Transcript Slide - AGI conferences

Artificial General Intelligence Conference, Washington DC, 2009
Holistic Intelligence:
Transversal Skills
& Current Methodologies
Kristinn R. Thórisson & Eric Nivel
Center for Analysis & Design of Intelligent Agents
and School of Computer Science, Reykjavik University
Transversal Cognitive Skills
Learning: An enormously wide range of things can
be learned, regarding the world and the self
Attention: To learn new things one must
understand what to pay attention to
Temporal grounding: Everything a natural
intelligence can do is grounded in time, including
its own operation
Affect every perceptual, cognitive and motor
skill in natural intelligence
Not bound to particular levels of detail
Intelligence = Architecture
To have transversal learning, attention and
temporal grounding means
the whole system learns through experience, which
means
in environments that change over time, such that
the architecture changes / evolves
Architecture = Intelligence
Implication for AGI
Architecture must have system-wide support
for transversal skills...
S/W Methodologies Focus on “objects”
Focus on “objects” in software engineering
renders (by and large) all current
methodologies inapplicable to AGI
E.g.: UML, component-oriented IT, etc.
These methodologies do not provide good
support for the construction of huge amounts
of interacting heterogeneous processes
Argument
For reaching AGI we need new methodologies
that support construction of large systems with
transversal cognitive skill properties
These methodologies need to enable us to build with
processes, rather than objects
We need to take transversal cognitive skills into
account from the very beginning
resulting in a breadth-first approach
The first systems may be dumb, but at least they will
have built-in transversal skills from the get-go