The Emergence of Interactive Meaning Processes in

AutonomousSystems

Argyris Arnellos, Thomas Spyrou, John Darzentas

University of the Aegean Dept of Product & Systems Design Eng

Syros, Greecehttp://www.syros.aegean.gr

General Problem

Theoretical frameworks of cognition are differentiated by the way they handle the notions of intentionality, meaning, representation and information.

One could ask: How is meaning generated and manipulated in natural and consequently in artificial cognitive systems?

MAIN APPROACHES to GOGNITION and to the DESIGN of ARTIFICIAL AGENTS

Cognitivism and Computational Artificial Agents• All intentional content is a kind of information which is externally

transmitted by a merely causal flow.• Meaning is externally ascribed

An Objection: Searle’s Chinese Room Argument An Answer: (Harnad, 1990): Symbol Grounding is an Empirical Issue

Note: A complete cognitivist grounding theory should consider both external and internal representational content, as well as, their transduction system and its interactive nature.

Main candidate for the transduction is a connectionist network

Connectionism• Connectionist systems as syntactically adaptive

systems (not all of them)

• Such machines receive (contingent) feedback from their outputs, which then directs the adjustment of their decision function (i.e.new percept-action mapping).

– No semantics as the system cannot decide on its own which aspects of the world must be encoded ("feature primitives") such that the machine can find a successful classification rule.

– Adding a "training" origin to correspondence works no better than adding lawfulness

– Connectionistic architectures cannot account for internal content.

Dynamic Systems and Cognition• Time-Dependence: Natural cognition happens in

real time, hence dynamics is better suited to model it than the a-temporal computational approach.

• Embodiment: Cognition is embedded in a nervous system, in a body, and in an environment, whereas computationalism typically abstracts this embeddedness away, and can incorporate it in only an ad hoc manner.

• Emergence: Dynamics can explain the emergence and stability of cognition through self-organisation, whereas cognitivism ignores the problem of cognitive emergence.

Dynamic Systems and Cognition

• No information processing (no symbols, no representations)

• The dynamics of the cognitive substrate (matter) are taken to be the only thing responsible for its self-organization

• System’s ability for classification is dependent on the richness of its attractors, which are used to represent events in its environment

System’s meaning evolving threshold cannot transcend its attractor’s

landscape complexity

Defining Agency

Strong notion: An agent is a system which exhibits:• interactivity: the ability to perceive and act upon its

environment by taking the initiative;• intentionality: the ability to effect goal-oriented interaction

by attributing purposes, beliefs and desires to its actions;• autonomy: the ability to operate intentionally and

interactively based only on its own resources.

[Collier, 1999] suggests that there is: no function without autonomy;

no intentionality without function;no meaning without intentionality;

Circle closes by considering meaning as a prerequisite for the maintenance of system’s autonomy during its interaction.

The Need for a New Kind of Representations

• “As tasks become more complex the use of internal states that carry information about the environment becomes less and less avoidable.” (Kirsh, 1991)

• “…even in the very simple cases mentioned above we find that individual units act as very simple representations in mediating interactions between the robot and its world” Brooks (1997)

What kind of representations do we need?No representations per se, but a different

type of representation

The Need for a New Kind of Representations

• Representations that can only be understood in the context of activity.

• For an adaptive system the primary problem is to produce action appropriate to the context, not referentially individuate a signal source (cognitivism).

• The content should be accessible to the system itself.

Functionality and Representations

• A behaviour is ‘really’ contributing to systems functionality if and only if it is mediated by representations and

• an information-carrier is only a representation if it plays an appropriate role in the systems functionality towards its self-maintenance

Where can these kind of representations be found and

What type will they be?

Code Duality in protein structure/sequences based on (Hoffmeyer & Emmeche, 1991)

Analog Information Space: protein functional conformations

Digital Information Space: amino-acid sequences

AIS

H K

10101010110100010011001DIS

Ksequence:structure

Levels of Interactive Representations (Bickhard, 1998)

Interactivism and Function• Function is a forward looking concept as it tries to explain

what is its future value to the system.

