[Fis] Contingency signals: AI Information, Decision and Learning

Mon Nov 24 14:39:33 CET 2025

Dear colleagues,

I have been following this thread with interest, and I would like to offer a contribution that speaks to all of the points raised so far, as well as to the question of how the conversation is itself developing. It seems to me that several different domains are intersecting here: cellular metabolism and signalling, physiological evolution and consciousness, cosmic constraints and elemental composition, management decision-making and informational contingency, and finally the role of AI in relation to uncertainty. Rather than treating these as separate topics, I want to suggest that there is a structural pattern that runs through them, one that might help unify the discussion without forcing any reduction or consensus.

Starting with Pedro’s reminder of Tomkins, the idea that metabolic “dead ends” can evolve into regulatory signals is a powerful example of how a system reorganizes itself so that previously irrelevant differences become meaningful. A metabolic byproduct becomes a signal only when the cell has developed the organization that can register it. This logic appeared again in John’s contribution, where homeostasis shifts from a purely physiological function to an evolutionary one, and then becomes the core from which consciousness is seen to arise. In that account, gravity, elemental distribution from nucleosynthesis, and symbiogenesis are all taken as constraints that organisms internalize through homeostatic regulation. The claim is that physiology stabilizes itself in relation to cosmic and environmental structure, and in doing so generates the conditions for subjective experience.

Mark extends this into the domain of information and decision-making, where the central point is that what matters most is not answers but degrees of contingency. The more contingent a judgment appears, the more human deliberation must be allocated to it. AI, in this framing, should not be an answer mechanism but a generator of contingency signals that help organize attention and expertise. He connects this with Shannon’s H as a formal measure of contingency, and wonders whether a similar gap between internal selection processes and external pressures exists in biological systems. His question, to put it simply, is whether there is a biological analogue to informational contingency, and whether AI systems could be designed to reveal their own uncertainty in a way that meaningfully couples with human decision structures.

John had earlier posed a related question about whether AI “parts” can ever represent a “whole,” or whether they are merely space-filling outputs. This question points to a basic problem: if AI systems do not reveal how their internal operations produce their outputs, then they do not generate signals that can be interpreted reliably by human organizational structures. They perturb, but without disclosing the enabling conditions of their perturbations. This is the opposite of what biological signalling does. In cells, metabolism produces its own enabling structures over evolutionary time; in effectively designed sociotechnical systems, technologies must surface the conditions under which their outputs become meaningful for human users. Without this, their contributions become noise rather than signals.

Across biology, physiology, organization, and technology, I think a central distinction may help clarify what is happening in all these discussions. It is the difference between signals and enabling structures. A signal is a difference that a system can register as meaningful. An enabling structure is the organization that makes such registration possible. These two are often conflated, but they represent different modes of operation. Metabolic anomalies become signals only when the cell’s internal chemistry reorganizes to treat them as such. Nitric oxide or oxytocin function as signals only within organisms whose physiological architecture has evolved the necessary receptors, pathways, and homeostatic roles. Informational contingency in decision-making becomes relevant only within a social or organizational structure capable of allocating effort, attention, or expertise in response.

This distinction can help address the question of whether AI parts represent a whole. An AI output is not a signal unless the system interacting with it has an enabling structure that can interpret it. Without an interpretive architecture, the output is just external noise. AI uncertainty measures may function as contingency signals only if the organizational environment is designed to metabolize them. In this sense, whole systems are not built by assembling parts; they are built by stabilizing enabling structures that make signals intelligible. Biology does this through evolution. Institutions do it through norms, procedures, and routines. Technologies must do it through transparency mechanisms and design choices that reveal rather than obscure their own operational constraints.

Something similar can be said about Shannon’s role in this. Shannon information is not meaningful because it measures something intrinsic. It is meaningful because it quantifies the variability that a system must be prepared to process. If we think of contingency as a signal, then the relevant question becomes: what is the enabling structure that renders contingency intelligible? In cells, this is metabolism and regulatory pathways. In organisms, it is homeostatic physiology. In societies, it is communication and decision structures. In AI-supported management, it is the architecture that allocates attention and responsibility. This gives a coherent way of thinking about how contingency signals appear in biology, social systems, and artificial systems without conflating them.

