School of Computer Science THE UNIVERSITY OF BIRMINGHAM CN-CR Ghost Machine

CoSy project CogX project XX

                    --------------- Talk for ---------------
The World Inside the Brain:
Internal Predictive Models in Humans and Robots

Thursday 23rd (09:30) - Friday 24th May 2013
The University of Birmingham

http://www.birmingham.ac.uk/research/activity/cncr/news/24May-wib-conference.aspx
https://www.dropbox.com/s/dekh0844jwndqgh/WIB_program.pdf


Online vs Offline intelligence: how and why the latter evolved and develops
         --- or ---
Evolved connections between worlds inside and outside brains.

Aaron Sloman
http://www.cs.bham.ac.uk/~axs

This paper is
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/wib2013.html
A PDF version may be added later.

A partial index of discussion notes is in
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/AREADME.html

Abstract:

There has been a vast amount of research in ethology, psychology, neuroscience, robotics
and AI concerned with how organisms or machines can perceive and act in a more or less
complex environment. In recent years a particular strand in this research has grown
followers proclaiming the importance of embodiment and enactivism, sometimes in opposition
to cognitivism or symbolic AI (including planning, reasoning, theorem proving, and
learning structural descriptions).

The Meta-Morphogenesis project offers a different stance: we can think of embodied agents
of various sorts as being embedded in environments of varying complexity and with varying
challenges and opportunities, which resulted in evolutionary changes in both sensory-motor
morphologies, and forms of information-processing, including development of new more
complex information-processing architectures, more varied forms of representation and more
complex uses of information.

XX

The need to understand evolutionary transitions

It is commonplace in discussions of evolution to talk about changes

Developmental biology also investigates such changes in the morphology, behaviours,
environments, capabilities, challenges, opportunities, during development of individuals
including changes before and after separation from mother or hatching from egg or cocoon, etc.

We need to add to such investigations the transitions in information processing including, for example, changes in

and many more?

There may be some continuous changes (e.g. increase in size, or speed) but the important
changes in information-processing are discontinuities.

Identifying the many past discontinuities may give us new deep insights into existing
information processing mechanisms and capabilities that are too complex and intricate to
be directly inspected, either in behavioural experiments/observations or physiological
observations/measurements. For a growing list of transitions see:
http://tinyurl.com/CogMisc/evolution-info-transitions.html

This talk will highlight particular types of evolutionary transition, from direct,
immediate and practical interaction with the immediate environment, using "online
intelligence" to increasingly indirect remote and theoretical engagement,
or "offline intelligence".

Online intelligence includes for example, chemotaxis, sensory triggering, reflex
responses, and various types of servo-control, including walking, running, grasping or
catching things, all of which require detailed and accurate information about states
and processes in the immediate environment, to be used and over-written almost
immediately.

Offline intelligence involves use of information about collections of possibilities,
constraints on possibilities, chains of sets of future possibilities, and also backward
reasoning concerning possible explanations of perceived states, events and processes.
Often instead of precise and detailed information (e.g. measures of distance, direction,
speed, angle curvature, etc.) it abstracts away from such details in order to represent a
range of possibilities and some of their invariants, or constraints, and may even include
branching sets of possibilities.

The information used in "offline intelligence" neither refers to the precise details
of what's going on here and now, nor is restricted to immediate use in controlling
actions, It is very closely related to human mathematical competences, I suggest that
detailed research will find many examples of offline intelligence that can be seen as
precursors to mathematical reasoning that evolves or develops later.

Note that possibilities are not probabilities: possibilities are intrinsically unordered,
for example, and need not form a metric space, though many sets of possibilities can over
time, have their ontology enriched to include metrical properties. For example, they may
start as partially ordered sets.

One of the striking results was production of abilities in our ancestors that led to
discoveries in Euclidean geometry before there were teachers or textbooks, and without
which Euclid's Elements, one of the high points of human intelligence, would not have been
possible. I suspect an earlier transition, in the evolution of many more species, was
evolution of abilities to discover and reason about affordances and constraints on
affordances prior to making use of those affordances, without which corvid nest building,
elephant adults helping their infants, and many other examples of non-human intelligence
would have been impossible. Likewise intelligence in pre-verbal humans, who discover what
I call 'Toddler Theorems'.

Unfortunately, we still lack theories or mechanisms able to explain the main reasoning
processes involved, and neither the vast amounts of research on bayesian mechanisms nor
the work on powerful logical and algebraic theorem provers seems able to fill the gap. I
can't yet provide mechanisms, but I'll present some clues about a way forward.

Design space and niche space and interacting trajectories

Mappings:
XX

Trajectories:
XX

Altricial-precocial Trajectories:
XX

Compare Karmiloff-Smith's, Beyond Modularity (1992):
http://tinyurl.com/CogMisc/ beyond-modularity.html

Information-processing architectures for behaving systems

The CogAff schema
XX

Possible background reading:


Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham