Last updated: 15 Aug 2010
Installed: 15 Aug 2010
Emil's original questions are in italics:In response to my circulating an invitation to discuss this Genome-to-Architecture problem, Emil Toescu made some interesting comments and asked useful questions. I have tried to use his questions as a basis for clarifying some features of the project I am trying to describe.
Emil's Responses on 17th August to my replies to his questions are in blue
EMIL: Hi Aaron, as per usual, you are coming forward from an unexpected angle. I think that the question posed in the initial text: "How can a genome specify an information processing architecture that grows itself guided by interaction with the environment?" is slightly different from the one you disseminated subsequently: "How can a genome specify an information processing architecture that builds itself?" and the difference is probably a little deeper that the semantic difference between "grow", with an implicit intrinsic rule of development and "build" which requires not only a master plan, a blue print, but also the means of understanding and implementing this master plan. A tree grows, and a construction is built, although for the uninitiated, it might appear as just growing (cf. the growth of the hospital nearby, for the last 2-3 years). That's an interesting observation. However, I have no reason to believe that the construction of the information processing architecture of an organism necessarily has a master plan that determines everything.
Emil to Aaron Neither did I want to imply that. I was just reacting to a possible semantic trap, and wanted to “expose” it, without me saying that I do prefer one option to another. In fact, if anything, I am more inclined towards the “growing” model, through which the system is endowed with a set of basic tools, and then thrown in the middle of “real life”, warts and all, and survive one that can. The “master plan” model involves, possibly, too many implicit assumptions.
There may be an implicit master plan or 'blueprint' for information processing systems in the genomes of simple organisms that do not survive long enough to learn much (e.g. the control mechanisms used in the winged phase of ephemeridae -- whose flying behaviour presumably can't be based on anything learnt in the larval stages??) Even some more sophisticated animals, e.g. grazing mammals that very soon after birth stand up, walk (or stagger) to the mother's nipple, and can run with the herd very soon after, don't have time to learn about the complex relationships between visual information, haptic and proprioceptive information, olfactory information, and required actions of walking, running, sucking, etc. In those cases, much of the detail must be either in the genome or some combination of genome and pre-natal environment (produced by the mother's genome?) In general, in organisms, it seems that a genome specifies only a schematic design for an information processing system, whose details are determined in part by processes of learning/adaptation. Of course, the amount and kind of learning or adaptation can vary enormously and that's in part what this proposed project is about.
Emil to Aaron I think that the previous two sentences are very important to the set of ideas that you are exploring. I am not enough of a zoologist (and maybe Jackie will contradict), but I feel that the ephemerides and the grazing new born calves do not show, in the context in which you described them, “learning”, but they are just executing some programs (genetically constructed and implemented). So, moving to the last but one para above, yes, the genome specifies the schematic design, and also delivers the tools – neurones, pathways, sense organs, the lot. To which “details” you are referring is less clear to me, and there is another potential conceptual trap – the genome produces the tools, and the potential for modifications, and the later will go as follows: at the moment of the initial stages of development of one cell or another, not all the genetic program is expressed. The genes are there, the potential is there for all the cells in an organism IDENTICAL, but the path taken by one or another cell type depends on the local environment, local signalling molecules, concentration gradients of chemoattractants and repellents, etc. And then, later on, the fine tuning and/or the workings of a mature, differentiated cell will continue to be influenced by all these signals. I would presume that in the context of your generic question, all this is learning, but I think you would agree that this is “post genomic”. An important NB to which I referred in one of my previous messages, there is another important modulator of genetic message, that can have effects over much longer periods of time by repressing or activating unexpected, or uncommon genome expressions – and that is the epigenetic modulation.
(NOTE: The papers Jackie Chappell and I wrote, referenced in the web site http://www.cs.bham.ac.uk/research/projects/cogaff/misc/gentoa attempt to show that the precocial/altricial distinction used by biologists needs to be replaced by a much richer specification of types of variation. We tried to show that some genomes provide for enormous influences from the environment, of different kinds at different stages of development, some of the influences used by genetically specified mechanisms that only come into play after a lot of other developments. E.g. an individual cannot learn to detect interesting patterns in its thought processes when planning until after it has developed planning capabilities.) As regards what you described as "the means of understanding and implementing this master plan" I take it you intended this to refer to a process of construction in which there is something (or someone) in control, or partly in control, that has information about what is going on and can use that information to alter the way things are developing -- e.g. like a builder or architect who notices during a construction process that the original plan needs to be modified either because of an intrinsic flaw (e.g. there is no way to get one of the pre-assembled components into the precise position intended for it within the partly constructed building) or because of changes in the environment (e.g. unexpected shortages in the supply of some material needed).
Emil to Aaron No, I did not mean in that anthropomorphic way; and there is a danger, in the way you phrased the above, and the characters that were brought in (the builder or the architect) that before one knows we might look at a kind of a “divine” intervention, and I would like to avoid that possible confusion. Probably the most useful concepts in your sentence for what I wanted to say, are those of comparison, feed-back and adaptation. Probably best to use that old model of feedback – the planning and execution of a movement, having a sort of blueprint of the move somewhere, and having at all times measurements of errors and deviations in guiding the continuation of the move. The same type of model can be used for thinking of metabolic outputs, with the blueprint being given by a certain homeostatic value of one or another metabolic parameter. It springs to my mind that another important element in what I am talking about is that the template, the blueprint might not be (=is not!) aware of itself.