Looking at the conversation itself, it seems to be evolving through a process similar to what is being described. Each contribution is not simply adding content but reorganizing the enabling structure of the discussion. Pedro’s reference to Tomkins reintroduces a foundational biological distinction. John extends the enabling structure of the conversation by introducing cosmic constraints, elemental composition, and physiological consciousness. Mark then shifts the conversation again by linking contingency signals to decision-making and AI, thereby reorganizing the interpretive domain. The thread does not seem to be heading toward agreement but toward a richer space of distinctions that allow each participant to integrate the others’ points while maintaining their own conceptual closure.

In communication theory, this is structural coupling: the conversation persists by selectively incorporating contributions that each participant can interpret within their own conceptual frameworks. No one’s perspective cancels the others; instead, they perturb one another in ways that the system of discussion can accommodate. The conversation survives because it is able to metabolize difference. If we apply the earlier distinction, the emails themselves are signals, while the implicit conceptual frameworks of each participant are the enabling structures that determine what counts as meaningful.

One way to bring this all together is to emphasize the idea of recursive enabling. Systems of all kinds seem to persist by producing and reproducing the conditions that allow their operations to recur. Cells do this through autopoietic metabolism. Organisms do it through homeostasis. Social systems do it through communication and norms. Decision-making systems do it through deliberative procedures. Conversations do it by generating distinctions that can be taken up by the next message. Technologies must do it by revealing enough of their internal operations that their outputs can become intelligible rather than opaque.

Perhaps what links the biological, physiological, organizational, and technological perspectives here is the observation that evolution, learning, and adaptation do not primarily involve adding new signals. They involve reorganizing enabling structures so that systems can integrate new perturbations without losing coherence. A metabolic anomaly becomes a hormone. A homeostatic shift becomes an evolutionary pressure. A physiological molecule becomes part of consciousness. A statistical measure becomes a management tool. An AI uncertainty estimate becomes a decision support mechanism. These transformations occur when systems expand what they can interpret.

In this light, the question raised in the thread about whether AI can ever represent a whole becomes clearer. AI systems can participate meaningfully in human processes when their operations become part of a larger enabling structure that makes their signals interpretable. Without that, they simply produce outputs that cannot be metabolized. If we think of biological contingency signals and informational contingency signals as parallel phenomena, then the relevant point is that both require internal organization capable of interpreting them. In cells, this is evolutionary. In social systems, it must be designed. The same applies to AI.

I hope these reflections help link the different strands of the conversation, not by reducing them to a single explanation, but by offering a way to see how metabolism, homeostasis, consciousness, information, contingency, and AI might be understood through a common structural logic. The thread itself seems to be demonstrating that logic in action, and it is a pleasure to observe and participate in it.

Warm regards,

Steve

Dr Steven Watson

Associate Professor Transdisciplinary Studies
Faculty of Education
University of Cambridge
Room 3S19, Donald McIntyre Building
184 Hills Road
Cambridge CB2 8PQ

Tel: 44 (0) 1223 (7)67584

https://urldefense.com/v3/__https://www.linkedin.com/in/steve-watson-8b130017/__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIKwuHKyr2$ 

Associate Editor: AI & Society<https://urldefense.com/v3/__https://link.springer.com/journal/146__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK_iRVhO0$ >

Co-Founder: Cambridge Global Knowledge Nexus<https://urldefense.com/v3/__http://www.cambridgeglobalknowledge.co.uk/__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK_NJNpOw$ > & EdgeLab<https://urldefense.com/v3/__https://cambridgeedgelab.org/__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK4qN_znI$ >

Fellow of Wolfson College
Wolfson College
Barton Road
Cambridge CB3 8BB

https://urldefense.com/v3/__https://www.educ.cam.ac.uk/people/staff/watson/__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK71tqlkk$ 

