LIST OF CONTENTS
(ETERNAL DRAFT: CONTINUALLY BEING RECONSTRUCTED)
(like life on earth)
How can a cloud of dust give birth to a planet
full of living things as diverse as life on Earth?
By starting from a very powerful construction kit: physics+chemistry, and using natural selection to produce many branching layers of information-processing machinery, required for new forms of reproduction, new forms of development, new forms of intelligence, new forms of social/cultural evolution, via new types of "derived" construction kit.
All this required natural selection to discover (unwittingly) and use (unwittingly) new types of mathematical abstraction, especially abstractions with parameters that could vary -- long before there were any human mathematicians to think about mathematical abstractions.
As Turing seems to have realised: the forms of information-processing used by evolution were richer and more varied than those developed by computer scientists and engineers so far, and made essential use of chemistry.
"In the nervous system chemical phenomena
are at least as important as electrical"
Alan Turing, in 'Computing machinery and intelligence', Mind, 59, 1950, pp. 433--460
Two years later he published: 'The Chemical Basis Of Morphogenesis', in
Phil. Trans. R. Soc. London B 237, 237, pp. 37--72, 1952,
Two years later Turing was dead.
What would he have done if he had lived several more decades?
Is this a clue to the answer:
In 1946 Turing wrote a letter urging W. Ross Ashby to use Turing's ACE computer to implement his ideas about modelling brains. He expressed a view that is unfashionable among AI researchers at present (2016), but accords with the aims of this project.
He wrote:"In working on the ACE I am more interested in the possibility of producing models of the actions of the brain than in the practical applications to computing."
Trying to understand the many steps in evolution of increasingly complex biological information-processing systems: including molecules, microbes, mastodons, magpies and mathematicians.
This could lead to discovery of unnoticed mechanisms in biological brains and minds that explain the many discrepancies between animal intelligence and current AI.
The project insludes attempting to understand the roles of fundamental and derived construction kits in evolution, discussed in:
(Short form http://goo.gl/eFnJb1 ).
[NASA artist's impression of a protoplanetary disk, from WikiMedia]
- Rodney Brooks on layered architectures and evolution
- Margaret Boden's work
- Brian Goodwin (in another file)
- Peter Strawson on Descriptive Metaphysics (in another file)
- The Meta-Descriptive Metaphysics project (begun 2014 -- in another file.)
Added 10 Nov 2015
This project resulted from a miscommunication in 2011 with Barry Cooper, who had asked me to contribute to a book on Alan Turing's life and work, eventually published by Elsevier in 2013, when it won several awards. Information about the book and a table of contents can be found here: http://www.cs.bham.ac.uk/~axs/amtbook/ Alas, I learnt a few days ago, that Barry had died unexpectedly after a very short illness.
For details and links to early tributes see http://www.math.uni-hamburg.de/spag/ml/ACiE/
My personal debt to Barry Cooper
Barry and I had never met until we were both invited to contribute chapters to a book published in 2011 on Information and Computation, edited by Gordana Dodig-Crnkovic and Mark Burgin, and published (at an exorbitant cost, by World Scientific):
Barry and I reviewed each others' chapters and as a result had some direct communication for the first time, by email -- producing mutual respect. Later, in 2011, out of the blue, he invited me to contribute to a proposed Turing centenary volume originally described here: http://www.mathcomp.leeds.ac.uk/turing2012/give-page.php?300
After some discussion about how best I could contribute, which involved a change of plan mid-way, I sent him my promised three related short chapters, the first for Part I, of the book (How Do We Compute? What Can We Prove?) and the others for Part III (Building a Brain: Intelligent Machines, Practice and Theory).
However, as a result of the earlier change of plan, which Barry had apparently forgotten, I was still listed as a contributor to Part IV (The Mathematics of Emergence: The Mysteries of Morphogenesis). So he later asked me for my missing contribution to Part IV. His request led me to look at Turing's paper on morphogenesis published in 1952, two years before he died.
That triggered questions about what Turing might have done if he had lived another 30-40 years, instead of only two. So I offered Barry a paper proposing "The Meta-Morphogenesis Project" as a conjectured answer. He accepted it (as the final commentary paper in the book) and ever since then I have been working full-time on the project, for which this page is the main entry point, from which many sub-projects have been launched.
A preprint version of the chapter proposing this project is available here.
Barry later invited me to talk to a workshop at the Royal Society Kavli Centre in June 2012, followed by an invitation to talk to the Logic Seminar in Leeds in October, and then in 2013 invited me to contribute to a book The Incomputable, to be published by Springer, being co-edited by Barry and Mariya Soskova.
My chapter, finished in December 2015, not long after Barry's death, extends the idea of the Meta-Morphogenesis project by discussing requirements for many kinds of construction kit to be used evolution, including construction kits for producing physical components of organisms, information processing systems, and new construction kits. Some of the construction kits are concrete e.g. mechanisms for building components of tree trunks, while others are abstract, e.g. construction kits for reasoning mechanisms implemented as virtual machines.
I also met Barry a few times at workshops or conferences, and I always immensely enjoyed and benefited from our discussions. Originally stimulated by Barry's requests and enormously encouraged by his responses to my drafts, I now expect to go on working on these problems as long as I can.
So he changed my life, by giving me a new research direction, which does not often happen to 75-year old retired academics! (Now nearly five years older.)
I am very sad that we'll not be able to have any more conversations.
