Re: The Feeling of Life Itself

The Feeling of Life Itself by Christof Koch

Note: many of the quotes use the term consciousness, which I think that has implications beyond what is actually implied — a better term would be awareness

Koch presents a broad, generous view of organic consciousness. He delineates 5 necessary properties of consciousness 

In summary, every conscious experience has five distinct and undeniable properties: each one (1) exists for itself, (2) is structured, (3) informative, (4) integrated and (5) definite. These are the five essential hallmarks of any and all conscious experiences, from the commonplace to the exalted, from the painful to the orgiastic.” (#) added by me

and Introduces Integrated information theory (IIT), a minimal theory that captures these properties. He then uses IIT as a filter to reject things that are too simple.

“Integrated information theory (IIT) does not bang its head against this cement wall; it doesn’t try to squeeze the juice of consciousness out of the brain.3 Rather, it starts with experience and asks how matter must be organized to support the mental.”

The IIT filter is extremely minimal and accepts a large number of biological organisms. However, I think that when it comes to inorganic systems, he is overly skeptical when examining configurations that could develop some self-sustaining feedback with minimal changes, e.g., the introduction of simple feedback loops. So, when he discusses inorganic systems he makes strawman arguments about how the feedback from tens of thousands of neurons is what makes something a candidate for supporting the mental, even though the number of neurons is not part of his core argument. 

A good example of this occurs in figure 13.1 of the book. It is true that, as drawn, the circuit on the right doesn’t have the intrinsic causal powers mentioned, but it seems that adding a few connections would allow it to develop them.

I find this a bit surprising given his support of Society of Mind-like modularization of organic consciousness 

The conventional interpretation is that driving is such a routine aspect of life that your brain has wired up an unconscious, zombie circuit. However, an alternate and more unorthodox explanation is that the sensorimotor and cognitive activities are each supported by their own Whole

So again, a deep appreciation of the complexity and (potential) partitioning of the organic systems (and also super organic systems, as I’ll mention below)

I found this organic bias a bit bewildering at first, but decided that it’s probably based at least in part on his deep understanding of the complexity of the biological systems and an ab-reaction to the tendency of many in the AI/neural net community to (severely) underestimate the complexity of even a single neuron.

I’ve thought about his characterizations for a while now and decided that more specifically they come down to a desire to primarily discuss that which he definitively knows about, with gestures to his proposed methodology for evaluating things outside of his sphere.

Further into the book the basis for his prejudice against feedforward systems becomes apparent: it’s a reaction to the primarily feedforward neural nets that account for the current buzz around AI. These “neural nets” require tremendous time to train, tend to be non-adaptive, lacking the ability to incorporate feedback. They are very capable, but very brittle, and their constituent “neurons” barely constitute a sketch of the operation of biological neurons. Their inadequacy causes him to underestimate the potential for improvement. 

I’m not in any way claiming that General Purpose AI is on the horizon, I don’t think we even know how to build it. 

However, I think it explains why his bias is to search for and accept the potential for awareness in minimal biological systems, can coexist with showing his greater caution in evaluating non-biological system.

For example, he does recognize that simple correlators are important contributors to conscious systems

Brains that incorporate the associated statistical regularities (e.g., antelopes usually arrive at the drinking hole just after sunset) into their own causal structure have an edge over brains that don’t. The more we know about the world, the more we are likely to survive

(Is this consciousness? Probably not, but certainly constitutes awareness, and is isomorphic to current AI/Neural Nets)

While simultaneously being dismissive of the potential of hardware systems

The difference between the real and the simulated is their respective causal powers. That’s why it doesn’t get wet inside a computer simulating a rain storm. The software can be functionally identical to some aspect of reality, but it won’t have the same causal powers as the real thing.12

I do think this analogy is flawed. After all, if you imagine a flower in your brain it doesn’t grow inside your brain. if you put a silicon “computer brain” into a robot with the appropriate capabilities it can potentially imagine a garden, build it, and grow it. 

Moving from a purely computational system to a computer in a robot that can operate in the environment is a big lift but has been under continual development for decades by academia and companies ranging from iRobot to Boston Robotics 

I was similarly surprised at his dismissal of the possibility of awareness/consciousness at the group (supra individual) level.

Closely interacting lovers, dancers, athletes, soldiers, and so on do not give rise to a group mind, with experiences above and beyond those of the individuals making up the group. John Searle wrote: Consciousness cannot spread over the universe like a thin veneer of jam; there has to be a point where my consciousness ends and yours begins.17 Panpsychism

This is the part of the book I most strongly disagree with. Supra personal groups can certainly have an awareness and perform concerted action. Think of mob psychology or pressure to conform or Group dynamics The wisdom of crowds, Herd Mentality, etc. Dismissing this capability out of hand just seems facile.

The most interesting pieces of the book dealt with IIT, while I found the other sections overly general. I did very much appreciate Koch’s measured approach when addressing topics outside of his core field — Christof Koch’s theory of everything it certainly wasn’t