A Biological Theory of Knowledge

Foundation Literature and a Sketch

William P. Hall


(originally posted 19 October 2004)

Explanatory Note: The following essay was originally written as a contribution to the ai-philosophy forum and summarizes in an abbreviated sketch format ideas from several disparate intellectual paradigms I am struggling to assemble into a coherent whole for academic publication. I'm publishing them in this format to facilitate criticism that may help me refine the argument and to encourage interdisciplinary exploration others may wish to do in this realm.


By way of background, I'm an evolutionary biologist by training (PhD., Harvard, 1973) based on a physics background. In the late 1970's I spent two years studying epistemology. In 1981 after giving up the quest for academic tenure, I began working in various technical communication roles, which gradually involved into documentation and knowledge management systems analysis and design roles in the defense industry. For the last few years in my spare time I have partially returned to my academic roots and have been working towards a biological theory of the evolution of knowledge that explains how symbolically expressed "knowledge" derives from observations of the world. (My background for the present work is summarised on my home page)

The core issue relating to the origin of "knowledge" in this theoretical framework is at what point the material processes of systems in the world become self-referential in such a way that experiences from the world are able to be applied to control the way the system responds to the world. Pei Wang's paper, Experience-Grounded Semantics, just mentioned in the ai-philosophy thread does a good job of addressing this issue, but in my research, I have encountered a number of mostly independent threads (i.e., I have found few if any cross references between them) that I believe deeply inform the approach Pei and I taking.

Because many of the threads I am bringing together in my work relate directly to the current "physical symbol hypothesis" discussion, I'll list some of the lines of work that I think help inform an understanding of how autonomous entities can transform physical experience into forms of "knowledge" that can be used to constrain physical events. I think these threads should also be of interest to the ai-philosophy, autopoiesis and post-Popper communities. After listing the threads and giving some of the basic references, I'll very briefly sketch the pattern that emerges when they are woven together, as I think aspects of the pattern may be tested via AI and AL modeling.

Background threads†


My interest in defining life in a way that helps us understand its emergence from originally abiotic physical and chemical processes traces back to the mid 1960's when I was teaching invertebrate biology (see my unpublished paper, Is the Plastid an Endosymbiont?, written in 1966). As I began my recent work on biological knowledge in 2000-2001, I discovered the literature on "autopoiesis" (literally, self-production) by Humberto Maturana and Francisco Varela (M&V) that now forms one of several competing theories of organization and has provided a basis for judging at least some AL work:

M&V's work is very difficult to understand because of its highly idiosyncratic, theory laden, and recursive terminology, but Randal Whitaker has done an excellent job of disentangling the terms as defined in M&V's own words:

A student/colleague of M&V, Hugo Urrestarazu, is currently working on a brilliant phenomenological analysis of the nature of autopoiesis starting from first principles that I find to be a much more useful basis for building theories and models:

Cognition and knowledge

Following from M&V's thinking, cognition is basically an extension of the cybernetic processes evolved by emerging autopoietic systems. My thinking about this has been strongly based on Karl Popper's works on critical rationalism, evolutionary epistemology (the latter is familiar to some AI investigators) and his three worlds of ontological abstraction.

Conjectures and Refutations is the best work for understanding Popper's critical rationalism. Objective Knowledge is the best source for his mature thinking on the origins of knowledge in evolutionary processes and his three worlds concepts. I reference the later works here only to show that he didn't retract the ideas presented in Objective Knowledge.

My early understanding of how knowledge could be rationally abstracted from observations of the world in a Popperian framework was published in work I did in the late 1970's as summarised in Fig 1, as explained in more detail in my 1983 paper (Modes of Speciation and Evolution in the Sceloporine Iguanid Lizards. I. Epistemology of the Comparative Approach and Introduction to the Problem) that on the surface would appear to have nothing to do with AI.

Figure 1. Flow chart tracing the informational relationships of the heuristic schemata used to discover and test understandings about the causal processes involved in the evolutionary development of classes of historical phenomena.

The central epistemology of my present approach is based on Karl Popper's three ontological worlds as described in Objective Knowledge - where world 1 (W1) is material reality, world 2 (W2) is the domain of the emergent dynamic cybernetic processes and cognition of living entities embodied in material (W1) structures, and world 3 (W3) is the objective "knowledge" produced cybernetic processes that is able to be stored and shared for later recall, to constrain W2 processes and cognition in the physical world. I totally failed to understand the depth of these concepts and only understood what Popper was getting at in these ideas after coming to grips with Maturana and Varela's concepts of autopoietic cognition. Fig. 2 from my 2003 paper, Organizational Autopoiesis and Knowledge Management, illustrates the relationships of the three worlds.

