Short works

Books : reviews

Irun R. Cohen.
Tending Adam's Garden: evolving the cognitive immune self.
Academic Press. 2000

rating : 2 : great stuff
review : 7 February 2002

This is a book of two halves, each of them excellent. Cohen is a professional immunologist, and his reason for writing is to explain his theory of the immune system as a complex self-organising cognitive system. But that's only the second half, because before he does so, he takes the first half to explain complexity, evolution and cognition. And that first half background is worth the entrance price alone.

The immune system works by recognising and responding to certain molecules and cells. Cohen's theory is that it has properties of a cognitive system, so he needs to define recognition, reaction, meaning, information, entropy, emergence, and so on. He does this all very nicely. (And he has a rather nice way of describing evolution in terms of attractors in a dynamical biological system, too.) His definition of a cognitive system is one that has three properties:

[I would quibble with his usage of the term self-organisation here, but not with the concept it is being used to label. I would prefer to call this property learning, which fits with the internal images being kinds of memories.] The important point to note in this definition is that it makes no mention of anything like consciousness, or even what we classically term as thinking. Thus is can be applied to non-brain-like activities, in particular, the immune system.

Having established a solid background in the first half of the book, Cohen goes on to describe the amazingly complicated immune system, and how it can be viewed as a cognitive system. He pays particular attention to the seemingly paradoxical properties of degeneracy (a receptor may bind to many different ligands), pleiotropia (an agent may do different things in different contexts) and redundancy (several different agents may perform the same action). He shows how these non-specific properties can lead to highly specific recognition and reaction, and why the resulting system has no simple one-to-one relationship between cause and effect.

There is a very strong sense of process and dynamics in this description. That life and other complex systems are not static things but dynamic processes has become the prevalent viewpoint over the last few decades, and helps to remove some of the paradoxes that the static view threw up. Indeed, Cohen's description is so process-oriented that he uses a form of Harel State Charts to depict some of the processes. [I would quibble slightly here too, because I think some of the "states" in these diagrams are not actually states, but things going through a series of states. However, the diagrams show clearly what is meant, and are much more useful than pictures of static things, or simplistic "lifecycle" pictures. Better use of State Charts for these kind of models can be found in Kam, Cohen, & Harel, The Immune System as a Reactive System: Modeling T Cell Activation with Statecharts.]

After describing his new cognitive theory of immunity, Cohen contrasts is with the 50-year-old "standard model" of Clonal Selection Theory (CST). He shows that this classical theory of immunology is correct insofar as it goes, but that it does not go far enough, and misses much of the picture. CST paints a picture of an immune system that is dormant when the organism is well, which awakens when infection strikes, deals with it, then switches off again. In reality, the immune system is constantly active. In particular, CST ignores the entire maintenance function of the immune system. The immune system does not just detect and kill intruders; it aids in repair and growth, too.

The last part of the book covers autoimmunity. Physiological autoimmunity is the recognition of, and reaction to, "self". When that reaction is inappropriate, there may be an autoimmunity disease. CST predicts that normal physiological autoimmunity is impossible, which is not the case. Physiological autoimmunity is actually an important property of the immune system. Amongst other things, immune system cells recognise the state of other immune system cells, so can tell when they are responding to something -- and they recognise maintenance molecules, in order to perform maintenance tasks. The cognitive theory readily accommodates physiological autoimmunity, and also explains the patterns of autoimmune disease much better than does CST.

In closing, Cohen notes that our immune system evolved in a world where we associated in sparse groups each of about 10 to 100 people, with no sanitation and no old people, very unlike the sterile but populous diverse world we live in today. The immune system needs to be understood much better, for it needs all the help it can get, operating so far from its evolutionary context. [This closing piece is wrapped up in some biblical metaphor, which did nothing for me. But it is just a metaphor, and the point being made stands alone.]

This is an excellent book, clearly written, with an interesting and well-argued thesis. Even if you are not interested in immunology, read the first half for the insights into meaning, evolution and cognition. Then read the second half anyway, and become interested!


I read this book mostly on the train while commuting. It is so packed with good ideas that I started carrying a little notebook with me, to jot down references to the good lines. I reproduce below the ones that caught my eye. These capture some of the breadth and depth of the material Cohen covers. (The cross referencing to other sections is useful, too. Rather than repeating a definition several chapters later, or leaving you hunting through your fading memory, a simple §nn in the text leads you to the original definition.)

On causality, evolution and cognition

[§8] information that causes some effect is information that bears meaning. Meaning is the impact of information. Meaning, in contrast to information, is extrinsic. Meaning is what the information does.

