CHAPTER 14 Intelligent Systems Development

n Overview of the expert system development process n Performed differently depending on the –Nature of the system being constructed –Development strategy –Support tools

Prototyping: ES Development Life Cycle n 6 phases n Nonlinear process

Phases I. Project Initialization II. Systems Analysis and Design III. Rapid Prototyping IV. System Development V. Implementation VI. Postimplementation

Phase I: Project Initialization (Table 14.1) n Problem Definition n Need Assessment n Evaluation of Alternative Solutions n Verification of an Expert Systems Approach n Feasibility Study n Cost-benefit Analysis n Consideration of Managerial Issues n Organization of the Development Team

Problem Definition and Need Assessment n Write a clear statement and provide as much supporting information as possible n Conduct a formal needs assessment to understand the problem

Evaluation of Alternative Solutions n Using experts n Education and training n Packaged knowledge n Conventional software n Buying knowledge on the Internet

Verification of an Expert Systems Approach Framework to determine problem fit with an ES (Waterman [1985]) 1. Requirements for ES Development 2. Justification for ES Development 3. Appropriateness of ES

1. Requirements for ES Development (all necessary) 1. Task does not require common sense 2. Task requires only cognitive, not physical, skills 3. At least one genuine expert, willing to cooperate 4. Experts involved can articulate their problem- solving methods 5. Experts involved can agree on the knowledge and the solution approach (continued)

6. Task is not too difficult 7. Task well understood and defined clearly 8. Task definition fairly stable 9. Conventional (algorithmic) computer solution techniques not satisfactory 10. Incorrect or nonoptimal results generated by the ES can be tolerated 11. Data and test cases are available 12. Task's vocabulary has no more than a couple of hundred concepts

2. Justification for ES Development (Need at least one) 1. Solution to the problem has a high payoff 2. ES can preserve scarce human expertise, so it will not be lost 3. Expertise is needed in many locations 4. Expertise is needed in hostile or hazardous environments 5. The expertise improves performance and/or quality 6. System can be used for training 7. ES solution can be derived faster than a human 8. ES is more consistent and/or accurate than a human Benefits Must Exceed Costs

3. Appropriateness of the ES (Consider 3 Factors) 1. Nature of the problem: Symbolic structure and heuristics 2. Complexity of the task: Neither too easy nor too difficult for a human expert 3. Scope of the problem: Manageable size and practical value Problem Selection is a Critical Factor

Feasibility Study (Table 14.2) n Economic (financial): Should we build it? n Technical: Can we build it? n Organizational: If we build it, will they come?

Cost-Benefit Analysis n Determines project viability n Often very complicated n Difficult to predict costs and benefits –many are qualitative n Expert systems evolve constantly

Cost-Benefit Analysis - Complicating Factors n Getting a handle on development costs –Consider (and revise) system scope –Estimate time requirements n Evaluating benefits –Some intangible –Hard to relate specifically to the ES –Benefits result over time –Not easy to assess quantity and quality –Multiplicity of consequences hard to evaluate (goodwill, inconvenience, waiting time and pain) n Key: identify appropriate benefits

n When to justify (very often!) –At the end of Phase I –At the end of Phase II –After the initial prototype is completed –Once the full prototype is in operation –Once field testing is completed (prior to deployment) –Periodically after the system is in operation (e.g., every six or twelve months) n Reality checks –How to justify?

Consideration of Managerial Issues n Selling the project n Identifying a champion n Level of top management support n End user involvement, support, and training n Availability of financing n Availability of other resources n Legal and other potential constraints

Organizing Development Team n Team varies with the phases –Typical development team –Expert –Knowledge engineer –IS person

Team May Also Include n Vendor(s) n User(s) n System integrator(s) n Cooperation and communication required!!! n Possible functions and roles in an ES team (Table 14.4)

Important Players n Project champion n Project leader

Phase II: Systems Analysis and Design n Conceptual design and plan n Development Strategy n Knowledge sources n Computing resources

Conceptual Design n General Idea of the System n General capabilities of the system n Interfaces with other CBIS n Areas of risk n Required resources n Anticipated cash flow n Composition of the team n Other information for detailed design later n Determine the development strategy after design is complete

Development Strategy and Methodology n In-house development n Outsourcing n Blended approach

In-house Development n End-user computing n Centralized computing n End-user computing with centralized control n High-technology islands n Information centers

Outsourcing n Hire a consulting firm n Become a test site n Partner with a university n Join an industry consortium n Buy into an AI firm

Blended Approach Mix both n In-house n Outsourcing

Selecting an Expert n Experts n Expertise is based on experience and can be expressed by heuristics n Selection Issues –Who selects the expert(s)? –How to identify an expert –What to do if several experts are needed –How to motivate the expert to cooperate

