.

Templeton challenge of FEP vs IIT as far as theories of consciousness.
(I don’t see them as conflicting. Addressing different aspects.)
Starting with what it means to be alive? Moving in a way that is in the service of sampling evidence, information and resources for its own existence. Sentience and consciousness? – Reflexive vs reflective-planned responses. I / me involved in the generative model. etc Agent that plans and selects its attention. Self-awareness at higher level, etc (Elaborating the levels of sentience from earlier.) Ad infinitum – in a meta sense.

The Meta-Hard Problem? After Andy Clark. Why does the hard problem matter, why is it a puzzle that interests us? Our model must include counterfactuals – must include philosophy and metaphysics. (Chiara Marletto) The existence of the philosophical zombie as a meme, says a lot about the kind of generative we must have. The fact we can have this thought experiment suggests we cannot be zombies.

Idealist or Physicalist? Bat for either side.  (Same model as mine – what we can know is always between reality and us – my triad. There are unavoidable implicit assumptions somewhere in our metaphysics. EXCELLENT!.)
Epistemic ontology – Yay!

Unlimited Scaling of Holons / Partons? Mathematically no constraint. Practical changes in speeds/frequencies/time-constants and predictability of processes at larger and smaller scales. Markov blanket states – internal / external independencies. Proof by induction.

The Meaning of Life? Deflationary – the meaning is in the existence – the existential imperative. Know thyself? (Aspiration, motivation, “attractive states” – Maslow?) Optimism bias in selecting our attention depending on agency is real. Interviewer Curt expressing interest in these levels of motivation – we really must rehabilitate Maslow. Beyond Self-actualisation to serving the gods, etc … Achieving congruence in the stack, acting to minimise cognitive dissonance.

Embodied Cognition? The “matter” of our actions in the world – yes a symmetry between internal and external – almost interchangeable, arbitrary which is which. Desire paths – the external world responds to / learns from its inhabitants. (My favourite is the course of a river constrained by the river bank that it is itself creating.) A “dance” – game theory again. Circular causality.

Overall impressionoverfitting was the one “new” idea I picked-up in this context. Interesting to hear so much elaboration and reinforcement through questions from another agenda, but the core is already there. Curt was clearly strongly affected by it – interesting in itself.


3 thoughts on “Long Friston Interview”

  1. Karl Friston is in the news today for teaching a vat of brain cells to play pong. I thought you’d want to know.


    “Remarkably, the cultures learned how to make their world more predictable by acting upon it,” Proffesor Friston said.

    “This is remarkable because you cannot teach this kind of self-organisation; simply because — unlike a pet — these mini brains have no sense of reward and punishment.”

  2. Funny, I saw several versions of the headline, but didn’t read this story – so didn’t spot the Friston connection. Thanks.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.