Tuesday, June 21, from 14:00 to 15:00
Room 2006, 20F, NII
On the Epistemic Relation between Data and Information
- Some
Projectionist Considerations on Designing a Computational Space of Intelligence
In explaining and utilizing intelligence, current theories commonly
have recourse to intelligence based on representation. A system (or
agent) models certain task-specific parameters, and that internal model
(based on specific input-data) then underlies operations and
manipulations that generate the desired results. Ascribing
"intelligence" in this sense to a system involves a behaviouristic
epistemic concept, since it attributes intelligence to a system
because it solves problems that we, humans, solve by using our
"intelligent capacities" (such as logical inference). The input-data
is automatically conceived of as information, i.e. task-relevant
data. On this basis artificially constructed systems are developed as
information-processing systems.
On the other hand, intelligence involves more capacities then merely
operating and manipulating a given symbolic structure. A distinction
between our subjective perspective and given (objective) input-data is
fundamental to human intelligence. By applying specific computational
(cognitive) categories and subjective constraints and distinguishing
these from the "real world", we exhibit a subjective perspective on
ourselves and the world. This perspective enables us to identify a
specific problem and to think of a solution. Information, in this
perspective, is input-data that is modelled according to the subjective
structure of an intelligent system (i.e. us), and needs thus to be
generated out of input-data.
Information isn't just outside in the world or in the data. To produce
real intelligence a computational space has to be modelled in which
the system or agent can process the input-data in order to generate
information.
When we construct systems, we assume the generation of information (by
precisely defining the relations between system and input-data) and let
the system or agent do the computational work. But a truly intelligent
system or agent must fill in the relations to the input-data by itself,
for example so that it is able to choose between different strategies
in reaction to a problem. This space of intelligent computation and
the informational content of computation rests on the system's
ability to differentiate between representing some given input-data (as
data about the world) and projecting its own constraints on these
data. In this way, an intelligent system does not just represent the
world and execute some programmed operations, but additionally takes
its own position to the input-data.
There must a computational space in which the system can make a
decision of its own by projecting its interests, needs, and capacities
onto the input-data. Therefore the system-specific constraints have to
be modelled by the system itself, and at a higher level of computation
the system has to distinguish between its constraints and the ones
input-data demands.
In my talk, I would like to present some basic concepts to distinguish
between data and information in order to gain a basic working-concept
to design a computational space of intelligence. Although there is a
lot of conceptual (philosophical) work on the subjective
characteristics of intelligence and its relation to a world, I am
aiming at an architecture that can finally be operationalized and
modelled. AI and computer science seem to be the appropriate field for
this research, because they provide the instruments for applied
modelling. I do not focus on consciousness explicitly, which is a
quite ambiguous concept. Cognitive science is no longer serviceable
when it focuses too much on neural structures, because the research on
the brain so far does not deliver results sufficient for understanding
how the brain can instantiate intelligence. So it could help to
advance research to concentrate on the formal (epistemic) structures
of intelligence independent of their instantiation-base (inbrain or
insilico).