Figure 2. Graphical representation of Karl Popper's (1968) three worlds of knowledge

Transforming data and information into knowledge and power

When I began working with organizational knowledge in the defense industry, I more or less automatically looked at organizations (firms) as complex living entities in their own rights. My organization wasn't doing a very good job of managing and reusing the semantic content in our documents, so I began to research paradigms I could use to explain to management the importance the knowledge we were capturing in documents to the adaptive success of our organization, and discovered Ian Coombe's hierarchy of knowledge related terms in the Australian Army's Information Management Manual, as summarised on http://web.archive.org/web/20000601203053/http://www.eclectic.au.com/speclty /im/iml.htm):

Col. John Boyd's OODA loop concept (Fig. 3) developed in a number of US military briefings provides a generic model for any complex autonomous entity working to turn experience into semantically grounded knowledge (See Boyd & Military Strategy). Basically, Boyd's loop is an iterated process of observation, orientation (processing and review of many alternatives to develop current view and understanding of the world), decision (selection of alternatives) and action (to apply the decision to the world). As I use the concept, as informed by ideas from Popper's epistemology and autopoiesis, iterated Boyd loops are the generic cybernetic process by which observations of the material world are turned into knowledge and strategic power to constrain events.

Figure 3. Col. John Boyd's OODA loop concept for winning competitions.


Last year I presented a conference paper under the title, Organisational Autopoiesis and Knowledge Management,† defining the threads from M&V, Popper, Coombe and Boyd in more detail, and showed how they can be combined to form a biologically based theory of organizational knowledge. This framework would, of course, apply to any autopoietic entity.

Since I finished the above paper, I have begun to incorporate two additional threads that substantially clarify the approach and provide a much more robust framework for the definitions and interpretations I have to make.

Hierarchy theory

What is the direct observation of the material world (W1) and what is semantically interpreted knowledge that can be used to constrain W1 dynamics via W2 cybernetic processes is determined primarily by the "observer's" level of analysis. Living things can have a deeply hierarchical structure building up from single autopoietic cells, through multicellular organisms, through social systems of organisms, or species, or colonies; to nations, and possibly even to Gaia. (In my own analysis, I make no attempt to go beyond the organizational level.)

The main proponents of hierarchy theory are Howard Pattee and Stanley Salthe:

Personally, I find Salthe's work to be too "post modern" to be useful to scientific endeavor, and I have not yet had time to lay my hands on Pattee's early work - although, as discussed below, I find Pattee's later work on the epistemology of complex systems to be profoundly astute.

On the other hand, Steve Gould's work on hierarchical selection and the existence of "evolutionary entities" at different hierarchical levels has proven to be quite useful. This provides a good definition as to what kinds of entities can be considered to be living in terms of their capacity to evolve through natural selection:

Epistemology of complex, hierarchical systems

Only in the last few weeks have I discovered Howard Pattee's recent works on the epistemology of evolving and learning systems. Pattee's idea of the "epistemic cut", distinguishing the observable activities of the material world from a living system's "observations" that can be fed back via activities of the living system to influence the material world, provided an insight that is helping me to clarify most of the remaining muddy aspects in my biological theory of knowledge. His idea of "semantic closure" in evolving and learning systems further clarifies what is happening in autopoietic systems. As well as speaking to biologists, Pattee is directly concerned with the issues of defining artificial life and artificial intelligence.

Of Pattee's papers I have read so far, those that most clearly define the concept and importance of the epistemic cut and semantic closure are:

A theory of emergent or open-ended evolution that is consistent with the epistemological foundations of physical theory and the logic of self-reference requires complementary descriptions of the material and symbolic aspects of events. The matter-symbol complementarity is explained in terms of the logic of self-replication, and physical distinction of laws and initial conditions. Physical laws and natural selection are complementary models of events. Physical laws describe those invariant events over which organisms have no control. Evolution by natural selection is a theory of how organisms increase their control over events. A necessary semantic closure relation is defined relating the material and symbolic aspects of organisms capable of open-ended evolution.