Latest writing:

https://urldefense.com/v3/__https://scholar.google.com/citations?user=6SuK584AAAAJ&hl=en__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK9V3_ik7$ 

https://urldefense.com/v3/__https://www.researchgate.net/profile/Steven-Watson-2?ev=hdr_xprf__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK-SFO4S2$ 

https://urldefense.com/v3/__https://stevewatson84.substack.com/__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK_B6gWa1$ 

https://urldefense.com/v3/__https://www.meer.com/en/authors/1707-steven-watson__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIKw5Y3_Cf$ 

Recent publications:

Watson, S. & Brezovec, E. (2025). Autopoietic Ecology: Rethinking Systems, Meaning, and Matter. https://urldefense.com/v3/__http://dx.doi.org/10.13140/RG.2.2.18462.24649__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK4sX4bwD$ 

Watson, S. (2025). Trust and the Crisis of Democratic Communication https://urldefense.com/v3/__http://dx.doi.org/10.13140/RG.2.2.19268.33923__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK0OL881j$ 

Watson, S. (2025). The integration of generative AI in education. https://urldefense.com/v3/__http://dx.doi.org/10.13140/RG.2.2.33570.75204__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK0ZXrfrZ$ 

Watson, S. (2025). Emergent discourses in generative AI in education. https://urldefense.com/v3/__https://rgdoi.net/10.13140/RG.2.2.29773.14562__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK51TY0Gu$ 
Watson, S., Brezovec, E., & Romic, J. (2025). The role of generative AI in academic and scientific authorship: An autopoietic perspective. AI & SOCIETY. https://urldefense.com/v3/__https://doi.org/10.1007/s00146-024-02174-w__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIKxv7DZt8$ 
Watson, S., & Romic, J. (2024). ChatGPT and the entangled evolution of society, education, and technology: A systems theory perspective. European Educational Research Journal. https://urldefense.com/v3/__https://doi-org.ezp.lib.cam.ac.uk/10.1177/14749041231221266__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK_t79PcV$ 

This email (together with any files transmitted with it) is intended only for the use of the individual(s) to whom it is addressed. It may contain information which is confidential and/or legally privileged. If you have received this email in error, please notify the sender by return email (or telephone) and delete the original message.
The sender has taken reasonable precautions to check for viruses but the recipient opens this message at his or her own risk.

From: Fis <fis-bounces at listas.unizar.es> on behalf of JOHN TORDAY <jtorday at ucla.edu>
Date: Sunday, 23 November 2025 at 13:19
To: Pedro C. Marijuán <pedroc.marijuan at gmail.com>
Cc: fis at listas.unizar.es <fis at listas.unizar.es>
Subject: Re: [Fis] Contingency signals: AI Information, Decision and Learning

To Pedro and FIS, I agree with Pedro that Gordon Tomkins' "Metabolic Code" correlates with what I have been saying about how and why cells evolve due to alterations in their homeostatic state in order to adapt to ever-changing conditions within the environment. Tomkins hints at homeostasis but doesn't use the term per se, even though it is the core principle for physiology (Bernard, Cannon)....

Survival would therefore have required the evolution of regulatory mechanisms that could maintain a relatively constant intracellular environment in the face of changes in external conditions.

