*
*
*
*
*The Center for Common-Sense

*Marvin Minsky, Push Singh, Walter Bender

*Our Vision

*We seek to build a machine that is capable of
commonsense thinking, one
that can solve real problems for real people. This
machine will be based on
a novel architecture envisioned by Marvin Minsky,
Push Singh and others and
will be able to reason about social relationships,
physical events,
linguistic actions, sensory perceptions, and other
central aspects of
everyday human thinking. We expect that the
knowledge we acquire from
building this machine will lead to both practical
applications as well as a
detailed model of human intelligence.

*Plan of Work

*We plan to start by developing a �cognitive
architecture� that will support
human-level commonsense thinking. We will apply it
initially to reasoning
about how people think about solving several types
of everyday problems;
this will lead to later versions that we expect to
go far beyond the
present-day �search engines� and �expert

that are used today to
help people solve the problems they face in life.

We anticipate that in three years we will be able to
produce a prototype of
this cognitive architecture. This prototype will be
based on the development
of several key mechanisms:

�* Panalogy*: a new, more flexible way to

information that allows
the use of multiple viewpoints simultaneously.
Unlike existing programming
systems, our Panalogy-based system allows us to use
multiple ways to
represent knowledge and processes, so that when one
approach fails we can
quickly switch to another.

�* Goals*: Knowledge in our system will be

around descriptions of
human goals, so that relevant information will be
automatically accessible,
given descriptions of the present problem situation.
Most present-day ways
to organize information are based on top-down
logical classifications,
whereas our system will be largely based on
�sideways� links that permit
more reasoning by analogy.

�* Critics*: In order to identify and solve the

common kinds of
everyday problems, we will program a large
collection of �critics� that
capture problem-solving methods�such as ways to
decompose each difficult
goal into several simpler ones that we already know
how to achieve.

�* Self-Reflection and Multiple Models*: Unlike
other present-day systems,
the Panalogy architecture will be able to do
�reflective thinking� by making
and using models of the processes and activities
that it has been using.
This will facilitate many new ways for the system to
learn to improve
itself.

We will test the architecture by developing a
web-based application that
will aid people with typical human goals in such
fields as health,
education, and entertainment. The system will
recognize its user�s goals
and help them plan courses of action by making use
of commonsense reasoning
machinery about time, space, events, beliefs,
causality, etc. We plan to
implement this application as a front end to an
existing search engine such
as Google, Yahoo, or MSN.

We have already implemented an initial prototype of
this architecture as
part of Push Singh�s PhD dissertation. This
prototype focused on the lower
levels of Minsky�s Emotion Machine architecture:

reactive, deliberative,
and reflective levels. Our new system will focus on
that architecture�s
higher level, more reflective and more
self-conscious ways to think. These
higher levels (which are missing in existing
systems) are populated largely
with knowledge and processes about problem solving
itself, such as:

---Ways to represent and organize goals,
--- ways to retrieve relevant information,
--- ways to represent what we know and what we don't
know,
---ways to use sources of external information.

To scale up this architecture we will use many
resources, including a large
community of web users that the system will use to
help build up specific
bodies of both commonsense and specialized
knowledge, as has been done in
Wikipedia and in our own �Open Mind� distributed
knowledge acquisition
projects.

*Schedule

*06 months: Initial specification of the
problem-solving architecture.
12 months: Initial development of its component
panalogies, goals and
critics.
18 months: Integration of components on test
problems.
24 months: Integration with other sources of
knowledge like Cyc, Open Mind,
and Web.
30 months: Initial interface to help people solve
problems using Google,
Yahoo, MSN, etc.
36 months: Prototype ready to show sponsors for
further development of
project

*Organizational Structure

*Our core focus is developing a computation-based
architecture upon which
application-level research and development can be
built. The cognitive
architecture will be developed by Panalogy
Foundation, a non-profit
organization being created mainly by MIT-based
researchers to support
human-level commonsense thinking, based on the
theories of mind pioneered by
Marvin Minsky, as well as the principles expressed
in Dr. Push Singh's PhD
thesis.

Panalogy will utilize a lightweight distributed
management structure. The
team will consist of part-time project and office
managers, two full-time
cognitive architects, three full-time
programmers/computer scientists, and
one and one half FTEs of consultants and visiting
scientists. An �inverse�
travel budget will be used to bring the best minds
to the project on a
quarterly basis.

The decision to build out from a non-profit
organization will simplify our
relationship with MIT, while reducing any pressure
to stray prematurely from
our basic research goals. The Foundation will be
�off-campus� to avoid
hiring restrictions at MIT, but it can take full
advantage of its proximity,
to provide support for graduate students,
undergraduate researchers, and
faculty consultation. Transitions from research to
consumer products will be
done by establishing for-profit spin-off vehicles.

It is expected that by end of Year Three, Panalogy
will be self-sustaining,
having developed ample funding from its spin-off
operations.

*Competition?

*The mainstream of AI community is divided into two
major