In my opinion, the measurement process in physics is the most convincing and fundamental example of the necessity of complementary models with semantic closure. On the one hand, it is possible to describe a measuring device in its material detail, and this may be necessary in its design and construction. On the other hand, if the measuring device is to perform its function (i.e., produce a symbolic record) these details must be selectively ignored. This is not a metaphysical position but arises from the pragmatic fact that to obtain a meaningful result we must be able to measure something without having to measure everything. This means that to function, the number of material degrees of freedom in the measuring device must be reduced to the few semantically relevant symbolic degrees of freedom of the result. Without such a classification process we have a divergent infinite regress of measurements, as von Neumann (1955) pointed out.


As in the case of measurement, in order to have any useful function, genes must be able to symbolize something without symbolizing everything. Otherwise genetic instructions would never end. Without simplification, heritable symbols would suffer the same infinite regress as measurement symbols. Therefore to allow open-ended increase of material complexity while maintaining heritability requires simplification of description....

Pattee's other works on the web and other recent works relating to the issue are:


Last, but not least, there are a number of papers in the discipline of "biosemiotics" that directly address the question of how symbolically expressed knowledge addresses the physical world:

Sketch: a biological theory of knowledge

Based on the threads summarised above, a biologically based theory of semiotically encoded knowledge emerges from the material world that can be defined phenomenologically.

Morowitz (1968, 2002) and Chaisson (2001) argue that environmental fluxes of energy between temporally or physically sources of high potential sources sinks low potential maintained by geophysical and astrophysical processes force the development of cyclical and dissipative processes to transport the fluxes from their sources to sinks. Sources and sinks may be such things as chemical and temperature gradients around deep sea thermal vents or the sun as a source of incoming solar radiation and deep space as a sink for reradiated infrared (heat) radiation.

Such dissipative cyclical systems are entropically driven away from thermodynamic equilibrium by the dynamic actions of transporting the fluxes of energy and material from high potential sources to low potential sinks.

These cyclical systems can be said to be "living" when they are identifiably bounded, complex, and self-regulated, with the capacity to self-maintain and self-produce their structure in a dynamic steady state in the face of internal or external perturbations.

In order to maintain their continuing autonomy or "life", such dissipative systems must maintain access to environmental or internally stored sources of potential differences sufficient to drive, regulate and maintain their "metabolic" structure in the face of entropic tendencies to dis-integrate. Where access to sources of energy/material required to fuel such maintenance is lost, continued dissipation will eventually lead to dis-integration of the system as its equations of state evolve towards entropic equilibrium.

Self-regulation and self-maintenance in the face of perturbations (noise) and entropic tendencies to dis-integrate implies:

  1. the existence of negative feedback mechanisms able to react to perturbations by channelling entropically driven fluxes of material and energy able to counteract or minimise the propagation through the bounded system of disruptive effects of the perturbation, and
  2. that such negative feedback mechanisms are also themselves able survive the effects of the perturbation.

Such cybernetic self-regulatory processes represent the origins of cognition.

Persistent structural aspects of the cognitive system that preserve and maintain the capacity to apply negative feedback to damp out the effects of otherwise destructive perturbations represent the origins of memory, and the origins of Pattee's epistemic cut. Such autopoietic systems are based on biophysical processes at the molecular and macromolecular level of organization.

Once capabilities to reproduce and multiply persistent regulatory structures evolved (through the selective elimination of those structures unable to persist), the capacity for memory could be propagated in the form of "knowledge" - in this case, information that has selectively survived testing against the material environment.

Eventually, in the evolution of cellular life, molecularly based systems of inheritance and development based on symbolically encoding experiential knowledge in the form of nucleotide sequences in the DNA molecule were established and provided the basis for further evolution and differentiation. (Note for future reference, that via processes of transformation, viral transduction, and sexual conjugation (see Hurlbert 1999 - Chapter 9), symbolic information encoded in selected DNA can be horizontally exchanged between cells, as well as propagated vertically from parents to offspring).

Thus, following from Pattee's ideas, the concept of symbolic memory only makes sense in a living system where there is an epistemic cut between the physical and material processes governed by immediate circumstances of a particular kind of perturbation (i.e., law), and selectively retained and persistent (i.e., remembered) semiotic controls able to observe (i.e., measure) and react to observations (i.e., decide) of a perturbation at some other time to constrain the then immediate physical and material process (i.e., to act) to maintain the dynamic steady state in the face of the perturbation.

