home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Frostbyte's 1980s DOS Shareware Collection
/
floppyshareware.zip
/
floppyshareware
/
LGB
/
HYPER.ZIP
/
FILE21
< prev
next >
Wrap
Text File
|
1988-02-01
|
7KB
|
127 lines
How do Expert Systems and Hypertext Compare?
============================================
An expert system moves through a set of rules in much the same way as an
individual who selects paths in a decision tree.
For dispensing knowledge, each technology has certain advantages. For
example:
Expert systems offer Hypertext systems offer
==================== =======================
Automatic goal selection Ease of construction
Multiple-variable processing Ease of use
Speed in reaching a decision Ease of modification
Sensitivity analysis (browsing)
Transmission of knowledge
Let's consider the importance of each of these factors.
AUTOMATIC PATH SELECTION An expert system can monitor time, pressure,
temperature, etc. to automatically eliminate certain decision paths
in the search for answers. In contrast, hypertext systems depend upon
operator responses to make selected paths to answers.
CALCULATED DECISIONS Expert systems often include formulas that convert any
number of variable inputs into a single path selection. Instead of
this parallel processing (multiple inputs -- single answer), hypertext
decision systems use a sequence of decision points (serial processing)
in order to convert multiple inputs into a single path.
SPEED An expert system may reach the appropriate decision within a
fraction of a second (i.e., avoiding an aircraft collision). With
hypertext, speed is limited by the ability of the user to complete
multiple sequences of reading, understanding, and selecting choices at
each decision point.
However, hypertext has several significant advantages over expert systems
<file66 EXCEPTIONS) in dispensing information or finding solutions --
construction speed, time to learn, knowledge representation, ease of
modification, sensitivity analysis, and transmission of knowledge. Here's
how.
CONSTRUCTION SPEED A good 200-rule expert system may take a team of
knowledge engineers two years to build. In contrast, most experienced
computer users <file53 TALENTS NEEDED> can build in one day a good
200-node decision tree leading to relevant advice. Our family of
hypertext products includes a number of utility programs that enable
you to do just that.
TIME TO LEARN It often takes many years for experts in other fields (e.g.,
PROLOG, LISP, SMALLTALK) to efficiently embed their knowledge into
automatic systems. In contrast, all experts can master the tools of the
hypertext system craft with just a few weeks effort.
KNOWLEDGE REPRESENTATION Two major difficulties exist in building expert
systems. First, how can users acquire expert-level knowledge? (often
the expert can't even explain it). Second, how can this knowledge be
represented so that a machine can generate solutions from it? In
hypertext systems, the knowledge <file55 DEFINITION> is represented
using existing everyday formats (i.e., text, diagrams, pictures).
MODIFICATION EASE Once completed, expert systems tend to be notoriously
difficult to update or modify (many interactions are often hidden
from users) and then to validate again (who knows when an expert
machine starts or stops producing expertise?). With hypertext
systems, modifications and improvements are as easy as adding
branching comments and footnote text using a word processor.
TRANSMISSION OF KNOWLEDGE Expert machines generally do not explain
to users the actual methods that will lead to a particular decision.
With hypertext, users directly participate in each and every decision
that leads to a particular expertise. This process of openly
displaying the structure <FILE75 STRUCTURE> and uses of knowledge
readily transmits it to users of hypertext systems.
SENSITIVITY ANALYSIS Expert machines usually provide a single answer
supported perhaps by a confidence factor such as 82 percent certainty
(whatever that means). Hypertext systems allow users to rapidly test
alternative paths <FILE56 BROWSE> to see how sensitive the advice may
be to changes in the initial assumptions.
In my opinion, given the years of AI promises and the relative lack of
success in creation and delivery of workable expert systems, the most
important pocketbooks are rapidly closing against further use of computer
processing of rule-based systems as a method of vending expertise from a
disk.
Generally, over the last 30 years, technologists haven't made operations
research practical. Consequently, if you can't build machines that are
effective in quantitative reasoning, how can you build machines that are
effective in subjective reasoning. Making subjective decision machines work
is at least several magnitudes more difficult. <FILE27 REASONS>
Why have such decision machines failed? I think the answer is simple.
At best, numbers and mathematical processing are a second-order
approximation to reality. In contrast, language is the first-order
approximation to reality as numbers only verify language. What does that
mean?
Consider the difference in views of reality contained in these two quotes:
┌───────────────────────────────────────────────────────────┐
│ "If you can't measure it, it doesn't exist" │
│ Fundamental maxim of quantitative thinking │
└───────────────────────────────────────────────────────────┘
┌───────────────────────────────────────────────────────────┐
│ "Language determines reality. If you haven't named it, │
│ you can't measure it...and measuring wrong things tends │
│ to produce useless answers." │
│ Fundamental maxim of linguistic thinking │
└───────────────────────────────────────────────────────────┘
From my viewpoint, experts don't measure or calculate. Instead, experts
have superior language that enables them to see, describe, and predict
reality in superior ways.
The foundation of expert machines is calculation while the foundation of
hypertext is classification. That's why I think hypertext is superior in
capturing and dispensing knowledge.
Neil Larson 1/16/88 FILE21
44 Rincon Rd., Kensington, CA 94707
Copyright MaxThink 1988 -- Call 415-428-0104 for permission to reprint