Download Making Expert Systems and Robots Smarter: presentation in Power

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
High Level Logic
Concept Presentation Suggesting
Higher Level Logic for
Expert Systems
Copyright © 2005 Roger F. Gay
[email protected]
HLL
August, 2005 Roger F. Gay
Formulation of the ideas
in this presentation began
in 1986.
The World is Catching Up!
HLL
August, 2005 Roger F. Gay
Past Proposals
SBIR proposals for development of an
HLL demonstration were submitted to
NASA, DARPA, and the NSF ~ 1990.
Funding was not approved.
HLL
August, 2005 Roger F. Gay
Inspiration
The proposals may have inspired
DARPA Project Genoa, initially known
as Collaborative Crisis Understanding
and Mitigation.
I did not participate in Genoa. I do not
believe My HLL has been developed.
HLL
August, 2005 Roger F. Gay
Evolutionary Pressure
HLL
August, 2005 Roger F. Gay
In 1986
Application oriented researchers
became disenchanted with the
limitations of rule-based technology,
even with meta-knowledge, objects,
frames, and Lisp machines.
HLL
August, 2005 Roger F. Gay
EMYCIN Generation
• Narrowly focused domains
• Very difficult to integrate
• No interprocess communication
(information exchange, multiple experts ...)
• Rules! So what?
HLL
August, 2005 Roger F. Gay
Practical AI
Artificial intelligence gets real. By Daniel Lyons. Forbes
Global (November 30, 1998) ."By contrast, Feigenbaum
succeeded by thinking small. Unlike his rivals, he
didn't set out to recreate all of human intelligence in a
computer. His idea was to take a particular expert -- a
chemist, an engineer, a pulmonary specialist -- and
figure out how that person solved a single narrow
problem. Then he encoded that person's problemsolving method into a set of rules that could be stored in
a computer."
HLL
August, 2005 Roger F. Gay
AI Evolution
Develop unconventional software
applications that require symbolic
representation and reasoning that
humans currently do better than
machines.
Separate general structure and logic
from domain specific knowledge.
HLL
August, 2005 Roger F. Gay
Modern Technolgy
• Internet
• XML, SOAP, etc.
• Rule engines and OO programs
• Intelligent Agents
HLL
August, 2005 Roger F. Gay
High Level Logic for Expert Systems
Think Just a Little Higher
Think Just a Little Bigger
The stage is set for a revolution.
HLL
August, 2005 Roger F. Gay
Organizational Context
Intelligence is all about acting in
context; adapting to an environment,
meeting basic needs, responding to
threats, living with other humans,
participating in organized activities.
Intelligence is a product of the context
that shaped it.
HLL
August, 2005 Roger F. Gay
Executive Role
• Define Goals
• Provide Strategy
• Assign Authority & Responsibility
• Assess Overall Results
• HL Corporate Relationships
HLL
August, 2005 Roger F. Gay
Manager’s HLL
• Collect Information
• Define a Problem or Opportunity
• Assign Tasks
• Coordinate Activities
• Finalize / Approve Decisions
HLL
August, 2005 Roger F. Gay
Expert’s HLL
• Refine Problem Definition
• Assemble Detailed Information
• Perform Analysis
• Identify Alternatives
• Recommend Solutions
HLL
August, 2005 Roger F. Gay
Drones
Perform specialized tasks:
Examples:
• Industrial Robots
• Mission Programmable Weapons
HLL
August, 2005 Roger F. Gay
Long-Term Goals
Large-scale, fully automated,
real-time, “corporate” systems.
Searchable, sharable,
“cooperating” components from a
variety of sources.
HLL
August, 2005 Roger F. Gay
Initial Expectations
• Medium Scope (An Application)
• Highly Integrated
• Internet Communications
• Domain Specific Language (XML)
• Focus on Experts
HLL
August, 2005 Roger F. Gay
Why Focus on Experts?
• Executives Need Corporate
Capabilities
• Managers Need Subordinates
• Drones “aren’t paid to think.”
• Experts can collaborate
HLL
August, 2005 Roger F. Gay
Component Technology
1. Define the Problem
2. Collect Information
3. Perform Analysis
4. Identify Alternative Solutions
5. Recommend a Solution
HLL
August, 2005 Roger F. Gay
Recursive
Slightly Generic Example:
• Define a problem
• Collect Information
– Which analysis should be used?