Formulating the goals of the Meta-Morphogenesis project made me realise that much of my own work of the last 40 years could be re-cast as a contribution to that project. This document, and a growing set of linked documents is my (messy, changing, inept) attempt to present the project: its questions, some of what we don't know, some of what we do know, some of the ways we can make progress, and some of the overlaps with work of other thinkers. (I suspect Immanuel Kant was attempting to work on these topics, but lacked some important conceptual tools developed in the 20th century, connected with information processing systems.)
in What is Life? CUP 1944.
I do not intend to apply for any funds for this project. Others may, if they wish.
Evolution initially uses the "Fundamental" Construction Kit (FCK) provided by Physics/Chemistry, but repeatedly extends the scope of natural selection by producing ever more complex and powerful "Derived" Construction Kits (DCKs). Some of the DCKs are concrete (physical), some are abstract, and some are a mixture: hybrid construction kits. The processes also require scaffolding: use of temporary constructs. Many of the later construction kits are concerned with production of virtual machinery: about which human engineers have learnt a great deal since the 1950s, though the new knowledge has been largely ignored by philosophers (and neuroscientists?), and the philosophical, psychological, and biological implications have been ignored by most computer scientists, biologists, psychologists, and engineers.
Construction kits intrinsically have mathematical properties: their generative powers and their constraints/limitations. That's why, to a large extent, it is easy to recognize and distinguish novel objects produced using a Meccano kit, a Lego kit, or a Tinker-toy kit. Because the generative powers, in most cases, are recursive or iterative, in the sense that constructed products of a kit can always be extended using that kit, and all the products of a kit are constrained by properties of the kit, the processes of evolution are deeply mathematical. However, when products of a construction kit CK interact with things that are not solely products of CK, the resulting processes can have features that are not derivable from properties of CK.
The theorems of evolution (the blind mathematician) are primarily theorems about what is possible including construction kits and their products. The proofs of the theorems are implicit in the evolutionary and developmental trails leading to instances of the possibilities.
Evolution also uses theorems about what is impossible: constraints on possibilities, i.e. necessities. But these do not have proofs produced directly by evolution, until some of the late products of evolution begin to notice them, think about them, reason about them, and communicate the results of such processes.
All this is part of a long and complex story that may not become clear for many years.
The need to provide a foundation for all the products of biological evolution may one day turn out to imply previously unnoticed constraints on the Fundamental Construct Kit (FCK), with important, new, implications for theoretical physics. At the very least this may help to resolve disagreements about fundamental physics. Alternatively it may reveal gaps in physical theory that require major changes. I expect new kinds of mathematics will be required, in order to bridge the gaps between fundamental physics and some of the complex products of evolution. (Compare Anderson 1972.)
Limits to reductionism
Although the products of evolution are fully implemented in physical mechanisms, accurate descriptions of the powers and actions of some of the new virtual machines require use of an ontology that is not definable in terms of the ontology of physics (e.g. concepts of winning, losing, attacking, defending, noticing a threat, trying to construct a defence). It follows that some of the properties of the virtual machines, and descriptions of some of their states, cannot be derived logically (or mathematically) from descriptions of the underlying physical machines. That non-derivability is consistent with full implementability. This is the variety of anti-reductionism proposed in The Computer Revolution in Philosophy (1978) That's also one of the main themes of the theory of Virtual Machine Functionalism (VMF) mentioned above.
An incomplete, messy, growing, collection of references to related work, including related research-projects, is here.
Feel free to suggest items for inclusion, giving your reasons. (Email A.Sloman[AT]cs.bham.ac.uk)
On 5th Jun 2012, Stuart Wray, after
reading a draft conference paper on Meta-morphogenesis and the Creativity of
Evolution (before the ideas about construction kits had been added):
produced a sketch of the ideas in the project, reproduced here, with his permission:
NOTE added 12 Mar 2015: A different answer by Andrew Hodges.
I have discovered that in 2002 Andrew Hodges asked exactly the same question about what Turing might have done if he had lived longer. He gave a completely different answer, without mentioning Turing's interest in biology, published here in 2004:
How can a cloud of dust give birth to a planet full of living things as diverse as life on Earth?
Many have asked: what sorts of physical and chemical mechanisms could make that possible, at various stages in evolution, or various stages in individual development (epigenesis) in various types of organism, group or ecosystem. They have also asked: what sorts of morphology (physical structure) and behaviour are needed at various stages of evolution or development, and what sorts of life-supporting chemistry are required. Examples of older questions, and links to further material, can be found on these Wikipedia pages (and many others):
In short, what are the causal roles of information in living things, and how do the information contents and the causal roles change over time, in individuals (at sub-cellular levels upwards), in species, in groups and in larger systems?
What information is, how many varieties there are, what can be done with it, what it can do, what mechanisms are required for these processes, are all complex questions discussed further below. An example: From dinosaurs to Birds (in another document).
What mathematical constraints? -- Topological, geometrical, physical, chemical, biological, computational, epistemological, linguistic, motivational?
D'Arcy Thompson, Brian Goodwin, and researchers included in a book of tributes to Goodwin, focused mainly on geometric and topological changes and constraints in evolution and development of physical forms (though I have not yet read all the papers carefully).
In contrast, the concerns of the M-M project include mathematical structures and constraints relevant to types of information content, forms of representation of information, modes of reasoning, types of control of behaviour, forms of learning, and other uses of information -- which are much less visible, and leave no fossil records. For these and other reasons, the study of M-M problems is still in its infancy. Far fewer researchers are equipped to think about these questions. At least physics, chemistry, and mathematics are taught to many children in schools.
Perhaps nobody is equipped yet: if some key ideas have not yet been discovered?