My own work on cellular evolution and consciousness ties what Tomkins is alluding to through the effect of the force of gravity as an 'organizing principle' (Maturana and Varela's 'Autopoiesis') that accounts for how/why we feel there is something greater than ourselves, ennobling ourselves to adapt to the ever-changing conditions in our environment as a consequence of the expanding Cosmos, but allow me to explain. The aggregate of that process is what we refer to as consciousness, mediatedd by Lynn Margulis Sagan's Symbiogenesis, assimilating factors in the environment to maintain homeostasis, along with their associated mathematics (Plato, Tegmark, Livio)....integrating such factors to form our physiology, replete with math (see Weibel's 'Symmorphosis'). In turn, the composite of our physiology constitutes synchronic local consciousness, and when challenged, the latter references the Cosmos diachronically as non-local consciousness. All of the above takes on a 'holism' when seen in the guise of Stellar Nucleosynthesis (Hoyle, 1946), the stars being formed from hydrogen and helium iteratively, the elements as byproducts in their exact order of their atomic masses as the 'logic of the Cosmos'. We living beings assimilate that logic through Margulis-Sagan's Symbiogenesis, the lighter (less than or equal to the atomic mass of iron) elements within the cell, the heavier elements (greater than iron) embedded in the extracellular matrix. As 'proof of principle', homeostatic communication between the cell and its matrix is mediated by nitric oxide (NO); the highest concentration of NO in the skin resides in the Acupuncture sites, alluding to the communication between the skin and visceral organs, NO being secreted into the circulation, and ultimately the exhaled breath. The consensus is that there's not enough NO in the breath to affect the behavior of other 'conversants', as in Gordon Pask's "Conversation Theory".....that as a conversation progresses it will lead to a higher level of consciousness. However, if oxytocin, the neuroendocrine hormone, is produced as a consequence of 'progressive' conversation, the oxytocin will amplify the NO signaling within and between the conversants. This is not unlike Suzanne Simard's ("The Mother Tree") observation that the leaves of trees use ethhylene to converse with one another. Much of the above is mediated by cAMP, so we're back to Tomkin's 'Metabolic Code'.

Please feel free to comment/criticize.....John

On Fri, Nov 21, 2025 at 3:22 PM Pedro C. Marijuán <pedroc.marijuan at gmail.com<mailto:pedroc.marijuan at gmail.com>> wrote:
Hi,

the messages below remind me Tomkins' "metabolic code " hypothesis. In a nutshell it says that most (?)  signaling evolves from the detection of anomalous inner states, accumulating unwanted metabolic dead ends, and later on these very substances become signals that circulate inside and outside to trigger functional responses (the case of cAMP is highlighted). So, the individual contingent becomes later on the social determinant, and also the vice versa.
My memory is weak, my time short. Anyone interested may go to journal Science. GM Tomkins · 1975 — The Metabolic Code: Biological symbolism and the origin of intercellular communication is discussed.
I think John's works are not far from these premises...

Greetings to all,
--Pedro

El 20/11/2025 a las 12:54, JOHN TORDAY escribió:
Any way to test whether the AI 'parts' are representative of a 'whole'? or just space-filling stuff....

John Torday

On Thu, Nov 20, 2025 at 4:29 AM Mark Johnson <johnsonmwj1 at gmail.com<mailto:johnsonmwj1 at gmail.com>> wrote:
Dear all,

A group of colleagues and I have been working on diagnostic AI for some years  using a comparison technique. A few of us just published this paper on the inter-relationship between management decision making, the learning of decision-making and technology: Developing judgement for business: an AI-based model of independent management learning - ScienceDirect<https://urldefense.com/v3/__https://www.sciencedirect.com/science/article/pii/S0148296325006654__;!!D9dNQwwGXtA!WPhKwNIaSyPehSx-1rv1IXkLmAKWvfhjcSoF8ycbMGpQLKjNmUeRHbMR7LBpcGoUSKLBgkQAwLAnZMgvKms0B4w$>

The role of information in this is obviously crucial - as is the criticism of AI that it is an unreliable assistant. How might AI tell us when it's not sure about things? While Shannon's H is a kind of contingency signal, it is curious how with AI this aspect of mathematical information theory is easily forgotten! However, there does seem to be an intersection between the way information is conceived and effective social decision-making where:

a. information is framed in terms of varying degrees of contingency
b. good decision concerns the effective allocation of scarce human expertise to maximise organisational effectiveness

In essence, the greater the contingency in a judgement, the greater the need to allocate human resource to debate choices; the less the contingency, the less the need for humans in decision processes.

The fundamental message is a management cybernetic one (i.e. Stafford Beer) - it is not what the technology itself does, it is how we organise ourselves with it that matters.

The technique in the paper illustrates a way in which degrees of contingency in decision-making can be identified by the technology. It is, we would suggest, this signal which we need from our technology, not an "answer" to questions. So the pursuit of "answer engines" (as Google and others are discussing) is barking up the wrong tree.