[§41] there are more E. coli organisms in any one of us than there have been of humans in all the world for all of time

[§41] evolution is the creation and occupation of attractors ...
The very existence of attractors ... makes new informational space (new room) for more complex attractors to emerge. ... the emergence of cells made room for the emergence of multicellular organisms.

[§59] Entities interact most naturally when they occupy similar scales.

[§60] Physiological systems ... are coded images of the needs they have evolved to satisfy. ...
the inside of the glove is a negative image of the hand it fits ... The glove is also a more abstract image of the cold weather

... that is, evolved systems are coded images of the needs they have evolved to satisfy; designed systems are coded images of the needs they have been designed to satisfy. (Note the essential change from active to passive voice.)

[§61] cognitive machines ... [construct] internal images that map the environment.

[§69] Internal images, like other physiological properties of systems, can fail to develop, or may function improperly

[§73] The second condition for self-organization ... is that there be extra, or redundant, copies of the old information. ... if the old information were not redundant, it would be destroyed in the process of transforming it into new information.

... hence we are talking of "embodied", or physical, information, not abstract information.

[§73] If the protein does something, participates in an attractor, then the information bears meaning. If there is no functional protein (no attractor), there is no meaning.

... so, information has "meaning" if the information receiver/bearer does something (participates in an attractor) based on the meaning.

[§76] experience can only be experience of something. ... Unless you have a notion of what you are looking for, you won't know it when you see it. Reality in itself is undivided; to observe, the observer needs pre-formed categories that divide reality into usable portions.

On immunity and autoimmunity

[§96] Immune maintenance of the functioning body, like the embryological building of the body, involves cell death as well as cell growth, movement, and support functions

[§99] recognition requires ... discrimination and response. ... Recognition ... requires both information and meaning.

[§101] a protein can respond to a ligand by changing its conformation when it binds the ligand. ... The reaction site can have one distinct shape in the absence of a ligand and another distinct shape when the combining site has bound (has sensed) the ligand. This shift in conformation of the reaction site constituted the physical response of the receptor ... Receptor-ligand interactions can be viewed as attractors.

[§102] The protein chain encoded by a single DNA sequence can have more than one stable conformation (and hence more than one function), and transitions between conformations are determined by ligands and other environmental factors. By itself, the DNA sequence does not suffice to determine the shape and function of its protein; [they] emerge in response to interactions wit the epigenetic environment.

[§103] When asked to computer the conformation of a protein based on its sequence, the computer flounders; there are simply too many computations that need to be done. ... It is a wonder how the protein, within seconds, folds itself into stable states of potential energy. Obviously proteins ... do not have to compute their steps one by one as they fold. But then how do they do it?

[§104] Protein synthesis is inherently hazardous ... because the contents of the cell could become exposed to incomplete protein chains. [They] lack the amino acids they need for their mature conformations and so might assume improper shapes and, even worse, poison the cell

[§105] A sufficiently flexible ligand ... can mold itself to fit different receptors

The "lock and key" metaphor is too binary:

[§105] affinity is a matter of degree.

alternative ligand can bind to the same receptor, so recognition is not just binding, it must also include response:

[§105] specificity is not a given

fidelity is a relative concept:

[§107] Any receptor will bind to more than one ligand, ... any ligand will be able to interact with more than one receptor.

[§108] pharmaceutical agents of all kinds work as artificial [high affinity] ligands to activate or block receptors that have evolved naturally to bind other [lower affinity] molecules

[§111] You do not inherit the DNA genes that encode your antigen receptors; you manufacture your own receptor genes epigenetically from genetic raw materials.

[§111] the epigenetic machinery of an individual ... is able to produce about a millionfold more diversity than does the germ-line of the species. Indeed, your immune system is as rich in the potential diversity of its antigen receptors as is your brain in the number of its nerve connections.

[§112] biologic specificity cannot be reduced to the chemistry and physics of ligand binding

[§119] The immune system ... responds simultaneously to different aspects of its target entities and to its own responses to these target features.
Recognition ... emerges as a property of co-responding populations of semi-independent agents.

[§124] Complex networking is the raw material from which cognition emerges.

[§125] all forms of life share molecular similarities. The self is really not very different chemically from its potential predators. ... Self-experience is the only experience accessible to the immature immune system, and, fortunately, protection can arise from self-experience.

[§126] If the first selection led to T cells with degenerate affinity for self, then the second selection leads to T cells with higher affinity for altered self. ... the second selection ... appears to focus on deviations from self.