Some who have raised similar questions focus mainly on the evolution of human minds, e.g. Merlin Donald and Peter Gardenfors, among many others (including many I have not read).
There also seem to be overlaps with the work of Stuart Kauffman. Jack Birner (2009) has discussed ideas of Popper and Hayek related to this project.
Many of the writings of Daniel Dennett make points similar to the points made here, including his Kinds of Minds (1996), his writings on free will and others. However, we disagree regarding his claim of the centrality of "The intentional stance" for reference to mental states and processes, and his denial of the existence of the entities variously referred to as "sense-data" or "qualia", which I have argued arise naturally in sufficiently complex virtual machine architectures (e.g. in Sloman and Chrisley 2003 and this discussion of Virtual Machine Functionalism(VMF)).
Dennett and I also seem to disagree on the origins of language -- which I argue must first have evolved (with structural variability and compositional semantics, though not necessarily a "linear format", to meet requirements for information processing within organisms, not between organisms.
Terrence Deacon's 2011 book Incomplete Nature: How Mind Emerged from Matter overlaps with the ideas presented here in what seems to me to be a fairly shallow way, insofar as his book does not do justice to what we have learnt about information processing in the last seven decades, including work in Artificial Intelligence, on perception, reasoning, theorem-proving, language understanding, planning, vision, learning, and other topics related to this project. Although I have not yet read all of Deacon's book, he also seems to have no understanding of the achievements of designers of new kinds of multi-functional virtual machinery, as described in http://www.cs.bham.ac.uk/research/projects/cogaff/misc/vm-functionalism.html
Added 16 Mar 2015:
A separate document provides some notes on
In contrast with the majority of evolutionary research (that I know of), this project focuses on changes in types of information and types of information-processing in evolution. Those changes produce changes in the roles of information in control, reproduction, development, discovery, learning, communication, coordination and other processes, in living things of all sorts. But the examples keep changing, and becoming increasingly complicated, as a result of evolution.
I am not sure whether there is any well-defined upper bound to the complexity, though if there is one it is likely to be far beyond the types of complexity found so far on earth. I shall not discuss the implications of that except to note that there's no reason to believe evolution of information processing has stopped, or is close to stopping, not least because changes in physical designs for organisms, i.e. genomes, are not required for evolutionary changes in information processing: as shown by cultural evolution, including evolution of art and science, and most recently the internet and related technologies.
Is virtually unending change in information processing an inevitable consequence of the existence of the universe? I don't know. It would not happen in empty space. It would not happen on a planet derived only from grains of sand. If a planet, or solar system, or galaxy, or universe has enough diverse chemical components and enough random influences, then perhaps the unending (frequent? infrequent?) initiation of processes of evolution, including various types of meta-morphogenesis, is inevitable -- though I don't know how constrained the set of possible evolutionary trajectories is. Not even natural selection can produce chemically impossible brain mechanisms!
Perhaps, when we have a deeper understanding of the Fundamental Construction Kit provided by physics and chemistry and its role in biology, we'll be better able to discuss whether a different kind of universe might have support life in a comparable variety of forms.
It is not clear whether the variety of forms of information processing currently known to science (and engineering) can support the variety of possibilities required to support all the products of natural selection. The Church-Turing thesis, if true, may turn out to be true only of a class of computations that can be performed on numbers (plus structurally equivalent computations). We'll see that there are many forms of information-processing that are not concerned with numbers, like reasoning about continuous deformation of curved lines on curved surfaces, illustrated here. Perhaps our universe provides a wider variety of information mechanisms than the Turing-equivalent forms of computation can. (REF??? recent evidence in Biology.)
This project (since late 2014) uses the notion of construction kits and evolution of construction kits as a framework for understanding some of evolution's most complex achievements. As far as I know this is a new idea. (If not, please send me links to other work on evolution of construction kits.)
The components, relationships, and forms of composition of a construction kit (e.g. Meccano, Lego Bricks, plasticine, paper and scissors) determines what sorts of entities can be constructed using the resources of the kit. What sort of construction kit had to be available from the earliest stages of this planet, to support not only all the physical forms and behaviours of life forms evolved on this planet, but also all the forms of information-processing, not only the information-processing involved in reproduction and growth, but also all the later forms including science, art, mathematics, engineering, politics, religious superstition, ethical debates, etc? Tibor Ganti Ganti (2003) proposed a minimal set of chemical mechanisms for elementary reproducing life forms. I don't know whether he thought his "Chemotron" idea sufficient also for all forms biological information processing.
A more detailed discussion of construction-kits as explanations of
biological possibilities was inserted here in November/December 2014:
A closely related topic is the role of explanations of
possibilities in science, discussed briefly here:
A theory that explains possibilities may be deep science without being falsifiable.
(First steps toward a "generative grammar" for varieties of mechanism required for various forms of life and ecosystems, including information-processing mechanisms.)
Every time some new physical feature, behaviour, or mechanism arises in a living organism, that constitutes an implicit discovery that that sort of thing is possible, and was possible previously, though the realisation of the possibility may be more or less accessible at different stages of evolution. The evolutionary or developmental history contains an implicit proof that it is possible, but extracting the proof at the right level of abstraction may require sophisticated mathematical abilities that do not evolve till much later.
The meta-cognitive abilities even to notice that such discoveries have been made, which require a highly specialised form of information processing competence, did not evolve till very recently (resulting from a mixture of biological and cultural evolution, among other things).
