Integrating OpenCog and DeSTIN
Ben Goertzel
I have thought a bit about how to best integrate DeSTIN with OpenCog
to create an AGI-worthy vision system.... The attached brief document
summarizes some of the improvements I'd want to make to DeSTIN to
enable this, as well as some aspects of my proposed integration
strategy.
I estimate all this at about 2 man-years of work... which is a lot
from one perspective, but not much in terms of the goal of an
AGI-worthy vision system....
The big new idea in the mix here is to modify DeSTIN to use imprecise
probabilities, and to allow outside systems such as
-- OpenCog for cognitive biasing of perception
-- SIFT or SURF etc., for attention focusing based on visual preprocessing
to affect DeSTIN's internals via impacting the confidences of the
imprecise probabilities associated with the DeSTIN nodes.
It also occurred to me that if we replace the simple probability
calculation in a DeSTIN node with supervised learning, then we can use
a learning method that uses background knowledge and experience to
bias the learning... thus causing OpenCog's long term memory to bias
the actual predictive modeling DeSTIN does within each spacetime
reason...
Putting these ideas together, one would have a perceptual hierarchy
tightly integrated with cognition...
-- Ben
--
Ben Goertzel, PhD
CEO, Novamente LLC and Biomind LLC
CTO, Genescient Corp
Chairman, Humanity+
Adjunct Professor of Cognitive Science, Xiamen University, China
Advisor, Singularity University and Singularity Institute
b...@goertzel.org
"My humanity is a constant self-overcoming" -- Friedrich Nietzsche