School of Computer Science THE UNIVERSITY OF BIRMINGHAM CoSy project CogX project

IRLAB Seminar: Introduction to Meta-morphogenesis (30 Apr 2012)
(DRAFT: Liable to change)

Aaron Sloman
School of Computer Science, University of Birmingham.
(Philosopher in a Computer Science department)

Installed: 30 Apr 2012
Last updated: 30 Apr 2012
This paper is
A PDF version may be added later.

A partial index of discussion notes is in

Date: Mon, 30 Apr 2012
From: Aaron Sloman
To: irlab
Subject:Irlab meeting today (30th April) 3pm room 225 meta-morphogenesis

Last Monday I wrote:

> If nobody else wants the slot next week I'll be happy to give an
> informal, interactive, introduction to some ideas about
> meta-morphogenesis, partly expanding on previous presentations of the
> ideas:
> (I am working on a conference paper on this.)

Nobody offered an alternative, so you'll get another of my rambles at
3pm today. For people who don't know me here's an introduction to the

Biological evolution (in cooperation with a host of environmental
features, some of them products of evolution) has obviously produced a
staggering variety of physical forms on many scales from sub-microscopic
to individual organisms weighing tons (the largest of which, I believe
is a fungus), and also multi-organism ecosystems of many kinds.

It has also produced a staggering variety of types of physical motions
and other physical/chemical changes -- of size, shape, speed, colour,
chemical constitution, strength, rigidity, flexibility, elasticity,
articulation, and individual behaviours in which those features and
relationships to things in the environment change.

Many of those processes remain utterly amazing, including multi-stage
developmental trajectories, such as: fertilised egg -> larva/caterpillar
-> pupa containing soup, -> flying insect -> new eggs, ....

I think much of that was obvious (after years of research) to Darwin and
his contemporaries, though new details are constantly being added to the

What Darwin and his contemporaries were unable to think about, or to
discover, because they lacked the required concepts, was the staggering
variety of forms of information-based control required for, and enabled
by, the other more obvious changes. Most people on the planet, including
many philosophers, psychologists, neuroscientists, educators,
politicians, social scientists, economists ... are still unable to think
about forms of control that are not merely physical (e.g. geometrically
constrained movement produced by changing forces) and, not merely
quantitative (like gravitational control of planetary motion, or changes
in signal strength or frequency), but are also *informed*: i.e. those
forms of control cannot be understood or modelled or replicated without
understanding the information-processing involved.

What is not at all obvious is how many types of information content
(ontologies), types of information encoding, and types of information
manipulation, biological evolution produced -- stimulated and aided by
changing features of physical and biological environments, then later
social environments.

Some of those forms of information processing produce new forms of
information processing, e.g. through learning and development.

The notion of "morphogenesis" was originally applied (e.g. by Turing and
others) to controlled physical changes of form. We can extend it to
include changes in information processing, especially types of informed

Some of the simplest types, though not necessarily the oldest types,
include homeostatic feedback loops using only quantitative variation in
sensory and motor signals, e.g. kinds that can be expressed in
differential equations. More complex forms require use of information
about structures and structural changes, e.g. using grammars, and their
derivatives such as parse-trees, graphs, etc.

Meta-morphogenesis includes changes in the morphogenesis of information
processing, e.g. evolving evolvability, learning to learn, developing
new forms of learning, and recursive interactions between these.

Previously evolved cognitive capabilities (e.g. perceptual, or reasoning
abilities) can be used in mate-selection processes or selective
nurturing of offspring, both of which can speed up evolution, as can
selective breeding of other species.

Understanding all of that (and more) requires us to understand the
variety of types of information content, the variety of types of forms
of representation/encoding of information, the variety of types of
mechanism for acquiring, interpreting, storing, manipulating,
constructing, deriving, and using information (is that list complete?),
and also the variety of types of information-processing *architecture*
within which such processes can occur, interact and cooperate,
especially self-constructing architectures.

And we need a way of investigating and characterising the variety of
biological (including cognitive, cooperative and social) functions that
all those things can serve.

The meta-morphogenesis project is not a three, five, twenty, or forty
year project -- though I don't know how long it will take to acquire
something like the breadth and depth of the physical sciences, which
took centuries. It is very difficult, for it requires us to do one of
the hardest things in science: extend our forms of thinking, as opposed
to applying existing forms to new problems.

The development of science is a type of example of meta-morphogenesis.[*]
Another, closely related, type is development of mathematics. Smaller
scale examples include the processes Karmiloff-Smith calls
"representational redescription" (in Beyond Modularity, 1992).

As many of you know, I have been struggling for a long time (with help
from many of you) to articulate the problems and identify possible lines
of progress (at least since 1971, when I wrote my first AI paper).

I think one way of pushing ourselves forward is to try to identify
changes in evolutionary pressures (changes in niches) that created
opportunities and/or requirements for new forms of
information-processing (including new information contents, forms of
representation, mechanisms, architectures, applications, interactions)
and in particular cases where the new developments can contribute to
future developments in new ways. (Hence the 'meta-'.)

One of the requirements for making progress is finding good ways to
decompose this huge and intractable collection of problems into
manageable sub-problems.

Today I'll invite people to consider some of the earliest forms of
meta-morphogenesis that might have been involved in transitions in types
of information processing from microbes in chemical soups to articulated
organisms, with multiple needs, in structured partly static and partly
changing environments requiring actions on different spatial and
temporal scales.

For practice, try to identify changing information-processing
requirements from single-celled organisms in chemical soups to things
more like ants.

I hope to learn from your ideas.

The earliest forms of information processing used chemical mechanisms.
I suspect chemical information processing continues to play a far more
important role than most researchers acknowledge.
E.g. it's required for building brains, a process that goes on in humans
for many years.
There was a wonderful BBC4 program last night (nearly ruined as usual by
obtrusive acoustic accompaniments inserted by knob-twiddling producers
who don't understand the difference between entertaining diversions
and mind-stretching educational communications) still available here,

on the development of ideas about electricity in the 18th century
(including people like Cavendish, Franklin, Volta, Galvani, and others,
all struggling to create an ontology to make sense of the exciting and
mystifying observations of electrical phenomena, not all originally
classified as such).