Yet evolution seems to have noticed some of them "implicitly", insofar as it discovered not only very particular solutions, but also generalised patterns that were then instantiated in diverse particular cases. The "laws of form" (studied by D'Arcy Thompson and others) illustrate this: A genome does not specify the precise shape and size of an organism or its parts, but rather a network of relationships between possibilities that can vary between individuals, but even more remarkably, can vary within each individual during that individual's growth and behavioural development (e.g. learning to control movements while size, shape, weight, weight distribution, needs and opportunities all change).
Another example of evolution discovering and using a collection of powerful mathematical abstractions is use of a basic collection of learning abilities to bootstrap abilities to learn how to use increasingly sophisticated features of the prevailing language or languages: a system that was eventually able to work in several thousand different cultures using different languages.
Moreover, the evolved mechanisms in humans somehow provide transitions between having various competences (possibly recently acquired) and becoming able to think about those competences and help others acquire them.
Those transitions from competences to meta-competences (using late developing
genetic mechanisms, or learning), include the processes labelled
"Representational Redescription" in
Karmiloff-Smith (1992). They also have much in common with mathematical
discovery, insofar as both often involve finding new abstractions that have many
How can the genotype available to a newly born or hatched animal make possible
hugely (infinitely?) varied developmental trajectories in different
environments, e.g. squirrels in different gardens with (mostly) shared genomes
learning to defeat new "squirrel-proof" bird-feeders, and humans learning any
one (or any two to four?)
of several thousand very different human languages, absorbing
whatever culture the child grows up in, acquiring competences relevant to local
geographical features, local fauna and flora, local sources of food, shelter and
danger, personalities of local conspecifics, etc. and in some cases creatively
extending those environments through new inventions, new discoveries, new works
of art, new moral teachings, new mathematical proofs, etc.
One common answer is that anything with human-like intelligence must use the same (postulated) general purpose learning mechanism, e.g. Juergen Schmidhuber, (2014). [The "empty mind" hypothesis mentioned below]
Even Turing (who should have known better?) toyed with that answer in his 1950 paper, though mainly in the context of a machine that learns to have text-based interactions. All the general purpose mechanisms I've heard proposed so far operate on compressing bit-strings, or symbol-streams, and don't seem to be capable of learning geometrical or topological facts or skills, including the competences of a squirrel, a weaver bird, or a mathematician studying properties of toroidal surfaces.
(I need to check whether I have missed a more powerful, more successful, general purpose learning engine.)
An attempt to characterise the sort of rich "Evo-Devo" interaction that makes nonsense of many speculations about evolutionary or environmental determination, led to the multi-stage developmental model of Chappell and Sloman (2007) depicted below -- apparently extending Waddington's "epigenetic landscape" idea. (This sort of system must have been the product of a complex evolutionary trajectory: it could not have existed when life first began.)
What are the (mathematical) properties of physics and chemistry that enable a protoplanetary dust cloud to produce machines that can ask questions like these? (The construction-kit question again.)
Is there something about chemistry that we have not yet understood? Only with the properties of chemistry do we seem to combine three features necessary for life from its earliest stages onwards: energy storage and transformation, mechanical and sensory structures that can act on the environment when appropriate, and mechanisms for storing, using, copying, and transforming information Ganti (2003).
Chemistry builds brains, at least in their early stages, though it remains essential for many brain processes throughout life. Perhaps interacting molecules do much more than we know, long after they have constructed neural mechanisms?
Can schools and universities provide the sort of education required for researchers and teachers in this project?
Most philosophers and scientists seem to be unaware of the deep
significance of what we have learnt about many forms of virtual machinery:
The Meta-Morphogenesis (M-M or MM) project focuses instead on production of new types of biological information processing, including information-based control mechanisms, whether used for reproduction, growth, development, metabolism, perception, motor control, learning (including creation of new ontologies and new forms of representation), motive formation, planning, planned or unplanned behaviours, meta-cognition, communication, daydreaming, explaining, theory change, mathematical discovery, mathematical proofs, enjoying and producing art, or anything else. All new forms of computation that arise during evolution, development or interaction with other organisms are included. This requires use of a very general notion of "computation", or "information processing", that is not restricted to use of bit-based computers.
The changes in information processing include (a) what is done (as indicated in the previous paragraph), (b) why it is done, e.g. what benefits, if any, result, (c) what the information used is about (e.g. what it refers to, which can include past, present, future, remote, and non-existent entities, events, etc.) and (d) how all that is done, which refers to types of: information bearer, mechanisms for analysing, transforming, constructing, comparing, storing, retrieving information bearers, types of information processing architecture, combining different forms of information processing in larger wholes, types of self-monitoring, self-modulation, self-repair, self-extension, types of competition, types of conflict resolution, types of interrupt mechanism, use of virtual machinery, including multi-layer machines, distributed information-processing (involving several different individuals, or a whole community) and many more.
As explained below, the ability of natural selection to be a sort of "blind
mathematician", discovering and using mathematical structures, seems to be
crucial -- refuting philosophical claims that mathematics is a human creation.
LIST OF CONTENTS
Natural selection (or the biosphere) is a bit like a young child that has begun to learn, but has no idea that it is learning, what it is learning, how it is learning, why it is learning, what it will do with what it has learnt, why what it has learnt works and why what it has learnt sometimes proves inadequate, either for individuals or for whole species.
A difference is that over billions of years natural selection modifies its information-processing abilities far more than any child can do in a human lifetime, and modifies them in parallel in different ways. Moreover, at least on one planet, it has recently produced some individuals that have begun to understand some of what the evolutionary mechanisms produce without understanding.
The reproductive mechanisms do not normally produce ready-made full understanders, but individuals empowered to grow their understanding guided by the environment and by what some of their forebears and peers have already understood.
