Intelligent agents and robots

1

• Intelligent agents and robots
• What are agents and bots?
• Characteristics of agents and bots
• II. How do they work?
• The structure of agents and bots
• III. Examples of agents and bots
• Resource discovery tools
• Agent mediated ecommerce
• Turing tools
• IV. The future of agents

2

• Searching for information
• What are agents and bots?
• An agent is a software tool for digging through
  data
• The software becomes an extension of the user,
  performing tasks on the user's behalf
• There are many agents and bots on the web
• Search engines use robots to crawl the web and
  compile the lists of URLs that are the heart of
  every search engine
• Shopping bots compile enormous databases of
  products sold at online stores
• Botspot. (2002). What is a bot?
  http//www.botspot.com/

3
An agent is software that Is autonomous and can
be personalized Is intelligent and can
learn Can perceive user actions and act though
the user interface to engage in
discourse Presents a human-like appearance and
engenders risk and trust Is context- and
domain-sensitive Has specialized knowledge Is
an assistant and not a toolbox
4
Agents and bots can be used for Information
search and retrieval Notifier agents,
personalized newspapers Organizers Ecommerce
agents Context-sensitive help Collaborative
agents Filtering agents Reconnaissance
agents Turing-test agents
5
Autonomy An agent should have a measure of
autonomy from its user It can pursue an agenda
independently of its user This requires periodic
action, spontaneous execution, and
initiative The agent must be able to take
preemptive or independent actions that will
eventually benefit the user This occurs without
the users direct intervention
6
Personalization Agents should help people to do
some task better Because of variability in the
way we accomplish tasks, agents must be able to
adapt to our different styles This requires
learning The agent should not have to be
programmed to handle the same task in different
ways (observation) Once learned, this
information should be retained (memory) This
allows persistence of interest You shouldnt
have to restate your interests each time
7
Discourse There should be some communicative
interaction with the agent This ensures that the
agent is operating according to our agenda It
will accomplish the task as we want it done The
interaction results in a contract This two way
exchange of information establishes intentions
and abilities It may be a single
conversation It may be a high level discourse
with the user and agent repeatedly interact with
both parties remembering previous interactions
8
Risk and trust We delegate tasks to our agent We
have to be trust that the agent will carry out
the task according to our specifications This
does not occur without risk The agent may do
what we want but because of its autonomy, it
may not We balance the risk that the agent will
do something wrong with the trust that it will do
it right Our internal mental model of what the
agent will do determines how much we trust
it Our domain of interest determines how much a
mistake will cost us
9
Domain Where we use the agent is the domain of
interest This is crucial to calculate risk and
trust In a game or a social pursuit, agent
failures carry low risk and we tend to have
greater trust For stock trading or auctions,
agent failures are costly Graceful
degradation When communication does not work,
agents should try to do as much as they can
before ending the exchange Communications
mismatch the two parties do not necessarily
communicate well, and may not realize it Domain
mismatchone or both parties are out of their
element, and may not realize it
10

• Intelligent agents and robots
• What are agents and bots?
• Characteristics of agents and bots
• II. How do they work?
• The structure of agents and bots
• III. Examples of agents and bots
• Resource discovery tools
• Agent mediated ecommerce
• Turing tools
• IV. The future of agents

11
Types of agents and bots Integration Information
integration, knowledge sharing Coordination Coop
erative problem-solving, multi-agent systems
Mobility Mobile agent/object solutions
Assistants Personal assistants, softbots, data
mining Believable agents A-life, simulation
12
There are static and mobile agents A static
agent does all of its work in one place An
email client that downloads your mail to your
PC A mobile agent can operate on a network
It is sent out on a mission, finds information
and reports back It traverses the
network, executing tasks at each node,
interacting with other agents that it
meets Mobile agents are useful in data mining
because they can find patterns in large data
sets Data mining is iterative and labor
intensive - agents save time, refining
the search over time and making decisions
based on past experiences
13
Mobile agents are based on a model combining
behavior, state, and location Agents inherit a
subset of behaviors from the model These
define the means by which they move through
the network Models also define a method of
interagent messaging The model defines a set of
events of interest to the agent Arriving at a
new location is an important event in the life
of an agent This will entail the invocation of
one or more of its behaviors Sommers, B.
(1999). Agents Not just for Bond anymore
http//www.javaworld.com/javaworld/jw-04-1997/jw-0
4-agents.html
14
This is a typical set of events used in a model
Creation This is a constructor event It
brings the agent to life A handler for this
event initializes the agents state and prepare
it for further instructions Disposal This is a
destructor event A handler for this event
frees the resources the agent is using and
prepare the agent for burial This is invoked
when the agent has completed its task
15
Dispatch Signals the agent to prepare for
departure to a new location This is generated
by the agent itself with a request to
migrate It can be triggered by another agent
asking this agent to move Arrival Signals
that the agent has arrived at its new location
and is beginning to perform its task
Communication Notifies the agent to handle
messages incoming from other agents (
interagent correspondence)
16
An example Mail carrier agent lifecycle
1. Post Office spawns an army of mail carriers
2. Each migrates to first destination in
itinerary 3. It interacts with agents at first
stop, delivers the mail, and continues to the
next stop 4. It returns to Post Office
Sommers, B. (1999). Agents Not just for Bond
anymore http//www.javaworld.com/javaworld/jw-04-1
997/jw-04-agents.html
17
Classes of agent applications User passivity/data
timeliness Applications demand an immediate
reaction to incoming real-time data streams The
agent is a digital proxy for the user,
interacting with data on the user's
behalf Multi-staged/multi-processed calculations
Calculations are broken into discrete units
Each unit is assigned to an agent, which is
dispatched to an agent farm where the work is
performed Upon completion, each agent returns
and the results are aggregated and summarized
18
Untrusted collaborators Mobile agents
collaborate by meeting in neutral turf They do
this with well-defined interfaces and are
protected from intrusion or inspection by other
agents by the agent host Low-reliability/partiall
y-disconnected networks Agents move executable
content to a data source rather than repeatedly
attempting network connections to the data
source An example is a network where users rely
on laptops networked via dial-up connections
Prior to disconnecting, agents are sent to a
server to perform offline calculations
Upon reconnecting, the agents return to the
laptop
19
How do agents learn? If the agent can take
instructions We teach them by telling them what
we want them to do They watch us to learn what
it is that we do Other ways agents can
learn Neural networks Statistical methods Data
mining
20

• Intelligent agents and robots
• What are agents and bots?
• Characteristics of agents and bots
• II. How do they work?
• The structure of agents and bots
• III. Examples of agents and bots
• Resource discovery tools
• Agent mediated ecommerce
• Turing tools
• IV. The future of agents

21
Types of bots Chat Bots News Bots Commerce
Bots Newsgroup Bots Data Mining Bots Search
Bots E-Mail Bots Shopping Bots Fun
Bots Software Bots Game Bots Stock
Bots Government Bots Surveillance Bots
Knowledge Bots Update Bots
22
Resource discovery agents ACORN (Agent-Based
Community Oriented Retrieval Network)
http//ai.iit.nrc.ca/ll.public/acorn.html An
architecture for the sharing, search, and
provision of information across networks
Several agents work together to ensure that
their users get the information that they are
interested in They can act when user sets a
task This is a traditional search task They
can also act with prior instructions Agents
perform continual community-based browsing
23
BFS Spider http//ai.bpa.arizona.edu/mramsey/S
PIDER The Best-First Search (BFS) bot/spider
crawls the Web in search of pages which interest
you You tell it what type of homepages you are
interested in by giving it URLs to start from It
will crawls the web by following these links and
links on these pages It reports back to you on
homepages of interest At each page the spider
pauses and compares it with the original
pages Those pages which are most similar to
the original pages get a higher ranking
24
Letezia is an interface and reconnaissance agent
for web browsing It acts as a scout It watches
what you look at , records URLs, and infers
your preferences When it understands the types
of pages that you search for, it looks through
the neighborhood for similar pages The longer
it has to examine your browsing behavior, the
better it can match your preferences Your
positive and negative feedback teaches through
explanation The interface is a web page with
links
25
Search engines are examples of agent
technology They act in response to user requests
and have considerable autonomy You send them into
action but cannot use the interface while they
are working You must state your request
explicitly While you are using the browsing
interface, the search process is idle
26
Agent mediated commerce NativeMinds
http//an1-sj.nativeminds.com/default.html
NativeMinds features virtual representatives
(vReps) created using the NeuroServer
Suite These bots can answer any questions you
have about NativeMinds and its virtual
representative technology They use facial
expressions that match their responses vReps
such as Nicole provide a best-fit response in
conversational language Nicole will even
remember your name each time you visit the site
27
mySimon http//www.mysimon.com/index.jhtml
This agent does comparison shopping across
the web mySimon uses VLA (Virtual Learning Agent)
technology The agent imitates human navigational
behavior mySimon's staff of shoppers surfs the
Web and interacts with the VLA system The
system translates human navigation behavior into
MySimon's proprietary programming language
In this way, it can be taught to shop at
ecommerce sites in hundreds of product
categories Simon shops in real time, so he
always finds the right products, at the right
place, at the best price
28
An agent called Julia, written by Loren Mauldin
is an example of a Turing agent
http//fuzine.mt.cs.cmu.edu/mlm/julia.html It is
in a class of agents called Maas-Neotek Julia
is a client bot TinyMUD robot and can run on
TinyMUDs, MOOs, or MUSHes She connects as any
human player on the mud would, via a telnet
connection She does not run on the mud server
itself She is written in C and runs on a
workstation in Pittsburgh In 1993 Julia ran
on DruidMUCK and elsewhere on the Internet
29
You interact with Julia as you would with any
person in the MUD gtpage julia You sent
your summons to Julia. Julia pages from
Stevi's Kitchen I'm on my way to you,
Lenny. Julia is briefly visible through
the mist. Julia says, I was called here by
Lenny. You say, julia? Julia
says, Yes? You say, julia?
Julia nods to Lenny. . . . Julia whispers,
Excuse me, Xerxes is paging me from Connie's
place. Julia walks south to the airship
landing field. Julia has left.
30

• Intelligent agents and robots
• What are agents and bots?
• Characteristics of agents and bots
• II. How do they work?
• The structure of agents and bots
• III. Examples of agents and bots
• Resource discovery tools
• Agent mediated ecommerce
• Turing tools
• IV. The future of agents

31
IV. The future of agents Q Are there any
problems that you see agents, and agents alone,
solving in the next few years? A Information
overload Users are increasingly dealing with
vast amounts of information that is unstructured
and very dynamic In order to keep track of
everything, and in order to find the information
relevant to THEM, they will have to use software
that knows their interests and can act on their
behalf Pattie Maes IC Online Virtual Roundtable
The Future of Software Agents http//computer.org/
internet/v1n4/round1.htm
32
Q What other technologies do you think will have
the greatest impact on the directions of agent
technology? A Learning and planning
technologies If you consider an agent to be a
sophisticated surrogate or advisor, theres
enormous potential to be had from both learning
and planning Learning has obvious applications
to agent creation and refinement Planning
technology, which addresses the automated
creation of sequences of activity to satisfy
goals, will have an equally significant role to
play in future agents Jeff Rosenschein
33
A Agents are components that execute in a
distributed heterogeneous and unstructured
environment So, the most important relevant
technologies are software component and
distributed computing technologies, both of
which are progressing rapidly Mani Chandy A If
we consider mobile agents from a performance
perspective, operating systems that would support
process launch/migration/shut-down quickly, and
network protocols that can be used for signalling
and efficient discovery of mobile objects From
an intelligent agent perspective, any technology
that deals with common sense understanding and
learning in a practical way Sankar
Virdhagriswaran
34
A Human-computer interaction design Agents will
not be accepted unless users feel they can
trust them We have to learn how we can build
agents that users can understand and
control Pattie Maes A The most needed
technology is a revolutionizing agent user
interface that makes agents (from different
providers) as easy and fun to use as Web browsers
make the Web fun to surf Secondly, we need an
agent interaction language that allows (1) the
endures to communicate with agents, and (2)
agents to communicate with each other in a
uniform manner Danny Lange