It is clear that normal humans have that sort of ability, though not necessarily at birth. Many of us have considered what we have learnt, what opportunities there are for further learning, and have taken explicit decisions about what else to try to learn, e.g. because of concerns about career prospects or because of what we like learning or doing. However, the ability to reflect on the processs of learning and development requires specific features in the information processing architecture that it is likely that not all species are capable of.
Emil to Aaron Statement of yours that brings me back to my initial query – what is this intended project about: biological handling of information, bacteria developing and navigating their environment, or cells in an embryo learning their fate (=specialization) from the environment, or even rodents learning how to manipulate the experimenters in order to get their extra rations of goodies; or, on the other hand, the higher level or learning, opened, from what we know, mainly to humans, and that involve a certain degree of self-awareness and separation between objects and subjects. You are referring to this dichotomy in your last sentence above, but later on you are talking about the AI people not being able to make any machine to have a nest-building capacity (NB – note the word commonly used in this sintagme: “building”)
It is also likely that some pathologies are related to either genetic, or environmentally caused, failures to develop the capability to exercise that sort of control. One of the reasons why I am interested in the study of types of development of animal information processing archictures is that knowing more about what they are and how they work will not only have scientific value but will also be relevant to a better understanding of certain pathologies and requirements for effective educational policies.
Emil to Aaron I feel, although I cannot cogently argue, particularly not with you, who had spent years thinking and playing with these concepts, that the learning algorithms, procedures and principles are different between the pragmatically based animal learning behaviour and the more “transcendental” (in the absence of a better word) nature of the self-aware processing of the human mind. Or, to phrase it as a question to you: is the same a mole searching for food in its burrows and expressing concerns and taking actions (networking, pleasing bosses, etc) in respect to a career prospect?
There are many people working in Artificial Intelligence and robotics who are trying to design and implement learning systems, but so far they are all extremely limited and capable of producing only tiny fragments of what intelligent animals can do -- e.g. there's so far nothing like what a nest-building bird can do. EMIL Notes -- as I have not yet been able to formulate and integrate a coherent train of thoughts on the topic. o the above note on the semantic differences maybe be worth exploring further, in an attempt to understand what is, indeed, the question asked. Insofar as I understand you I agree with you and have tried above to spell out some of the implications, though there's much more to be done. o there is always the possibility that the genome simply provides a set of generic tools (read: enzymes) and the use of information to generate a construct is independent of the nature of these tools (ie, of the genomic information), leading to the idea of different types of information: genomic information handling some - to generate transcripts of aminoacids, building up the enzymes; once these are around, they are handling a different type of information - metabolic, chemical, that is informed by the environment and generate another type of construct. Yes: there are many different kinds of information that are important for organisms of different sorts, and there are some types of organism (e.g. humans, primates, many other mammals and birds) that go on for some time acquiring abilities to use new forms of information. In computer systems engineering we have learnt an enormous amount over the last 60 years or so about kinds of mechanisms that operate at different levels of abstraction, performing very different tasks -- often referred to as using virtual machinery. (But *real* machinery which is virtual only in that it isn't detectable and describable by the methods of the physical sciences.) I suspect biological evolution 'discovered' the need for such multiple layers of control long before we did and has produced far more sophisticated multi-level information processing architectures than humans have. One of our tasks is trying to find out what the problems were (i.e. what sorts of requirements) that made such developments useful. There's a lot more to be said about this. I have tried to indicate how Darwin might have answered some of his critics regarding evolution of mind and consciousness if he had known about virtual machinery in this pdf presentation mentioned previously: http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#darwin o a third point - it becomes clearer and clearer that the genomic information is only a part of the whole story - the other part is the field of epigenetics, referring to the modulation of the genetic information (both repressing and activating it) by chemical means (methylation, acetylation). These modulatory activities are metabolically controlled (ie, at the mercy of both momentary metabolic states but also influenced by "learned" events (ie, exposures to various stimuli), to an extent that is almost generating a new wave of lamarck-ianism). We have on site a person that is heavily and from the early days both in epigenetic control and reassessment of Lamarck's ideas - he is Prof. Brian Turner. Yes I agree. The book by Jablonka and Lamb Evolution in Four Dimensions: Genetic, Epigenetic, Behavioral, and Symbolic Variation in the History of Life also makes this point. But I don't believe that enough different disciplines have combined their resources in the study of epigenetics, and for good reasons nearly all of it so far has been concerned mainly with biochemical mechanisms.
FORMATTING TO BE FIXED: Emil to Aaron You might be sceptical, but the field is progressive apace, both in linking epigenetic to normal brain functions but also to disease states.Very general:
Epigenetic learning in non-neural organisms - at http://www.ias.ac.in/jbiosci/gins68.pdf
In normal cognitive states:
Eur J Neurosci. 2009 Jul;30(1):1-8. Epub 2009 Jun 10. Epigenetic dysregulation in cognitive disorders.
Gräff J, Mansuy IM.
In various diseases states:
Neurobiol Dis. 2010 Jul;39(1):3-12. Epub 2010 Mar 18.
Disruption of the epigenetic code: an emerging mechanism in mental retardation.
van Bokhoven H, Kramer JM.
Neurosci Biobehav Rev. 2009 Sep;33(8):1227-42. Epub 2009 Jun 24.
Autism: a world changing too fast for a mis-wired brain?
Gepner B, Féron F. http://www.ncbi.nlm.nih.gov/pubmed/19559043
The developmental psychologists who should be able to contribute have generally not learnt how to think about architectures for information processing systems. I also think we can't just look at humans: we need to understand the variety of possibilities to get the deep insights we need. To be continued...
School of Computer Science
The University of Birmingham