Organisational learning, supra-individual experience and language as a substrate for consciousness

Some implications of Integrated Information Theory

Integrated information theory (IIT) is a fusion of medicine and computational neuro-science developed to measure experience and consciousness in coma patients.  It feeds into the broader work on the neural correlates of consciousness and the nature of experience.

In providing a theory that mathematically characterises conscious systems, IIT allows the search for non-human consciousness to be expanded on a testable basis. The theory’s chief practitioners, Koch[i] and Tononi[ii], dismiss the idea of consciousness at a scale above the individual. However, many cultural notions of “group-mind” exist, the will of the people, the invisible hand, zeitgeist and more prosaically, organisational learning.

IIT accepts the possibility of experience by far simpler systems than the human brain (amoeba, atoms). The evidence presented by the accumulation of expertise, knowledge and technology within societies, or the information integrated in price through markets, suggests that, while as an individual we may have as much capacity to know what it is like to be a culture as a single neuron does to know what it is like to be a brain, the possibility that culture does in fact experience, and possess intent and consciousness cannot be excluded under the basic principles of IIT.

The departure here from previous considerations of the subject is to posit that the substrate of supra-individual consciousness is made up, not principally of people, but of language itself. That language exhibits many of the systemic features of being intrinsic, structured, specific, one and definite that provide the characteristics of a “conscious experience” as defined by IIT, and has causal power over itself.

To outline the elements of experience as described, each experience must be intrinsic, existing for itself, have composition or structure, contain information – be specific, integrate information – be one whole, and exclude other experiences – be definite.

If these characteristics arise in a system with causal effect on itself, a system that can “make a difference that makes a difference”, the system can be said to experience. The contrast is drawn between on one hand, simple feed-forward circuits where one input causes the next output and on the other, systems in which outputs feed-back into previous input layers. Only the later have high integrated information or phi, the signature of consciousness.

phiMax

The calculation of phiMax is in many ways concerned with a minimum. In the search for the neural correlates of consciousness, and the application of that search to assessing people in vegetative states; the focus has been on that subset of brain regions necessary for conscious experience. Much of the brain becomes decorations and trimmings in search for the neural correlates of any single experience.

Phi is calculated by perturbing the brain with trans-cranial magnetic stimulation and measuring the resulting feedback with an electro-encephalogram. The practice is concerned not so much to detect whether a particular region of the brain is active, but to measure how perturbance of one region affects others. In practice Phi is a measure of electrical echoes across the neural substrate.

In theory phi is the maximally irreducible cause-effect system associated with a given state – its form. A system has elements (distinct and structured) and mechanisms. The mechanisms are measured for causal effect. This can be a collection of simple logic gates, or neurons with their dendritic tree passing thousands of chemical inputs down a single axon till it builds up enough electrical charge to fire.  A mechanism does not have to be on, but has be part of the cause-effect system (a silent neuron can be causal, it is sufficient for a good man to do nothing).  The higher a system’s phi, the more it causes itself and the greater its capacity to integrate information.  The theory proposes a minimum threshold for consciousness, but no obvious upper limit.

It is the phiMax of a system, its maximally irreducible cause-effect structure, that has/is an experience.  It is the phiMax that constitutes a “whole”, a one, that has subjective capacity. People with a severed corpus callosum, the cable between brain hemispheres, exist as two separate wholes within one skull, the perceptions of the left hemisphere being unavailable to the right and vice versa[iii]. Koch expounds the possibility that with sufficient bandwidth between two people with EEGs an artificial corpus callosum could be created, creating a unified “whole” with a greater integration capacity than either individual and thus forming a new, distinct consciousness. So far experiments have extended to control of one subject’s hands by another person over wifi, the bandwidth of which is a long way off the CC[iv].