[§128] Human mothers transmit a sample of their antibodies to their children. [via placental blood, and milk] ... These antibodies not only passively supply the newborn baby with ready-made antibodies against environmental pathogens common to mother and child, the maternal antibodies actively prime the immune system.

[§132] A monoclonal population of antibodies is intrinsically poor in ligand specificity because the population is uniformly degenerate. Polyclonal antibodies tend to be much more specific because each of the different antibodies features a different pattern of degeneracy.

[§138] the immune system connects the experience of the species to the experience of the individual. Immune interactions are an on-going association of somatic particulars with germ-line classes of behaviour.

[§139] Memory is another cognitive concept whose mechanism is clearer in the immune system than it is in the brain. Memory is the expression of learning from past experience.

Is there an immune system analogue of "false memory syndrome"?

[§139] Memory is the replacement of a context of infection by some antigens.

[§139] Vaccination is a way to supply the immune system with the experience it needs to learn an effective response pattern to an agent of disease without incurring the actual disease. ... the classroom lesson is never as effective as is real experience on the street. ... That's why protection by way of vaccination usually needs boosting.

[§148] [Autoimmune disease] susceptibility genes are prerequisites, not causes. ... most [identical] twins are discordant (differ) when it comes to expressing an autoimmune disease.

[§162] the immune system of [a Non-Obese Diabetic] mouse, by practising on infectious agents, can learn to become more adept at controlling its genetic tendency towards autoimmune diabetes.

[§162] The immune system, like the brain, needs experience to self-organize.

[§163] Immune responses to some infectious agents have been shown actually to induce autoimmune diseases. A clear example in humans is acute rheumatic fever...

[§163] Reactive arthritis and possible other infection-associated autoimmune diseases in humans are thought by some to be triggered by self-antigen mimicry. ... the self-antigen becomes confused with a persistent infection. ... Conflicting signals can drive brains crazy and immune systems mad.

[§177] There is no absolute antigenic distinction between the self and the not-self; the immune system can recognise and respond to either. ... the self, like life itself, is an emergent property.

[§181] creatures with complex tissues also have complex immune systems. ... The plants and the invertebrates, with their few and relatively simple tissue types, have populated the earth and thrive to this day without the benefit of even a single lymphocyte. An innate, germ-line immune system of macrophage-like cells satisfies them completely.

[§181] the complex composition of vertebrates requires extreme maintenance. Complexity dazzles, but costs; simple tissues and innate immune systems don't get cancer.

[§184] The ever-increasing complexity of human culture ... has created new opportunities for parasites and, consequently, new challenges for the immune system.

Lee A. Segel, Irun R. Cohen, eds.
Design Principles for the Immune System and Other Distributed Autonomous Systems.
OUP. 2001


Steven A. Hofmeyer. Introduction to the immune system. 2001
Thomas N. Denny. Cytokines: A common signalling system for cell growth, inflammation, immunity, and differentiation. 2001
Dragana Jankovic, Alan Sher. Th1/Th2 efferctor choice in the immune system: a developmental program influenced by cytokine signals. 2001
Howard L. Weiner. Oral tolerance. 2001
Charles G. Orosz. An introduction to immuno-ecology and immuno-informatics. 2001
Irun R. Cohen. The creation of immune specificity. 2001
Jose A. M. Borghans, Rob J. de Boer. Diversity in the immune system. 2001
Andre J. Noest. T cells obey the tenets of signal detection theory. 2001
Lee A. Segel. Diffuse feedback from a diffuse informational network: in the immune system and other distributed autonomouse systems. 2001
Eugene C. Butcher, Ellen F. Foxman, Junliang Pan, Eric J. Kunkel. Multistep navigation and the combinatorial control of cell positioning: a general model for generation of living structure based on studies of immunne cell trafficking. 2001
Ele E. Sercarz. Distributed, anarchic immune organization: semi-autonomous golems at work. 2001
John Ross, Marcel O. Vlad. New approaches to complex chemical reaction mechanisms. 2001
Eric W. Bonabeau. Control mechanisms for distributed autonomous systems: insights from social insects. 2001
Deborah M. Gordon. Task allocation in ant colonies. 2001
Luis Mateus Rocha, Johan Bollen. Biologically motivated distributed designs for adaptive knowledge management. 2001
Melanie Mitchell. Analogy making as a complex adaptive system. 2001
Stephanie Forrest, Steven A. Hofmeyer. Immunology as information processing. 2001