Download Powerpoint - WordPress.com

Technical Goals
for the BICA Community
Mark R. Waser
mailto:[email protected]
http://BecomingGaia.wordpress.com
Goal - 2008
creating a computational equivalent of the
natural mind in its higher cognitive abilities
Specific topics include
•cognitive architectures inspired by the brain,
•constraints borrowed from biology,
•human-like learning and self-sustained cognitive growth,
•self-regulated learning assistance,
•natural language acquisition,
•emotional and social intelligence,
•metrics and
•a roadmap to solving the challenge.
Goal - 2009
creating a real-life computational equivalent
of the human mind
Specific topics include
•Bridging the gap between AI and biology: robustness, flexibility, integrity
•BICA models of learning: bootstrapped, self-regulated (SRL), meta-learning
•Scalability, limitations and ‘critical mass’ of human-like learning
•Biological constraints vital for learning
•Physical support of conscious experience
•Formal theory of cognitive architectures
•Emotional feelings and values in artifacts
•Measuring minds of machines and humans
Subgoals – 2008 & 2009
Part I.
cognitive architectures inspired by the brain
Formal theory of cognitive architectures
constraints borrowed from biology
Biological constraints vital for learning
human-like learning and self-sustained cognitive growth
self-regulated learning assistance
BICA models of learning: bootstrapped, self-regulated (SRL),
meta-learning
natural language acquisition
(NONE)
Subgoals – 2008 & 2009
Part II.
emotional and social intelligence
Physical support of conscious experience
Emotional feelings and values in artifacts
metrics
Measuring minds of machines and humans
a roadmap to solving the challenge
Bridging the gap between AI & biology: robustness, flexibility, integrity
Scalability, limitations and ‘critical mass’ of human-like learning
Goal - 2010
creating a real-life computational equivalent
of the human mind
four schools of thought:
(1) computational neuroscience, that tries to understand how the brain
works in terms of connectionist models;
(2) cognitive modeling, pursuing higher-level computational
description of human cognition;
(3) human-level artificial intelligence, aiming at generally intelligent
artifacts that can replace humans at work; and
(4) human-like learners: artificial minds that can be understood by
humans intuitively, that can learn like humans, from humans and
for human needs.
Subgoals – 2008-2010
Part I.
computational neuroscience (connectionist modeling)
cognitive architectures (low-level)
biological constraints (low-level) ???
cognitive modeling
cognitive architectures (high-level)
biological constraints (high-level)
human-level artificial intelligence (that can replace humans at work)
human-like learners/human-like artificial minds
human-like learning
natural language acquisition
emotional and social intelligence
Subgoals – 2008-2010
Part II.
metrics
a roadmap to solving the challenge
2008 - creating a computational equivalent of the
natural mind in its higher cognitive abilities
(human-level AGI)
2009-2010 - creating a real-life computational
equivalent of the human mind
(human-like AGI+)
safety!
Toward a Comparative Repository of Cognitive
Architectures, Models, Tasks and Data
•Introduction (discussion panel agenda - by Christian Lebiere)
First Step: Comparative Table of Cognitive Architectures
•Current comparative table: HTML | XLS | PDF
•Old comparative table - from Pew & Mavor, 1998
Complementary Frameworks for Comparison (4)
Related Sites (3)
What Is Our Goal?
– OR –
What Do We Want To Be?
A united working community dedicated to a
specific common goal (2008 or 2010?)
A social networking community dedicated to
sharing/collecting information and recruiting
Thursday, November 5, 4:00 pm – 5:45
pm, Westin Arlington Gateway Hotel
1
AAAI 2009 Fall Symposium Series
Arlington, Virginia – November 5‐7, 2009
Panel Discussion:
Comparative Repository of
Architectures, Models, Tasks and Data
Chair: Christian Lebiere
Objective
To identify the necessary means to achieve greater
rates of convergence and incremental progress in
cognitive modeling through the use of a shared
repository of computational cognitive architectures,
models, tasks and data.
Why do we need a repository?
1. To facilitate direct comparison of different
architectures.
2. To provide a centralized resource, that modelers,
students, and teachers can access when they want to
start a modeling research project.
3. To have an immediate and organized way to access
an overview of relevant information.
4. To enable the reuse of models.
5. To encourage the development of modeling tools and
standards.
How are we going to spread it?
How are we going to make it work?
Uploading tasks and code as currently existing is not enough.
The following issues should be considered.
1. A standard API between cognitive architectures and task
simulation environments is needed to assure portability
across tasks and models.
2. Models need to be updated and kept current.
3. Infrastructure funding should be provided by some source,
4. Before proceeding with the implementation, some informal
polls or surveys should be taken to study the modelers’ habits
and needs