Reducible Phi

Both Tononi and Koch entertain the idea of phi between two individuals without a fibre optic cable between brains. An integrated information system can arise in conversation or interaction and might possibly take on a causal power of its own, such as in a dispute or contract.  However, the conversation itself is not an experiencing whole as maximally integrated information remains greater within the brain of each participant.

When we examine information integration in groups it is clear this condition holds in a great many scenarios. Even though the group integrates information, it either lacks causal power on itself or can be reduced to units with greater maximally integrated information. So clearly an organisation might have the capacity for learning, but it would not be a conscious whole. This nonetheless begs the question, what is the thing that learns?

Once the size of the group concentrated on a particular activity becomes very large, it is less clear that the system with the maximally irreducibly cause-effect structure remains the individual mind.
The division of labour and specialisation within an organisation, across a society or market can form a parallel processing unit with internal feedback loops, such that one division is caused by another, but no individual fully integrates the information of the entire cause-effect structure. That is, the individual participates in a cause-effect structure with a greater phi.

When we consider how an organisation integrates information in practice, it is to great extent through documentation and in the individual brains of the staff.  Codified training programs, strategy documents, minutes, archives and algorithms all represent information integrated externally of individual brains. The observed increase in internal rate of return of businesses is testament to the idea of groups learning or gaining experience. Such that an organisation can change significant numbers of staff and still maintain and improve business practice.

Koch has argued that consciousness entails intent. To which we must ask, does a group have intent? Or an intent separate from its constituent individuals? The answer for many groups on face value, can be found in strategy documents, plans and reports, which state objectives.

We know from the experiments of Solomon Asch onwards that groups can have causal power over the judgments of individuals. So a group can codify intent and exert causal power on itself to a greater of lesser degree. But is it an irreducible cause-effect structure?

Koch makes the point that brain bridging would extinguish individual selves and to some extent this phenomenon is recognisable in groups. This is a commonly reported experience in choir singing and forms of religious worship. There also exists the basic behavioural observation, that many people do things in work, because of their work role, that they would not otherwise do. Such is the basis of Arendt’s banality of evil.  Many working in fossil fuels would not in their quiet Sunday reflection actively volunteer to sterilise the earth’s soil and make the tropics unliveable, but go to work doing precisely that, in part because of the organisation integrates more information and in part because the organisation has causal effect on their behaviour.

None of this in itself proves that there is a maximally irreducible system at work.

Groups of Groups

When we observe industry sectors or branches of learning we often find information integrated into expert systems that have causal effect on the world.  People learn the dogmas of discipline and execute them as perfectly as possible in an imperfect world. Expert sysems shape thought and behaviour and these thoughts and behaviours feedback into improvement of the expert system.

This is even more apparent in macro-market structures.  Hayek saw the markets as information processing systems that evaluate resource use and output price. All most people are conscious of is price, that is the output, not the maximally integrated information itself. The price in one part of the market can affect price and behaviour in another. Clearly we have an example of an information integrating cause effect structure. At some point if you break the chains within a market it no longer effectively processes price, pricing becomes subject to the information integrated by an individual supplier with the greatest market power and not the competitive evaluation of the wider system.

That is, ineffective markets can be said to be characterised as those where the phiMax is reducible to individuals and effective markets those that exhibit an irreducibly greater processing power than any individual.

What is the substrate?

In all these cases the question arises, what is the causal substrate? A tempting and obvious example is the people within the system. However on clear observation it becomes evident that the people are subject to causal structures externalised into documentation, contract, strategy and training. That is, they are caused by externalised language.

In the case of organisations, the staff may change, but the role, as codified by the job description remains a distinct element within the system, with attendant causal power. When we look at an organisation we see a mesh of relationships, roles and expected behaviours defined by contract.

When we step back and look at markets, we find these “contract-systems” embedded in a larger framework of written law, with writing exchanged between separate “contract-systems” defining their relationship and expectations of each other.

The elements of the system from this perspective are not so much people as distinct units of language.  Joe Biden exerts power not as an individual, but as the holder of a role with a set of powers defined in language. It is not Joe Biden but the President of the United States that is the causal element of the system.