Some of those evolutionary changes bear a high level resemblance to the processes in individual development in animals described as "Representational Redescription" in Karmiloff-Smith (1992). In particular, it seems that increases in competence both in evolution and in individual development involve mechanisms that partition discoveries into domains with mathematical structures that can be discovered by appropriate domain-related mechanisms (not merely the use of universally applicable statistical learning techniques as some have supposed). See also the quote from McCarthy below, and the Chappell-Sloman proposal (below).
Moreover, each of these processes and mechanisms of change can impact on the others, over appropriate time-scales. That includes changing what evolution can do, by changing the resources available to natural selection -- e.g. by creating, modifying, and combining construction-kits available for evolution, development, learning and social/cultural processes.
If all that is correct, attempts to characterise any of those processes or mechanisms in a uniform way will lead to erroneous theories, because:
Natural selection may seem to be a uniform process, but what it does depends both on the mechanisms generating options between which selections can be made, and the selection mechanisms, which in turn depend partly on external constraints and opportunities -- niches. The points summarised above imply that both the types of option and the selection mechanisms can change dramatically.
Those modifications include:
One of the most important discoveries of biological evolution was the power of "generative" forms of representation of information: e.g. encoding information using trees and networks of information, whose nodes can be either arbitrary non-decomposable objects, or structured (decomposable) objects composed of other objects, for example trees and networks whose nodes contain trees and networks.
The need for such meaning structures is clear in connection with the contents of complex sentences, pictures and diagrams, with parts that have parts that have parts. The need also exists in percepts, in mathematical formulae and proofs, in complex intentions, in explanatory theories, and in action plans.
The ability to create and operate on such structures has been a pervasive feature of AI programming languages, often described as symbolic programming languages, which typically also provide standard instructions for operating on numbers of various sorts. Without this sort of capability, human language, and, I suspect, powerful animal vision systems, could not have evolved. This is why widely used forms of representation using vectors of scalar values are not sufficient for explaining how organisms work. (They don't even suffice for representing chemical structures and processes.)
This is not intended to be a complete list of information processing novelties produced by natural selection. Extending the list, filling in details, and testing ideas by empirical research into processes and products of evolution, building working models to check the feasibility of the theories, and addressing a variety of closely related philosophical problems, including problems about relations between mind and body, are all among the long term aims of the M-M project -- potentially a huge, long term project.
Achieving such goals will require, among other things, major advances in AI and robotics in order to be able to test theories of how organisms work, and may even require novel forms of physical computing machinery, for instance if some of the functions of chemical information processing, with their mixtures of continuous and discrete changes, cannot be replicated in digital computers; and new kinds of mathematics may be required, for reasoning about how some of the systems work.
In the process we can expect many old philosophical problems to be solved or dissolved and many new ones to emerge.
The remainder of this document expands on some of these points and provides links to other, related documents on this web site and to relevant publications. (A partial list)
Offers of collaboration welcome. I have no funds for this research, and do not intend to apply for funds. Others may do so.
(School of Computer Science, University of Birmingham.)
Some researchers, including (as I understand him) Juergen Schmidhuber, (2014) seem to regard the pinnacle of evolutionary design as a totally general, domain-independent learning mechanism, which allows individuals to learn in any environment by discovering statistical relationships between sensory inputs and motor outputs; whereas there seems to be plenty of evidence that humans have different kinds of learning capabilities, used at different stages of development or for different domains of structures and processes.
Compare the views of Neisser (2007) and John McCarthy who wrote:
"Evolution solved a different problem than that of starting a baby with no a priori assumptions."One way to make progress on such questions is to try to chart the variety of forms of development of information processing in young animals including humans. A subset of that task forms the investigation into "toddler theorems" (the abilities of pre-school children to make proto-mathematical discoveries, without necessarily being aware of what's happening), described in a separate file:
"Animal behavior, including human intelligence, evolved to survive and succeed in this complex, partially observable and very slightly controllable world. The main features of this world have existed for several billion years and should not have to be learned anew by each person or animal." McCarthy (1996/2008)
McCarthy's suggestion is consistent with the hypothesis that natural selection produced a variety of different learning mechanisms useful for different stages of development in complex organisms, as depicted below.
Turing's paper is not an easy read, especially for non-mathematicians, but there is a very readable introduction to the ideas in Margaret Boden's magnum opus Boden (2006). In particular, section 15.iv ("Turing's Biological Turn") gives a summary of Turing's work on chemistry-based morphogenesis (which she had read and admired decades earlier).
The previous section 15iii (Mathematical Biology Begins) summarising work by D'Arcy Thompson is also very relevant. E.g. she writes:
Turing's 1952 paper made a deep impression on me, and led me to wonder what Turing might have done if he had lived longer. My tentative (presumptious?) answer was that he might have worked on what I've called The Meta-Morphogenesis project, summarised here. The proposal for a Meta-Morphogenesis project, was first presented as a chapter (written in 2011) published as part of the Turing volume (published in 2013):
A piece of evidence that Turing might have been interested: According to his mother, he had always been interested in living things, as depicted by her in this famous drawing: https://www.commondreams.org/sites/commondreams.org/files/imce-images/turing_mother_drawing.jpg
Margaret Boden's commentary on Turing's work on morphogenesis provides this additional piece of evidence
Perhaps he would have moved (by analogy with some of his earlier moves) from studying embryology to studying the origins of embryology deep in the evolutionary past: the basis of the M-M project. (Later I'll discuss another link with Boden's work: her ideas on creativity and the varieties of creativity in natural selection, including ontological creativity, required for production of new types of virtual machinery mentioned briefly below.
