INTELLIGENT AGENTS

1
INTELLIGENT AGENTS

• IS3301 Session 15
• Prof. Mark Nissen

2
Agenda

• Travel Agents Vignette
• Agent Background Examples
• Third-Party Agent Design Issues
• The Intelligent Mall
• Agent Experiment
• Summary

3
Travel Agents Vignette

• Plan overseas vacation
• How does current process work?
• How would it work with agents?
• How futuristic is this redesign?
• Military applications?

4
Agent Functionalities

• Overcome information overload
• Automate mundane info tasks
• Persistently vigilantly serve master
• Interface with complex systems
• Support/make decisions
• Cognitive prostheses
• Design keys
• Easy to build small, limited KBS
• Link together in cooperative federation

5
Agent Capability Framework
6
Agent Definitions

• Disagreement on agent definition
• Classify by attributes (intel, mobility)
• Classify by intention (transact, represent)
• Useful abstraction (module, object)
• Many diverse agent examples
• Information filtering
• Information retrieval
• Advisory
• Performative

7
Agent Examples

• Information filtering - I search
• User e-mail preferences (Maes, Malone)
• NetNews preferences (Sycara Zeng)
• FAQs search filtering (Whitehead)
• Arbitrary text (Verity)
• Information retrieval - I search
• Compact disks (Krulwich, BargainFinder)
• Computer equipment (uVision)
• Advertising rates (PriceWatch)
• Insurance services (Insurance)

8
Agent Examples

• Information retrieval (cont)
• Web indexing robots (Etzioni Weld)
• Web report writing (Amulet)
• Web publishing (InterAp)
• Assisted browsing (Burke)
• Tech info delivery (Bradshaw)
• Info gathering (Knobloch Ambite)
• Books (BargainBot) http//www.ece.curtin.edu.au/s
  aounb/bargainbot/

9
Agent Examples

• Advisory - I recommendations
• Music recommendations (Maes, Firefly)
• Electronic Concierge (Etzioni Weld)
• Campus visit host (Zeng Sycara)
• Manufacturing plans (Maturana Norrie)
• Strategic planning (Pinson)
• Software project coordination (Johar)
• Interface assistance (Ball)
• Reconnaissance (Bui)
• Portfolio management (Sycara)

10
Agent Examples

• Performative - I behavior
• Agent2agent markets (Chavez Maes)
• Negotiation (Bui)
• Scheduling (Sen, Walsh)
• Cooperative learning (Boy)
• Digital services (Mullen Wellman)
• Software purchasing (Mehra Nissen)

11
SOAR Agents

• SOAR is 20 years old
• Architecture for intelligent behavior
• Used to construct intelligent agents
• TACAIR SOAR - mil application
• What does it do?
• How well does it do it?
• What class of agent is it?
• Other applications of this technology?

12
Third-Party Agent Design Issues
Agents Orientation Design Cooperative C
ompetitive Centralized Predetermined n/a
Distributed Trust-based Negotiation Job
specialization Game-theoretic
13
Third-Party Agent Design Issues

• Inter-agent protocols
• KIF - Declarative - predicate calculus
• KQML - Standard communication types
• IIOP - Net inter-ORB protocol
• Negotiation - multiple bids quotes
• Game-theoretic reasoning - problems?

Unknown/Unfriendly Agent Friendly
Agent Cooperative Competitive
Cooperative 10 -5 Competitive 5 -10
14
Supply Chain Management

• SC management is critical biz competency
• Longstanding focus on profit/cost risk
• Speed flexibility increasingly important
• Trend toward disintermediation (in/external)
• Question role value of intermediaries
• Aggregation, trust, facilitation, matching
• Examples brokers, agents, market makers,
  wholesalers, distributors, purchasing, others
• Value added vs. cost/time of services
• Potential for I-agent (dis)intermediation

15
Agent Vision

• Vision
• For most procurement items (80)
• Any user in the organization
• Can effect own procurements
• Obtain best value
• OOM more quickly and inexpensively
• Without understanding procurement
• Without an intermediary (e.g., Supply)
• Adhere to all laws, rules regs

16
Multi-Agent Approach

• Continue from Nissen Mehra (1998)
• Performative agent federation
• Procurement order-fulfillment processes
• Represent as single integrated process
• Capture formalize process knowledge
• Operational supply chain
• University procurement process
• Hi-tech order fulfillment process
• Agent integration social conformance

17
Supply Chain Process
User Procurement
Vendor
ID rqmts
Advertise
Market survey
PR form
Verify form
Research sources
Issue RFQ
Prep quotes
Analyze quotes
Select source
Fulfill order
Issue order
Send invoice
Receive goods
Use goods
Make payment
Deposit funds
18
B2B Commerce Model
B1 B2 B3 B4
B5
Use,
Buyer
ID
Find
Arrange
Purchase
maintain,
need
source
terms
dispose
Intermediary
Money
Information
Information
Influence
goods
(0
thru
n)
X1 X2 X3

X4

X5

X6
Seller
Arrange
Find
Arrange
Fulfill
Support
to provide
customer
terms
order
customer
S1 S2 S3 S4
S5
Process flow
19
Process Design General Commerce Model
20
Agent-Integrated Process
User I-Mall
Vendor
ID rqmts
Market survey
Launch agents
Launch agents
ltSelect sourcegt
ltFulfill ordergt
Receive use goods
Deposit funds
21
The Intelligent Mall

• Proof-of-concept multi-agent-system
• Mall metaphor intelligent agents
• Agents represent users (engr, mgr, CO)
• Agents represent vendors (FPI, Dell, GSA)
• Share responsibilities with people
• (Dis)intermediate procurement process

22
The Intelligent Mall

• Intelligent agent federation
• Legions of autonomous, persistent agents
• Surpass some human performance
• Socially-conforming behavior
• Maintain agents via inheritance
• Work within SPS environment
• Autonomous, long-lived commerce
• Procurement domain important (thesis)
• NPRDC officer billets market (thesis)
• Intelligent Mall demo

23
Preliminary Results

• Developed proof-of-concept MAS
• Agent (dis)intermediation feasible
• Demonstrated good performance
• Correct behaviors, justifiable decisions
  actions
• Process speed flexibility
• Super-human vigilance, persistence, reliability
• Problems
• Emergent, distributed behaviors are complex
• Trust remains huge impediment (credit cards)
• When to (dis)intermediate via I-agents?

24
Grafcet behavior flows
Objects, methods messages
Virtual mall/supply chain Intelligent
procurement agents
25
Web-based I/O
Suppliers register product data
Users specify requirements
Item, qty, budget, date, etc.
Items, specs, price, delivery, etc.
26
Regulatory environment
Agents locate select suppliers, virtually,
arrange terms
27
Agents complete transactions, return with
goods/contracts
28
Experiment Procurement Test

• Experiment
• Continued agents research
• Humans agents along supply chain
• What do humans/agents do best?
• Design
• 2 blocks acquisition IS students
• Task purchase items of varying complexity
• 4 groups varying degrees of agent support
• Judgment, reasoning, accuracy, conformance
• Data good thesis opportunity!

29
Summary

• Disagreement on agent definition
• Unique set of capabilities
• Many examples
• Military applications
• SOAR addresses mil/sim environment
• Supply chain management is key
• Agent offer excellent potential
• Can exceed human capabilities
• Intelligent Mall POC implementation