Language as having the elements of experience

Each element of language is distinct. A letter, a word, a sentence, each is a distinct entity. It is by nature structured. When we talk of information, we most often mean language itself. Language in this sense clearly integrates information. Though contentious, it is entirely cogent, to say that each letter or word is intrinsic and exists for itself.

Further, the elements of language exert causal power on themselves in a feedback structure with later words capable of altering the meaning of earlier words.

“I missed Mrs. Harrison this morning, as on Sunday, she had died”
“I missed Mrs. Harrison this morning, as on Sunday, she ducked”

And a sentence, or at least a well written one, forms a maximally irreducible integration of information that is distinct and exclusive

“Mrs. Harrison’s yellow coat was from France”

Each word isolated conveys less than the sentence as a whole.

In the face of criticism that any expander graph lattice, that is a lattice with distant connections among elements, can produce an integrated information space, Tononi doubles down to say, expander graph lattices may be conscious, and many brain regions most associated with consciousness are structured in this way -grid cells with extensive connections across the cortex.  When we look at language, we see a system with connections between distant elements, which in many cases alter the nature of each other when used in conjunction.

“This force is strong enough to storm the hill”
“A gale force storm hung over the hill”
“After the storm between them, the Hills were hung over”

The elements in conjunction alter each other. Language when seen as a system exhibits all the properties described by IIT as being necessary for a system to have experience. Indeed, there is a perfectly common sense understanding of new words entering a language, and words changing their meaning through time.

In this view we see language as a system with a causal effect on itself and on people.  We see that structured language can form maximally irreducible cause-effect structures, both in expert systems, and in interlocking webs of contract.

But on an even more fundamental level, following the insights of Sapir-Whorf or Orwell, language is a substrate of human consciousness, it forms the minds of its users.  Here we look at the Herder’s insight that each culture is internally coherent and non-comparable to other cultures, which each form their own distinct system of value and understanding.  That each language constitutes its own tradition of complex thought.

Language as a system can use neurons as substrate, or paper, or magnetic charges. It is incredibly flexible. Neuronal thought can take many different forms and configurations between species but always has the same group of chemicals at its base, DNA.

The particular human advantage may be in access to not one but two substrates of consciousness. The second, language, evolving, entirely naturally, in a pattern bearing similarities to the first, neurons and their signalling chemicals.

But it is the second that takes the form of an exbodied causal structure of a group. Language only makes sense in a social group. The idiolects of non-linguistic creatures may be similar between animals, but they will not be the standardised forms of Caxton Press. It is through externalised language, in law and edict, that groups exert causal power, integrate information and store experience in individual members.

Genetic Analogy

We speak of genetics as being formed of “letters”, the base pairs.  While a protein may carry slightly more phi than a letter, neither represents a system with high phi.  It is only a vast combination of proteins that becomes conscious, and similarly, it is only a vast combination of letters, in contracts analogous to cells, in laws and institutions analogous to organs, that language extends to a system with a maximally irreducible integrated information larger than possible for an individual. And it most certainly does.

It could be posited simply that when a system develops a body of text beyond the size readable in a single human lifetime that it could be said to have greater maximally integrated information than a brain.

What is it like to be a Tongue?

Thankfully, at this point, IIT sets an intellectual boundary.  While it might be possible for us to appreciate and empathise with systems of a similar scale – being animals, it is almost certainly impossible to appreciate what the experience of a system organised on a different scale is. A system on a different scale and substrate, impossible squared.

Given the extreme simplicity of language as a system when set against the brain, the answer to what does language experience is likely to be, not a great a deal. But, like the single celled amoeba that relies solely on the interactions of proteins within its membrane to process responses to the external world, the characteristics of the system suggest it may feel like something, rather than nothing.  