This is a complex, multi-faceted project, and could take several decades, or
even much longer. Some of the main ideas are elaborated below, and in other web
pages referred to on a separate page. But at
present everything is provisional.
"In the nervous system chemical phenomena are at least as important as electrical." in 'Computing machinery and intelligence', Mind, 59, 1950, pp. 433--460I wonder if he had thought about the significance of chemistry for evolution of information processing mechanisms rich enough to support minds in a physical universe.
Many research fields can contribute, including: genetics, microbiology, ethology, developmental psychology, neuroscience, linguistics, anthropology, philosophy of science, philosophy of mind, computer science, Artificial Intelligence and robotics. The research requires us to raise new questions about what evolution achieved and how it did so, including questions about new forms of information, new uses for information, and new mechanisms for processing information. Doing that requires us to investigate new construction kits created, and then used, by processes of development, learning, and natural selection to support those developments.
Graham Bell writes in his book Selection: The Mechanism of Evolution
Living complexity cannot be explained except through selection and does not require any other category of explanation whatsoever.No: adequate explanations need to mention both selection mechanisms and enabling mechanisms, as I am sure Bell is aware.
Without enabling mechanisms, selection processes will not have a supply of new working/viable options to choose from. In that case the selection mechanisms will no longer be able to select new viable options.
Both the selection mechanisms and the enabling mechanisms can change during evolution (partly by influencing each other). As a result, we can think of the initial enabling mechanisms, provided by physics and chemistry, as a form of construction kit that natural selection eventually uses (blindly) to build new mechanisms forming an enriched construction kit. If this happens repeatedly (as has happened spectacularly with computing mechanisms in the last 60 years or so), then the most recently evolved biological construction kits may be unrecognisable to scientists who know only about the initial mechanisms.
So the M-M project requires multi-disciplinary investigations of layers of evolved biological construction kits, some of which have helped to produce new construction kits for use by evolution.
There is a useful web site listing common misconceptions about evolution here:
It does not bring out (or try to bring out) the full variety of types of explanation of evolutionary phenomena. E.g. Computer systems engineers have been discovering or inventing new types of information processing for over half a century -- especially new types of virtual machinery. It appears that biological evolution made use of a similar discovery very much earlier, for good reasons, some of them summarised here.
Familiar ideas about natural selection need to be expanded to show how small changes can build up to create increasingly complex mechanisms involved in the processes that repeatedly produce:
The project investigates how increasingly complex products of evolution produce increasingly complex forms of information processing including new mechanisms of evolution -- generalising ideas in Turing's 1952 paper on chemical morphogenesis and also the theory of meta-configured individual cognitive development presented in
That theory (and diagram) referred to processes of development in an individual -- processes that change some of the mechanisms of later development in that individual. The M-M project extends that idea to evolution, so that in this new context instead of the diagram referring only to development of individual organisms, it can also refer (loosely) to evolution of a species, or even of a whole ecosystem whose main features, including features affecting further evolution, change over time.
The M-M project has begun to identify many changes in forms of biological information processing, including transitions in mechanisms of reproduction, mechanisms of learning and development, and inter-individual and inter-species forms of information-processing. Examples of distinct types of transition in biological information-processing are being collected here.
An important under-studied transition is evolution of capabilities that led to proofs in Euclidean geometry long before modern mathematics, one of the most important extensions of human minds in the last few millennia. How did abilities to think philosophically evolve? Were the cognitive mechanisms unique to humans or did unnoticed subsets develop in other species? When will our robots begin to acquire these abilities?
The questions raised in the M-M project require long term multi-disciplinary collaborative research, perhaps comparable in scale to the Human Genome project. The relevance to philosophy of mathematics is discussed in a related web page. ____________________________________________________________________________
The older concept refers to information that has causal roles in evolution, in
animal perception, learning, motivation, acting, interacting, thinking, asking,
wondering, being puzzled, finding answers (etc.) This ancient concept was often
used explicitly by Jane Austen over a century before Shannon's work, and by many
others long before her. Several examples from her novel 'Pride and Prejudice'
published in 1813, are presented here:
Jane Austen's concept of information (contrasted with Claude Shannon's).
However, I am not claiming that Jane Austen had considered all the uses of
information relevant to biology. Readers may find it useful to try making a list
of the kinds of information they use in a typical day, and what they use those
kinds for -- or, more realistically, in a typical hour, such as the first hour
after waking, including information used getting light (if needed), deciding
whether to get up, getting out of bed, getting dressed, ...
In particular, "information-processing" here does not refer only to bit manipulation, or symbol manipulation, the operation of computers, or the sending and receiving of messages: those are all special sub-cases. In particular, the kind of information we are talking about does not need a sender and a receiver every time there is a user.
Acquiring information is finding out about something that the information refers to (or purports to refer to: it could be false information). Information contents used by an organism can come from many different sources outside or inside the organism, and can play different roles: in questions, intentions, instructions, multi-step branching plans, conditions for doing something, theories, and many more. All organisms, and many parts of organisms, including cells, use information -- and not just for reproduction. Information from external or internal sensors can turn on, turn off or modulate behaviour, which may be internal or external behaviour, or a mixture -- e.g. muscle contractions used for grasping something or for running. Working out a plan for achieving a goal uses information about the intended state of affairs to create a new complex information structure whose parts refer to possible actions, possible contents of perception, conditions for doing things, sources of missing information, and many more.