Note that the symbol is persistently encoded in the material structure of the cybernetic system (i.e., is embodied in W1), but is an abstract representation (in W3) of a surviving, tested (i.e., selected) solution to a real-world problem, that is cybernetically enacted (i.e., in W2) at some future time as a tentative solution to a similar type of problem.

To understand and describe the origins and evolution of higher orders of autopoiesis and kinds of "knowledge" relevant to each order, one has to take care to situate our point of view as an observer and establish an epistemic cut between the cognitive phenomena being described and the properties and processes in the material world the autopoietic cognition of that order is considered to control.

For example, M&V and Urrestarazu have argued that associations of cells cannot be considered to be autopoietic in their own right, because they are hierarchically removed from the material processes of biochemistry that account for cellular autopoiesis. In other words, M&V and Urrestarazu have left their epistemic cuts between the cell (where symbolic controls are enacted) and the biophysical material world of macromolecules. I would argue that the autopoietic nature of associations of cells (i.e., multicellular organisms) in relationship to autopoietic single cells can be properly considered only if the epistemic cut is established between the hierarchical level of single cells and the multicellular organism. Here the W1 material processes contributing to the autopoiesis involve physical contacts, the exchange of particular chemicals and macromolecules between cells in contact, and the establishment of molecular or other chemical gradients across the system of cells. Multicellular cognition (W2) includes the propagation of developmental and regulatory signals between cells to regulate processes within individual cells. In the early evolution of multicellular organisms the hereditary knowledge for these signaling processes is still almost entirely encoded in DNA (W3) not relating to cellular autopoiesis, but as encoded informatoin that makes symbolic sense (i.e., as knowledge) only in the environment of the integration, development and behavior cells working together to form the multicellular organism. In higher organisms, with the development of a specialised central nervous system, it became possible to encode (w3) knowledge in the disposition of specialised nerve cells; and with the further evolution of consciousness it became possible to exchange and criticise symbolic information between autopoietic individuals, via pheromones, visual or tactile observation and imitation and eventually via spoken language and writing.

Social organizations can analysed to see if they are autopoietic. However, for this analysis to make sense, the epistemic cut must be made between the higher organizational level of the autonomous social entity and the lower level of multicellular organisms. In the case of social entities the autopoietic cognitive processes are enacted in W1 via gestural communication, language, writing, and economic exchanges (e.g., buying, selling, salaries and wages). Lower level components (e.g., individuals - for parts or segments of their lives) are made part of the organization by processes such as induction training, employment agreements, team building etc., and identifiable boundaries are established by tagging individual members (e.g., ID badges, uniforms, etc.). Organizational memory and learning is encoded symbolically in W3 in the form of routines, processes, documents, and relevant memories of the individuals comprising the organization. Knowledge is assembled from W1 via W2 processes of data collection, aggregation, circulation, criticism (e.g., experimentation and peer review) and decision (e.g., consensus building, voting, management, etc.)

To conclude

Obviously, in the thinking summarised here I have informed, modified and extended concepts derived from my primary sources by combining them with other disciplines and paradigms the primary authors had individually not considered and my own still unpublished thinking from nearly 40 years of thinking about biological and organizational issues. As I can steal time from my more than full-time employment, I am now working towards putting the ideas into print in peer-reviewed journals and at least one book. However, I am happy to share the ideas in any form possible.

I needed to write the above sketch to test my own understanding. Although the thinking presented in it is highly compressed, I believe that I have given enough substance to the threads to provide a framework for modeling some of the ideas in computer environments; and I will also welcome any critical comment from autopoesis and Popperian communities - although I may respond slowly or not at all to the critiques as I concentrate on the main writing exercises.

Other References

Chaisson, E.J. 2001. Cosmic Evolution: The Rise of Complexity in Nature. Harvard University Press. - http://www.tufts.edu/as/wright_center/cosmic_evolution/

Hurlbert, R.E. 1999. Fundamentals of Microbiology 101. Washington State University - http://www.slic2.wsu.edu:82/hurlbert/micro101/pages/101hmpg.html

Morowitz, H.J. 1968. Energy Flow in Biology: Biological Organization as a Problem in Thermal Physics. Academic Press, New York. 179 pp;

Morowitz, H.J. 2002. The Emergence of Everything: How the World Became Complex. New York: Oxford University Press