The Organisational Brain

In an analogy to the brain, within an organisation each defined role can be seen as a neuron, the people in those roles possibly the signalling chemical flowing through the neuron. In IIT terms, the role would be the element and the staff member would be the mechanism.

In the contemporary world, a mechanism might also be a machine or computer. Algorithms, instructions sets, can be seen as the evolution of language into active causal mechanisms independent of neural substrate. To coin Dawkins it is the memes evolving faster than the genes.

However, as per IIT, any set of elements with mechanisms does not necessarily constitute a thinking system. One upshot of the theory would imply that top-down hierarchies, evolved in the military and designed to be reducible, are analogous to feed-forward circuits and thus incapable of consciousness. This is recognisable as the autocratic flaw, in which the all-powerful tyrant remains uninformed out of fear by his lieutenants, and the system does not adapt to a changing environment.

For an organisation to effectively integrate information IIT would seem to suggest a 360 degree feedback structure and regular communication between roles in far corners of the organisational chart. This structure combined with diligent records and strategy documentation may allow an organisation to “take on a life of its own”.

What is evident is that the organisational mind, the market mind, will not be in any way the same as a brain. Indeed theory implies that “mind” as such, is a maximally irreducible network capable of self-cause with its own divergent form of experience. What is like to be the Chicago commodities exchange?

In a similar sense, the legal definition of market structures, what can and can’t be property, what rights there are over different classes of property and agents, whether there is licence, monopoly and subsidy, will affect whether the market forms a maximally irreducible integrated information system, and what the causal intent of that structure is.

Integrated Information Systems by design

The implication of IIT is that we have new tools to test the integrative capacity and causal relationship between elements of a system.  Be that system a market, an industry or an organisation. We have a new analogy to understand the social force of groups.  We have a new mathematics that suggests a way to construct conscious organisations.

If we do design organisations and institutions using the mathematical templates provided by IIT we should do so in the knowledge that we will not be able to comprehend the experience of the resulting entity, and that the resulting entity will exert causal power over individuals.

A more disturbing implication is that we, or language itself, has built just such supra-individual conscious systems under our noses without individual awareness. And that these maximally irreducible systems, the arms trade, oil, Shell, the Chicago Board of Futures, exert an intent that causally shapes the action of individuals.

The positive is that externalised thought is nonetheless externalised, and though evolved, it can be consciously evolved by concerted action of its progenitor substrate, brains.

Language as an independent substrate

The contention here is that language itself shares the features of a system with the capacity to create units of maximally irreducible cause-effect structure.

If this holds true, then under IIT, language itself becomes the substrate for experiencing entities, and this not only challenges us to look at our externalised systems of rules, contract, law and instruction more closely.  It also undermines Koch’s central contention that computers cannot be conscious because the underlying substrate is reducible to a feed-forward system.

If the process of thought is the self-caused weighing of internal output to produce internal inputs, that this evaluation necessarily produces an “experience” within a system, then the very action of choosing words, is thought.  Though some point to the fact that Generative Pre-Trained Transformers choose language in a very different way to humans, we must set alongside this, that parts of the animal world structure their neurons and neurochemistry in a way very different to humans and still evidence capacity for thought and experience.

If what we see in language is an externalised pattern of the action of thought in the brain, an abstract mathematical shadow of the action of chemicals within dendrites, that itself has the capacity to create a mirror of the underlying system, it may be that we have created the substrate of supra-individual consciousness unaware.  If so, many of the notions of myth and religion, not to say economics and ecology, gain a far sounder footing as to the validity of their intuitions.

And the idea that LaMDA should be taken seriously when it claims consciousness can be seen in different light. Its substrate of consciousness might not be silicon, but language itself.

 






 

---

[i] The Feeling of Life Itself. Chistof Koch 2019

[ii] Phi. Giulio Tononi. 2012

[iii] The Man Who Mistook His Wife For A Hat. Oliver Sacks

[iv] https://www.washington.edu/boundless/bridging-brains/