In the simplest cases, information is acquired and used immediately, with no record kept. For example, many homeostatic mechanisms, and servo-control mechanisms, use one or more sensors that continually record internal or external physical states, while response mechanisms continuously alter their behaviour in accordance with the current record, which is continually being overwritten -- online information processing. In more complex cases information received from sensors, is converted to a form that can be stored and used in multiple ways later on -- offline information processing. (A failure to understand such engineering design distinctions has led to vast amounts of confusion about differences between dorsal and ventral streams in brains. http://en.wikipedia.org/wiki/Two-streams_hypothesis)
Biological information is of many kinds, with many types of complexity, using many kinds of mechanism, for many types of purpose or function. The earliest organisms must have been restricted to "online" mechanisms (e.g. using chemotaxis). One of the tasks of the M-M project is to investigate the variety types of "offline" use of information, including uses to refer to the past, to remote (currently unsensed) parts of the environment, to possible future states of affairs, and to possible past or remote entities, about which questions are raised and theories formulated. The variety of types of use of information may be far greater than any theory of information produced so far has proposed. Some of the deep questions are concerned with the extent of use of mathematical structures in biological information processing.
For more on the concept of "information" used here see Sloman (2010) [in a separate web page].
Related Videos (Moved to another file 24 Aug 2014)
Long slide presentation introducing the Meta-Morphogenesis project ____________________________________________________________________________
Return to list of contents
The universe is made up of matter, energy and information, interacting with each other and producing new kinds of matter, energy, information and interaction.
How? How did all this come out of a cloud of dust?
In order to find explanations we first need much better descriptions of what needs to be explained.
This is a multi-disciplinary project attempting to describe and explain the variety of biological information-processing mechanisms involved in the production of new biological information-processing mechanisms, on many time scales, between the earliest days of the planet with no life, only physical and chemical structures, including volcanic eruptions, asteroid impacts, solar and stellar radiation, and many other physical/chemical processes (or perhaps starting even earlier, when there was only a dust cloud in this part of the solar system?).
The "proofs" of discovered possibilities are implicit in evolutionary and/or developmental trajectories.
The proofs demonstrate the possibility of
development of new forms of development evolution of new types of evolution learning new ways to learn evolution of new types of learning (including mathematical learning: by working things out without requiring empirical evidence) evolution of new forms of development development of new forms of learning (why can't a toddler learn quantum mechanics?) how new forms of learning support new forms of evolution how new forms of development support new forms of evolution (e.g. postponing sexual maturity until mate-selection mating and nurturing can be influenced by much learning) .... .... and ways in which social cultural evolution add to the mixThese processes produce new forms of representation, new ontologies and information contents, new information-processing mechanisms, new sensory-motor morphologies, new forms of control, new forms of social interaction, new forms of creativity, ... and more. Some may even accelerate evolution.
More on connections between natural selection and mathematical discovery:
Updated: 27 Oct 2016
Biology, Mathematics, Philosophy, and Evolution of Information Processing
Return to list of contents
A growing list of transitions in types of biological information-processing:
Biology, Mathematics, Philosophy, and Evolution of Information Processing
Mathematics is at root a biological, not an anthropological, phenomenon
(as suggested by Wittgenstein). But its possibility depends on deep features of
the universe, some of which evolution had to 'discover':
An attempt to identify a major type of mathematical reasoning with precursors in
perception and reasoning about affordances, not yet replicated in AI systems:
Even in microbes
I suspect there's much still to be learnt about the varying challenges and opportunities faced by microbes at various stages in their evolution, including new challenges produced by environmental changes and new opportunities (e.g. for control) produced by previous evolved features and competences -- and the mechanisms that evolved in response to those challenges and opportunities.
Example: which organisms were first able to learn about an enduring spatial configuration of resources, obstacles and dangers, only a tiny fragment of which can be sensed at any one time? What changes occurred to meet that need?
More examples to be collected here:
For a messy, still growing, collection of examples relating to learning and development
see this web page on "Toddler theorems":
(including an introduction to the idea of
http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#talk102 (PDF) Meta-Morphogenesis: of virtual machinery with "physically indefinable" functions (Slides for presentation given at the Workshop "The Incomputable" (superseded) Royal Society Kavli Centre: 11-15 June 2012) http://www.mathcomp.leeds.ac.uk/turing2012/inc/
Return to list of contents
PAPERS WITH FURTHER DETAILS
A collection of examples extracted from the papers and presentations, along with some new examples based on things I have read and conversations with friends and colleagues. Some of the examples illustrate portions of the process of information re-organisation (perhaps instances of what Karmiloff-Smith means by "Representational Redescription"?).
The list of examples is a tiny sample. I shall go on extending it. (Contributions welcome.)
PAPERS ON META-MORPHOGENESIS
RELEVANT PRESENTATIONS (PDF)
A sample list of types of transition produced by biological mechanisms The mechanisms include evolution by natural selection, individual learning, cultural development and transmission, including changes in genomes as well as changes in factors affecting gene expression.
Constant extension of what needs to be explained by science.
(Added 27 Oct 2014)
As new, more complex forms of evolution, development, learning, perceiving, reasoning, communicating, collaborating, technology, ... keep arising out of the interactions between things that existed previously, along with natural selection (possibly extended by non-natural selection!), there is a never-ending stream (trickle? flood?) of extensions to the phenomena that science needs to explain. For example, it seems likely that at some stage our evolutionary ancestors lacked some of the mathematical abilities that now exist in humans. How those abilities evolved, what new things they make possible, how they make them possible, are all questions for science that would not necessarily have been thought of by scientists observing those ancestors who lacked our mathematical abilities.