– What information is needed?
• Perform Analysis
– Interpreting input?
– How should the output be presented?
HLL
August, 2005 Roger F. Gay
A Simple Start
HLL
August, 2005 Roger F. Gay
Define a Problem
Identify the type of problem from the
choice of subject experts currently
available using a rule-based expert
system.
HLL
August, 2005 Roger F. Gay
Define a Problem: Examples
General consultant on soil-water
decides whether there is a drainage,
irrigation, or nutrient problem.
Ship board defense system consults
with aircraft identification sub-system
to decide whether to choose com.
protocol or perform threat assessment.
HLL
August, 2005 Roger F. Gay
Collect Information
Suggest stored data and data
generated by system components
are exchanged in XML.
To visually demonstrate this step,
an example might include user
input using an NL Menu system.
HLL
August, 2005 Roger F. Gay
Collect Information:
Examples
Drainage Problem: Need detailed
information on soil type, slopes,
and water volume.
Threat Assessment: Aircraft type,
flight vector, and nationality.
HLL
August, 2005 Roger F. Gay
Perform Analysis
This component should support use
of whatever analytical software is
needed by the application. Analysis
software needs to take set-up data
from the HLL system and the HLL
system needs to make use of results.
HLL
August, 2005 Roger F. Gay
Identify Alternatives
Match detailed problem
characteristics with alternative
solutions.
HLL
August, 2005 Roger F. Gay
Identify Alternatives:
Examples
Drainage Problem: Alter slopes,
modify irrigation schedule, or sell
the farm.
Threat Response: Hail the pilot,
call the President, fire a warning
shot, or shoot to kill.
HLL
August, 2005 Roger F. Gay
Recursion: Example
An alternative solution to the drainage
problem is to modify the irrigation
schedule. Irrigation is a “problem type” in
the general soil-water consultant’s
knowledge base.
The generalist can have a consultation
with the irrigation specialist before
recommending a drainage solution.
HLL
August, 2005 Roger F. Gay
Recursion
We can reuse the general flow of logic for
sub-problems and sub-problems of subproblems; compartmentalizing specialized
expertise; facilitating reuse of knowledge.
Easier to build complex “thinking”
systems.
HLL
August, 2005 Roger F. Gay
Recommendations
Traditional role of rule-based
expert systems.
HLL
August, 2005 Roger F. Gay
Simple Start: Is it of value?
• Design HLL for Experts
• Prototype HLL Engine and
Development Tools
• Wide Range of Applications
• A Step Toward Long-Term Goals
HLL
August, 2005 Roger F. Gay
The Future in Space
The intelligent space station has a responsibility
to maintain its orbit and protect itself and its
contents – including humans. The maintenance
manager (software) is informed of a change in
resistance in the outer hull and sends a
maintenance drone to look it over. The drone
returns imagary and measurements to the
manager who consults with specialized experts.
The drone is ordered to perform repairs.
HLL
August, 2005 Roger F. Gay
The Future in Agriculture
The farm manager (software) is informed of a
change in the average reflectance of soybean
crop leaves. It consults with a weather specialist
and then initiates soil-water measurements.
After consulting with specialists in irrigation and
agricultural economics, it adjusts the short-term
irrigation schedule to maximize profit.
HLL
August, 2005 Roger F. Gay
The Future Battlefield
The battle manager (software) makes field
assessments, defines targets, and orders
drones to attack. It regularly consults with
specialized experts to determine the choices
that will best implement a defined strategy.
Outcomes are returned to an executive
(software) that compares battle status to
expectations and alters strategy accordingly.
HLL
August, 2005 Roger F. Gay
The Future Consultant
HLL applications with the addition of
stored commentary that fills out
written reports in “standard HLL
form” explaining the problem,
providing detailed information,
analytical results, alternatives, and
recommendations.
HLL
August, 2005 Roger F. Gay
The Future Divorce Court
The “judge” (software) collects
detailed information, uses a specialist
to set standard of living goals for
parents and children and to find the
sound mathematically balance
between spousal and child support;
issues a written order and informs
payment tracking systems.
HLL
August, 2005 Roger F. Gay