Recursively Self-Maintenant Systems• System has alternative ways of self-maintenance available

and it can switch one alternative to another in case of failure. • The conditions under which the serving of a function

succeeds constitute the dynamic presuppositions of those functional processes.

a minimal ontological representative system (S) has to include a subsystem, a differentiator (Dif), engaging

in interaction with its environment (Env).

Levels of Interactive Representations (Bickhard, 1998)

• Internal course of that interaction will depend both on the organization of the subsystem and on the interactive properties of the environment.

• Each final state classifies all of the environments together that would yield that particular final state if interacted with. Each possible final state (FS) will serve as a differentiation of its class of environments

Env1

Env2

Env3

System

Dif

FS1

FS2

Goal System 1

Goal System 2

Pi

Pj

Emergent Levels of Interactive Representations

Level 3: Implicit definitions of environmental categoriesFinal states reached can be considered as a digitalisation of the

analog-analog interactions in the internal of the system due to its contact with the environment.

Minimal information: • There is no information concerning anything about that

environment beyond the fact that it was just encountered and that it is not the same as those environments differentiated by any of the other possible final states.

• There is no representational content involved

• System has no information about the classes of environments that it implicitly defines.

Emergent Levels of Interactive Representations

Level 4: Functional Interactive PredicationSystem strives to achieve maintain self-maintenance and through its

interactions builds a new level of organisation (new representational level) where the implicit environmental differentiations of Level3 are re-organised as quantitative variety of functional predications about the environment.

From Level3 Level 4 - Minimal representation: • The whole system at this moment (FI) interprets the signs of

Level3 as Dynamic Interpretants at Level4• Differentiator’s final states (FS) (Representamens) indicate

which further procedures might be appropriate and the goal-system selects from among them.

Analog-driven Emergence where new predicates are formed

Levels of Interactive Representations

Env1

Env2

Env3

System

Dif

FS1

FS2

Goal System 1

Goal System 2

Pi

Pj

System’s HabitRepresentamen

Dynamic Interpretant

Levels of Interactive Representations

Level 5: Implicit definitions of environmental propertiesInteraction with the environment continuous and the AIS of

Level4 are locally interacting in various time scales in order to reduce uncertainty for the environment.

A in a way more compressed digital record emerges and we have a transition from

implicit definitions of environmental categories implicit definitions of environmental interactive properties

From Level4 Level 5: Emergence of functional relations among system’s

organisations that involve such implicit definitions:

Implicitness and presupposition is observed which can account for unbounded representationality.

Levels of Interactive Representations

Level 6: Emergence of organisations of interactive potentialities

Level’s 5 representations are implicitly being selected by system’s differentiating interactions in a “statistical manner” formation of aggregates of properties that are presently available.

– These aggregates are ongoingly updates construction of new indications and changing old ones formation of apperceptive procedures.

From Explicit Situation Images Implicit Situation ImagesImplicit definitions of environmental interactive properties –

Level5) ENGAGE in various apperceptive procedures driven by Level4 and forms organisations of interactive potentialities

Levels of Interactive RepresentationsLevel 7: Emergence of objects and Constructive MemoryThe organisations of indications of interactive potentialities are used in

system’s interaction and in some cases they tend to remain constant (invariant) as patterns.

The quantitative variety of the organisations of interactive potentialities of Level 6 is re-organised as certain types of organisations (based on their temporal coherence).

Such types of organisations of interactive potentialities constitute objects for the system itself.

Memory: System is able to expand its situation image without explicit bound system represents such invariances in its situation image.

Constructive MemorySystem is able to test past apperceptive processes in present and differing

directions.

Levels of Interactive RepresentationsLevel 7: Emergence of objects and Constructive

Memory

At this phase, icons and indexes can emerge into the system but not symbols, for which genuine social communication is needed.

Symbols will need the lower levelsNotes: • A useful framework for Alife and AI experiments

(since, interactive representations need only simple control systems)

• It seems that initially two non-semiotic levels should exist?

IOII

FI

sos of represe

ntatio

nal structure

DI

pragmatics system’s history

Testing anticipations

object

II

FI

Morphod

ynam

ics

action

COGNITIVE SYSTEM

ENVIRONMENT

DO

Memory-based analogy making

measurement

PS PS

PSPS

Rule-based syntactic complexity

DO DO

DO

signs

Abduction

Deduction

Induction

Full Semiotic and Representational Capacity