Transcript Ethical and Legal issues

Commercial tools
•
Size of system:
– small systems
• 400 rules
• single user, PC based
– larger systems
• narrow, problem-type specific or hybrid shells using many
problem-solving paradigms (rules, induction, NN, GA, ...)
• 1000+ rules
• can be multi-user (esp. WWW)
•
Type of system
– expert system consultation
– decision-support system: high-end analysis of data (data
mining)
B. Ross Cosc 4f79
1
Commercial Tools
p.93-95
B. Ross Cosc 4f79
2
Commercial Tools
B. Ross Cosc 4f79
3
Commercial Tools
B. Ross Cosc 4f79
4
Commercial Tools
• Evaluating knowledge engineering tools
- Consultation paradigm:
diagnosis? planning? configuration?...
- AI paradigms: Representation, inference, control, uncertainty, neural nets
- Implementation: Lisp, Prolog, C based, speed, transportable, interfacing,
compiled code, WWW
- User interface: explanation, graphics, windowing, knowledge engineering,
- Applications: what applications have been implemented with the system
- Support: documentation, training, support services
Contemporary shells:
- multi-paradigm, PC-based (and up), integrate with std languages and
applications (eg. databases)
B. Ross Cosc 4f79
5
M.1
• by Tecknowledge Inc.
• prototyping tool
• can handle small systems (100-200 rule systems)
• implemented in Prolog
• EMYCIN strategy: backchaining with uncertainty, unknown values
• supports "variable" rules: rule macro's
• window, menu interface
• M4: latest version ($1000)
- embeds expert system code into applications
- VB version embeds into Visual Basic ($199)
B. Ross Cosc 4f79
6
M.1
p. 107
B. Ross Cosc 4f79
7
Flex
• Bundled expert system toolkit with Quintus Prolog, Macprolog, others
• multiple paradigms: forward and backward chaining, frames
• forward-chaining rule selection is flexible, and permits builtin
or user-defined algorithms to be used
• Can use Prolog's inference engine: directly call prolog code
• "data-driven" programming: frame demons
• procedural control
• Macintosh interface primitives
• fairly rudimentary explanation: must indicate explicitly which rules
can be in explanation -- and explanation is text
(hybrid systems mean that explanation is a problem)
B. Ross Cosc 4f79
8
Flex
B. Ross Cosc 4f79
9
OPS5, OPS83
OPS5:
•Carnegie-Mellon
•Used to implement XCON
•Production-rule, forward-chaining system
•Uses time stamps to fire rules (least-recently used strategy)
•intended for larger expert systems
•Implemented in Lisp
•interface permits a programming environment
OPS83:
•successor to OPS5
•more generalized rule format, control
•embeddable in C
B. Ross Cosc 4f79
10
OPS5
p.118
B. Ross Cosc 4f79
11
RuleMaster
• by Radian Corporation (Texas) and Intelligent Terminals (Scotland)
• inductive inference system, intended for small to moderate systems
• modular approach: expert system components encoded in modules
(procedures), which hierarchically call one another
• Can create rules usig ID3 algorithm, or encode if-then rules directly
• runs on unix or PC-DOS,
• down-compiles into C or FORTRAN if desired
• spreadsheet interface for creating example sets
• example systems: Willard (severe storm forecasting), OIL (oil tank diagnosis)
B. Ross Cosc 4f79
12
ART
• The Automated Reasoning Tool (by Inference Corporation; now owned
by Brightware Inc))
• hybrid tool kit for knowledge system development
• 4 components:
- knowledge language
- compiler (knowledge language --> Lisp)
- applier (inference engine)
- development environment
• Uses a number of inference paradigms, including frames, logical represenations
forward and backward chaining, procedural execution, uncertainty
• systems range from $8000-$150,000
B. Ross Cosc 4f79
13
ART
p.122
B. Ross Cosc 4f79
14
Intellicorp
KEE: Knowledge Engineering Environment by IntelliCorp
• hybrid system, used for number of genetic expert systems
• uses frames, rules, procedures, backward and forward chaining
• $60,000 in 1985 (today’s price ???)
Kappa-PC: PC Windows shell
• object-oriented, rule-based GUI environment
• interface builder
• $995
B. Ross Cosc 4f79
15
KEE
p. 124
B. Ross Cosc 4f79
16
Trends for commercial shells
•
•
•
•
•
•
•
Most run on PC’s or distributed networks of PC’s
WWW is a hot area! Web is now a standard interface for
production expert systems.
Java is becoming an implementation language
Highly interactive development environment are the norm
Most systems include a library of various tools & technologies
– forward & backward rules, rule induction, NN’s, fuzzy, GA’s
The difference between decision support environments and expert
systems is becoming vague
– data mining applications use same tools as expert systems,
although they are applied for often different purposes
– main dogmatic difference: expert system KB needs human
expert, while data mining uses auto techniques on large DB
– both are merging for some problem domains
New AI technologies will find their way into shells
B. Ross Cosc 4f79
17