3. Engineering
Natural Artificial
-Made of autonomous agents capable
of playing several roles.
- Autonomous actions by agents based
on self interest.
- Structure is a result of evolution and
local adjustments.
- Made of parts custom built for a
specific purpose
- Well defined functionality for each
part.
- Structure is designed into its present
shape.
Living beings behave more like societies than
machines according to the definitions.
Courtesy: Course Slide
5. A hybrid model which
mimics natural engineering.
A related video is available on http://riteshnayak.com/xconf
6. The goal of multiagent systems’ research
is to find methods that allow us to build
complex systems composed of
autonomous agents who, while operating
on local knowledge and possessing only
limited abilities, are nonetheless capable
of enacting the desired global behaviors.
7.
8. What’s different
• Completely autonomous agents that act out
of self interest.
• No constraints on inputs (Open World
Programming)
• Limited knowledge (don’t think bluegene)
• Interaction between autonomous agents
determined based on self interest function.
11. Why is this important
• To simulate events with real life actors.
• To learn about evolutionary characteristics of
a model.
• Model societies, colonies, groups etc and
learn about herd behavior (ex: How do
perfectly rational humans create traffic
deadlocks?)
Humans are individually smart and collectively stupid.
12. Desirable Characteristics of Agents
• High predictability in behavior
• Operability in uncertain conditions.
• Resilient to failures.
• Optimal performance for given situations
• Mathematical tractability to the finest detail.
13. Emergence
• Designing Objective
functions and payoffs in a
way so that local decisions
of agents collectively result
in optimizing a global
objective function.
• You cannot code for
emergence.
• Emergent behaviors are not
always desirable.
18. Features
• NetLogo is a cross-platform multi-agent
programmable modeling environment.
• Built and maintained out of Northwestern
University.
• Uses a variant of LOGO
Lets see a demonstration
20. Termite mound
• A termite mound is work of art
• The temperature inside the mound has to
remain a constant 31 deg.
• Termites are autonomous beings.
They are hard wired to do only one thing!!
A related video is available on http://riteshnayak.com/xconf
21. Modelling
• Autonomous actions by agents based on self-
interest functions like beliefs, desires,
intentions, etc (Rational Choice – or utility
maximization – John Stuart Mills)
• Interaction between Agents can be modeled
as a game, auction (common resources), vote
(master-slave setup) etc.
24. Equilibrium Concepts
The simplest form of Nash equilibrium
is one where each player makes a
rational choice with no belief (or a
least biased belief) about the other
players.
25. MAS Borrows from
• Rational Choice theory
• Game theory
• Stochastic Networks
• Auction theory, negotiations and mechanism
design.
• Chaos theory, complex systems and theory of
emergence.
26. Mechanism Design
• Refers to the design principles behind an
auction or a voting process that can be used
to favor specific outcome
• In an auction, the seller’s choice is to sell at
the highest possible price. How do you get the
agents to quote higher prices.
28. Weather Forecasting
There is a lot of work being done
to model climate and
implications of climate change
etc. A domain that has seen a lot
of action in last decade.
31. Realism in Games
A related video is available on http://riteshnayak.com/xconf
32. The Grand DARPA Challenge
Requires teams to build an autonomous vehicle capable of
driving in traffic, performing complex maneuvers such as
merging, passing, parking and negotiating intersections.
Prize money is $2 million, $1million and $500k respectively
A related video is available on http://riteshnayak.com/xconf
33. Disaster Recovery
• Work done by my classmates at CSTEP.in
• Using technology to shape public policy
• Use SimCity as a base framework for
modelling agents.
A small video of the simulation
34. My Project
• Multi-Agent based simulation of a
Normative/Incentive system for Content
Aggregation on Online Forums
• Main objectives
– To build a system of norms and incentives for
knowledge aggregation on an online forum
– Mechanism design to increase activity on the
forum and also keep the network from saturating
36. LoyalUser
• Posts
Regularly
• Forgiving
• Satisfied
with
answers
• Closes q’s
early
• 1 category
RegularUser
• Less
frequently
than loyal
• Less
forgiving
• Waits for
more
answers
• At least 2
categories
BounceUser
• Participates
rarely
• Reply
sometimes
• Mostly
simulate
redirects
from
search
engines
• Don’t close
questions.
37. Last but one slide
• MAS research is a relatively new field for
computer scientists.
• Lot of applications in many different fields.
Will gain a lot of prominence very soon.
• Skeptics doubt results due to inconsistency.
• Hope you figured the playing God part.
38. References
• Fundamentals of Multiagent Systems - Jos´e M Vidal -
http://jmvidal.cse.sc.edu/papers/mas-20070824.pdf
• Course on MAS at my institute IIIT – Bangalore (course
page http://osl.iiitb.ac.in/wiki/index.php/Multi-Agent_Systems)
• Prof Srinath Srinivasa for all anecdotes/ examples etc.
• Evolution of Co-operation – Robert Axelrod
• C.H. Papadimitriou. Algorithms, Games, and the
Internet. Proc. STOC-2001, ACM Press, 2001. Invited
talk write-
up.(URL:http://www.eecs.harvard.edu/~parkes/cs286r/spring02/papers/stoc01.pdf)
• Thanks to DARPA, Google Image search, wisegeek.com
and Wikipedia for the images.