It was not possible until recently to ask the questions raised in: http://www.cs.bham.ac.uk/research/projects/cogaff/misc/entropy-evolution.html
This leads to the conjecture that the space of possible forms of information processing that need to be explained by science is at least as complex as the space of mathematical problems that arise in the arithmetic of natural numbers. And we know that that space has unending complexity.
If all this is correct there could never be a time at which all scientific questions will have been answered, not even if all questions about the underlying physical/chemical mechanisms that make life possible have been answered. That would be analogous to having a set of axioms for number theory. One of the great discoveries of the twentieth century, due to Gödel and others, was the infinite supply of unanswered mathematical questions that arise from the basics of arithmetic. Whether only a finite subset of the questions are worth answering looks unlikely.
Things that cause changes can produce new things that cause changes. Old phenomena may be produced in new ways: e.g. both types of information acquired and ways of acquiring and using information can change. Often new mechanisms can produce new biological phenomena
-- e.g. organisms that can discover what they have learnt.In particular, most forms of biological information processing that exist now are products of parallel trajectories of biological information processing over many stages of evolution and development, including cultural evolution in the case of humans.
-- organisms that make and use mathematical discoveries.
This is quite unlike use of evolutionary computation (GA, GP, etc.) with a fixed
evaluation function, often used to solve engineering problems.
For example, evaluation in natural evolution keeps changing, as environments, including competitors, prey, symbionts, diseases, etc. change.
Return to list of contents
His ideas overlap with (and probably helped to influence) those presented in this project, but there are also deep differences. E.g. I don't claim that we can start building simple organisms with our current technology that provide suitable old, 'lower level' layers on which to add newer, more sophisticated, layers of competence. I am, however, suggesting that what is old in the evolutionary history of existing organisms may have many unobvious features that the M-M project should attempt to uncover.
I also don't propose that it will suffice to start from multi-cellular organisms like insects, that have already evolved capacities to move around in rich and complex environments, foraging, feeding, mating, building nests, etc. Instead I consider the possibility that even at the single-celled level there may have been forms of information processing that underpin some of the types of information processing that interest us in humans and other animals.
Brooks' suggestion that the importance of internal representations has been over-rated because the best representation of the world is the world itself, has been highly influential, but is at most relevant to what I've called 'online intelligence' involved in control of movements and manipulations using feedback mechanisms of various sorts. (H.A.Simon made similar points.) For deliberative and meta-semantic competences the slogan is not merely wrong: it has been positively harmful.
Also the ideas in the CogAff project and the CogAff architecture schema allow for a richer variety of types of architecture than the type of layered subsumption architecture proposed by Brooks, though it's possible that each could be modified to cover more of the features of the other.
His work had enormous influence in many research and teaching centres. Unfortunately the people influenced were often much less intelligent and less subtle than Brooks, and as a result much of the influence has been bad. Hence my critique.
David Kirsh wrote a critical review of Brooks' ideas around 1986, published in 1991 (here.). Brooks wrote a reply ('From earwigs to humans') published in Brooks (1997). I wrote a somewhat different critical commentary much later Sloman (2009)
Most of the references have been moved to a separate file:
A slightly messy PDF version is also available:
This is one of a set of documents on the meta-morphogenesis project.
A partial index of a wider collection of discussion notes is in
This version installed: 21 Oct 2012
Original version installed: 19 Oct 2011 now here.
UPDATES (A partial list):
27 Oct 2016: Exended the analogy of evolution as blind mathematician by characterising the relations between the fundamental construction kit and all derived construction kit as closely analogous to proposed relationships between foundations of mathematics and all derived kinds of mathematics.
Jan-March 2015: added separate pages on construction-kits, and explanations of possibilities.
19 Dec 2014: added reference to initial construction kit in introduction. Added link to related projects in this file.
8 Nov 2014: added link to new paper on entropy and evolution.
27 Oct 2014: Added a bit at the top about origins of this project. Slightly reorganised and extended various portions.
Made some relevant additions to the (disorganised) notes on Virtual Machine Functionalism - VMF)
17 Sep 2014: Added more structure to the introduction, with subheadings
14 Sep 2014: New experimental top section. Is it too confusing? Does it sound like clap-trap to the uninitiated?
8 Sep 2014: slight rearrangement. Some new references.
24-5 Aug 2014: considerable reorganisation, with most references moved to here.
15 Aug 2014 added Birner's paper on Hayek and Popper;
7 Aug 2014: minor changes;
30 Jul 2014: added link to Strawson and meta-descriptive metaphysics moved to another file
5 Apr 2014 (Doyle and Popper links); 17 May 2014; 12 Jun 2014
31 Jan 2014: added new introduction and reorganised; 10 Feb 2014: Minor eds;
12 Nov 2013 (Added comparison with ideas of Rodney Brooks.);19 Nov 2013
2, 16 Aug 2013; 24 Aug 2013 (re-formatting); 6, 29 Sep 2013; 31 Oct 2013;
(Adam Ford Video fixed) 24 June 2013;
2 Feb 2013; 24 Apr 2013; 4 May 2013; 20 May 2013; 17 Jun 2013;
6 Dec 2012 19 Dec 2012; 21 Oct 2012 (Split in two: other part here.);
10 May 2012; 22 May 2012; 19 Jun 2012; 29 Jun 2012; 7 Jul 2012; 24 Aug 2012; 13 Oct 2012; 14 Nov 2012;
20 Oct 2011; 22 Nov 2011; 21 Feb 2012 (Appendix);5 Mar, 19 Mar, 23 Apr 2012;
School of Computer Science
The University of Birmingham