A deeper question is whether some kind of "contingency signal" lies at the heart of biological information itself. This would have resonance with deeper cybernetic ideas about contingency (Luhmann, Leydesdorff, Shannon, etc) and cellular organisation (Torday and colleagues, Levin, etc) - whether a biological "contingency signal" is produced in the gap between the internal biological selection process of a cell (referencing its evolutionary history acquired through symbiogenesis, for example) and external selection pressure.

Any thoughts?

Best wishes,

Mark
--
Dr. Mark William Johnson
Faculty of Biology, Medicine and Health
University of Manchester

Department of Eye and Vision Science (honorary)
University of Liverpool
Phone: 07786 064505
Email: johnsonmwj1 at gmail.com<mailto:johnsonmwj1 at gmail.com>
Blog: https://urldefense.com/v3/__http://dailyimprovisation.blogspot.com__;!!D9dNQwwGXtA!QN0fp1BHdBfkCzyyfGFvr2ouR9sXbBb0zk6_F741Vgtjft8ruOu72d2uq0mdZFRgl_mJwZTo6BPIK-bGLfWL$ <https://urldefense.com/v3/__http://dailyimprovisation.blogspot.com__;!!D9dNQwwGXtA!WPhKwNIaSyPehSx-1rv1IXkLmAKWvfhjcSoF8ycbMGpQLKjNmUeRHbMR7LBpcGoUSKLBgkQAwLAnZMgv378sKxM$>
_______________________________________________
Fis mailing list
Fis at listas.unizar.es<mailto:Fis at listas.unizar.es>
http://listas.unizar.es/cgi-bin/mailman/listinfo/fis 
----------
INFORMACIÓN SOBRE PROTECCIÓN DE DATOS DE CARÁCTER PERSONAL

Ud. recibe este correo por pertenecer a una lista de correo gestionada por la Universidad de Zaragoza.
Puede encontrar toda la información sobre como tratamos sus datos en el siguiente enlace: https://sicuz.unizar.es/informacion-sobre-proteccion-de-datos-de-caracter-personal-en-listas 
Recuerde que si está suscrito a una lista voluntaria Ud. puede darse de baja desde la propia aplicación en el momento en que lo desee.
http://listas.unizar.es <http://listas.unizar.es/ >
----------

_______________________________________________
Fis mailing list
Fis at listas.unizar.es<mailto:Fis at listas.unizar.es>
http://listas.unizar.es/cgi-bin/mailman/listinfo/fis 
----------
INFORMACIÓN SOBRE PROTECCIÓN DE DATOS DE CARÁCTER PERSONAL

Ud. recibe este correo por pertenecer a una lista de correo gestionada por la Universidad de Zaragoza.
Puede encontrar toda la información sobre como tratamos sus datos en el siguiente enlace: https://sicuz.unizar.es/informacion-sobre-proteccion-de-datos-de-caracter-personal-en-listas 
Recuerde que si está suscrito a una lista voluntaria Ud. puede darse de baja desde la propia aplicación en el momento en que lo desee.
http://listas.unizar.es <http://listas.unizar.es/ >
----------

_______________________________________________
Fis mailing list
Fis at listas.unizar.es<mailto:Fis at listas.unizar.es>
http://listas.unizar.es/cgi-bin/mailman/listinfo/fis 
----------
INFORMACIÓN SOBRE PROTECCIÓN DE DATOS DE CARÁCTER PERSONAL

Ud. recibe este correo por pertenecer a una lista de correo gestionada por la Universidad de Zaragoza.
Puede encontrar toda la información sobre como tratamos sus datos en el siguiente enlace: https://sicuz.unizar.es/informacion-sobre-proteccion-de-datos-de-caracter-personal-en-listas 
Recuerde que si está suscrito a una lista voluntaria Ud. puede darse de baja desde la propia aplicación en el momento en que lo desee.
http://listas.unizar.es <http://listas.unizar.es/ >
----------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listas.unizar.es/pipermail/fis/attachments/20251124/ad0d76c0/attachment.html>