Software Classification: Technology Levels (Figure 14.2) Expert System Applications (Specific ES) Shells Hybrid Systems Support Tools, Facilities, and Construction Aids Programming Languages

Software Classification: Technology Levels (Figure 14.2) n Specific expert systems n Shells n Support tools n Hybrid Systems (environments) n Programming languages n NEW –Object-oriented Programming (OOP) –Internet/Web/Intranet-based Tools

Building Expert Systems with Tools 1. The builder employs the tool's development engine to load the knowledge base 2. The knowledge base is tested on sample problems using the inference engine 3. The process is repeated until the system is operational 4. The development engine is removed and the specific expert system is ready for its users (using a separate runtime (executable) component of the tool)

Shells and Environments n Expert systems components 1. Knowledge acquisition subsystems 2. Inference engine 3. Explanation facility 4. Interface subsystem 5. Knowledge base management facility 6. Knowledge base n Shell: Components 1-5 (Figure 14.3)

Shell Concept for Building Expert Systems (Figure 14.3) Knowledge base (rules) Consultation manager Knowledge base editor and debugger Inference engine Knowledge base management facilities Explanation program Shell Tip of the iceberg

Rule-Based Shells n Exsys n InstantTea n XpertRule KBS n G2 n Guru n K-Vision n CLIPS n JESS

Domain-Specific Tools Designed to be used only in the development of a specific area n Diagnostic systems n Shells for configuration n Shells for financial applications n Shells for scheduling

Development Environments n Support several different knowledge representations and inference methods n Examples –ART-IM –Level5 Object –KAPPA PC

Software Selection n Complex problem –Frequent technology changes –Many criteria n First check out: –“PC AI Buyer’s Guide” –“Expert Systems Resource Guide” in AI Expert –Newsgroup FAQs on ES Major Issues in Selecting ES Development Software (Table 14.7)

Shells vs. Languages n How to select which n Try the Analytic Hierarchy Process (Chapter 5)

Hardware Support Software Choice Usually Depends on the Hardware n PCs n Unix workstations n Web servers n AI workstations n Mainframes

Phase III: Rapid Prototyping and a Demonstration Prototype n Build a small prototype n Test, improve, expand n Demonstrate and analyze feasibility n Complete design

Rapid Prototyping n Crucial to ES development n Small-scale system n Includes knowledge representation n Small number of rules n For proof of concept n Rapid prototyping process (Figure 14.4)

Phase IV: System Development n Develop the knowledge base n Define the potential solutions n Define the input facts n Develop an outline n Draw a decision tree n Create a knowledge map (matrix) n Create the knowledge base n Test, evaluate, and improve (knowledge base) n Plan for integration

Use a System Development Approach n Continue with prototyping (yes) n Use the structured life cycle approach (rare) n Do both (rare)

Develop the Knowledge Base Acquire and Represent Knowledge Appropriately n Define potential solutions n Define input facts n Develop outline n Draw decision tree n Map matrix n Create knowledge base

Test, Validate and Verify, and Improve n Test and evaluate the prototype and improved versions of the system –In the lab –In the field –Initially - evaluate in a simulated environment

n Modified Turing Test: Compare ES performance to an accepted criterion (human expert's decisions) n Experimentation n Iterative Process of evaluation: –Refine the ES in the field –Use new cases to expand the knowledge base n Validation - determine whether the right system was built Does the system do what it was meant to do and at an acceptable level of accuracy? n Verification - confirms that the ES has been built correctly according to specifications

Phase V: Implementation n Acceptance by users n Installation, demonstration, deployment n Orientation, training n Security n Documentation n Integration, field testing

ES Implementation Issues n Acceptance by the user n Installation approaches n Demonstration n Mode of deployment n Orientation and training n Security n Documentation n Integration and field testing

Phase VI: Postimplementation n Operations n Expansion: maintenance and upgrades n Includes periodic evaluation

n The environment changes n More complex situations arise n Additional subsystems can be added (e.g., LMS) Upgrading (Expansion)

n Maintenance costs versus benefits n Is the knowledge up-to-date? n Is the system accessible to all users? n Is user acceptance increasing? (feedback) Evaluation (Periodically)

The Future of Expert Systems Development Processes Expect Advances In n Flexible toolkit capabilities, including inferencing hybrids n Improved languages and development systems n Better front ends to help the expert provide knowledge n Improved GUIs via Windows-based environments n Further use of intelligent agents in toolkits n Better ways to handle multiple knowledge representations

n Use of intelligent agents to assist developers n Use of blackboard architectures and intelligent agents in ES n Advances in the object-oriented approach, for representing knowledge and ES programming n Improved and customized CASE tools to manage ES development n Increased hypermedia use and development (Web) n Automated machine learning of databases and text