Learning from Live Simulations
Live in-class simulations can be a great way for students
to experientially learn systems concepts like non-linearity, iteration, self-organization
and emergence. They are also a useful way for the instructor to experiment
and gather research data regarding “real” multi-agent systems in action.
Setups for three simulations and a corresponding software model will be discussed,
with a focus on key principles and major learnings for this author based
on dozens of simulations. Briefly overviewed are a:
50-minute simulation
2-hour simulation
8 week (1 hr/week) simulation
Discussed are principles such as:
The power of simple rules – with some
thoughts on good design
How agents (students) respond when
there is no apparent purpose to an activity
The relationship between instructor
authority and student behavior
The importance of minimal (but optimized)
“interference” from outside the simulation (particularly by the instructor)
The role of paradox and double-binds
as catalysts for group emergence (self-organization of common purpose)
Some useful things to say, and not
say, just prior to a simulation
Methods for debriefing; particularly,
documenting process and avoiding unnecessary judgments
A software model will also be demonstrated that allows
students to simulate group experiences on the computer. The software, a “virtual
group” model, allows users to specify agents and “typical” or “fuzzy” social/cultural
rules and then set the group in motion. As the group settles to a (strange)
attractor state, the user can view the mental models agents build, such as
which attributes and behaviors lead to positive and negative feedback from
other agents. The virtual agents can also employ meta-strategies such as
conformity and rebellion, and evaluate other agents’ behavior.
Example instructions to run the simulations and handouts
that have been developed to help students connect systems concepts with simulations
and software models will be provided; also provided is a link to download
the software.
Raising Truman – Working with Advanced Robotic Agents
The author shares his experiences working with an open-architecture
robot, the ER1, that has been become available commercially only in the last
year. With sophisticated vision, speech and movement capabilities, the ER1
is easy to program using a graphical laptop interface and, most excitingly,
is easy to link to one’s own software. Using a robot, or a small group of
robots, allows one to explore multi-agent systems, self-aware agents, and
similar ideas in a more “real world” away than relying upon a computer model
only. The presentation explores four areas:
The ER1 itself, as-is out of the box,
as a physical “thing”
Behaviors that can be programmed into
the robot, such as learning to identify flashcards and faces, or navigating
a room
What happens when a conversation AI
(a sophisticated chatbot) is linked to the robot – for example, when the
robot is presented with a series of flashcards showing math operations, the
robot is able to calculate and speak the results; or the robot can solicit
information like someone’s name, age of interests, and remember and use that
information for later in conversation.
The author’s experiences outside the
lab environment, including its experience lecturing a college class.
The robot itself, named Truman, will be a part of the
presentation and present information about himself (itself).
The possibility (and possible experiences by conference
time) of working with several robots will also be discussed.
Dario Nardi
University of California in Los Angeles
Program in Computing
www.darionardi.com
dnardi@math.ucla.edu