Representational Content in terms of Dynamics for Meta

Representational Content in terms of
Dynamics for Meta-Cognition
Jan Treur
Vrije Universiteit Amsterdam
Department of Artificial Intelligence
Utrecht University
Department of Philosophy
1
Meta-Cognition
one of the principles used to obtain cognitive
architectures for self-consciousness
e.g., Zalla in TSC’00:
the phenomenal character of experience
is conscious due to the fact that it is
introspectively accessible
2
Meta-Level Architecture
has been investigated in some depth within
Artifical Intelligence
 problem: what is the representational
content of meta-level representations ?
 dynamics and interaction between the levels
are essential 
problematic to define semantics only in
terms of the state of the object level

3
Dynamics Perspective


dynamics of a meta-level architecture:
transitions over time of combined
(object level and meta-level) states
semantics of a meta-level representation:
temporal property of traces of the object
level process
4
Interactivist Perspective
on Mental States (1)
Dynamics of mental states and their interaction
with the environment are central:
‘When interaction is completed, the system will end in
some one of its internal states - some of its possible
final states. Some environments will leave the system in
that same final state, when interactions with this
system are complete, and some environments will leave
the system in different possible final states.’
(Bickhard, 1993)
5
Past Interaction Histories
and Present Internal States
past interaction
present
internal states
6
Interactivist Perspective
on Mental States (2)
‘The overall system, with its possible final states,
therefore, functions as a differentiator of environments,
with the final states implictly defining the differentiation
categories. (..)
Representational content is constituted as indications of
potential further interactions. (..)
The claim is that such differentiated functional
indications in the context of a goal-directed system
constitute representation - emergent representation.’
(Bickhard, 1993)
7
Present Internal States and
Future Interaction Traces
past interaction
present
internal states
future interaction
8
Object Level Representational Content
as a Dynamic Interaction Property
present
object level representation
past

future
world interaction
trace
time
9
Interactivist Perspective
on Mental States (3)
In summary, representational content of mental
states need to be
grounded in interaction histories
 related to future interaction possibilities

Formalisations are needed that cover this
10
Pain Example: Mediating Role




tissue damage causes pain
heat causes pain
pain causes ouch!
pain causes future avoidance behaviour
for possible sources; e.g., wasps
Note: occurrence of this avoidance behaviour
depends on events in the world
11
Pain Example: Past Traces
the set of histories of mental property pain
PTRACES(InOnt, pain)
an example member is the following interaction trace:
t0. input:
no tissue damage,
no heat
t1. input:
tissue damage,
no heat
t2. input:
tissue damage,
no heat
12
Pain Example: Future Traces
the set of future traces for mental property pain
FTRACES(InterfaceOnt, pain)
an example member of this set:
t0. input:
output:
t1. input:
wasp present
ouch!
no wasp present
t2. input:
wasp present
t3. input:
output:
wasp present
move
13
Formalisation:
Temporal Trace Language
expressive language to specify dynamics:
 traces M as first class citizens:
explicit reference to, comparison of and
quantification over interaction histories and
interaction futures

state properties p as first class citizens: explicit
reference to and quantification over p

explicit reference to, comparison of and
quantification over time points t and durations d

discrete, dense or real time frame possible
14
Pain Example: Past Formula
M
t
a trace
a time point
a past formula representing the set of histories
of the mental property pain
P(M , t):
t1 ≤ t state(M , t1, input) |= injury 
t2 ≤ t state(M , t2, input) |= heat
15
Pain Example: Future Formula
M
t
a trace
a time point
a future formula representing the set of future
traces of the mental property pain
F(M , t):
t1 ≥ t state(M , t1, output) |= ouch! &
t2 ≥ t [ state(M , t2, input) |= wasp_present
 t3 ≥ t2 state(M , t3, output) |= move ]
16
Dynamics as Three-Level Traces
a combined three-level state:
< I, M, N >
where
N a meta-state
M an object state
I an interaction state
a three-level trace:
a sequence of combined three-level states
17
Meta-Level Representational Content
as a Dynamic Object Process Property
present
meta-level representation
past

future
object process
trace
time
18
Three Levels of Representational
Content and Dynamics
meta-level
process trace

object level
process trace

world
interaction
trace
time
19
Possible Transitions
within a Combined Trace
meta-state

meta-state
(meta-processing)
meta-state

object state
(control effectuation)
object state

meta-state
(introspection)
object state

object state
(object processing)
-------------------------------------------------------------------------------object state

interaction state
(effectuation of action)
interaction state

object state
(conceptualisation of sensory state)
interaction state

interaction state
(sensory or motor event)
20
An Architecture for Meta-Cognition

semantic content of object-representations:
past and future interaction processes

semantic content of meta-representations:
past and future mental processes

introspective capabilities in self-monitoring;
e.g., monitoring of the sensory processes by
which information is acquired, and, in
particular, of the modality (cf. Zalla, TSC’00)

meta-representations have control impact on
the agent’s own future mental processes,
focusing of sensory activities, and action selection
21
Related Approaches

Zalla in TSC’00 (nr. 278) on source modelling
difference:
no formalized architecture proposed

Cunningham in TSC’00 (nr. 272) on axiomatic
theory
difference:
no explicit reference to traces within language
22
Conclusion (1)

interactivist perspective on grounding of mental
states; relation between:
mental state
and
- interaction with the environment in the past
- potential further interactions in the future
23
Conclusion (2)

application of interactivist perspective on grounding
of meta-cognitive states as well; relation between:
meta-cognitive state
and
- mental processes in the past
- potential mental processes in the future
24
Conclusion (3)

result: three-level architecture for meta-cognition
that supports introspective capabilities and self-
awareness
25