Barry Kort, Ph.D.; The Orenda Project; The MuseNet K-12 Project and Mary Harrsch.
Comment: Rik Min
Barry Kort writes:
Even the most sophisticated (model-based) simulations do not yet
include machine-based models of peers, because machines have not yet
evolved to the stage where they can instantiate a learning being
comparable to a human learner.
--> My reaction: That is true until now.
What makes multi-person role-playing interesting is the presence of peer learners. It is well-known that an emergent property of the capacity to learn is the presence of emotions (such as curiosity, fascination, confusion, puzzlement, bewilderment, anxiety, boredom, frustration, annoyance, embarrassment, satisfaction, and confidence).
--> My reaction: That is true.
Mary Harrsch writes:
Barry, I would be interested in a further explanation of these elements.
Perhaps because of my work with computer programming, I view most
simulations as rule-based. Computers, because of their binary nature,
perform functions by analyzing condition states of true or false then
executing the instruction provided by the "rule" as specified in the
program- if true, do this; if false, do that; Rules can be nested to
create complex, cascading conditions but in the end, the computer is
still following a rule.
--> My reaction:
Even the most complex rules or rule bases are in fact a serie of (simple) if-than-else-rules.
Barry writes:
Rule-based systems are predicated on the notion that every proposition
is either true or false, and that the behavior of the system can thus
branch on the truth value of each proposition.
This model goes back to Aristotle, and descends to us as bivalent logic
which nearly all digital computers employ.
There are two places where bivalent logic is too weak to model a
system. THe first place is where an analog calculus is required.
Modeling Newtonian Gravitational Mechanics, for instance, requires
continuous functions which must be solved mathematically to compute
thrust programs to carry out the necessary "rocket science."
The second place is handling uncertainty, where the truth value of
propositions is not known. In this case, we move from bivalent logic
to continuous-valued logic (also known as Bayesian Analysis), in which
propositions are assigned a "degree of belief" that varies continously
from 0 (false) to 1 (true). This method is popularly known as "Fuzzy
Logic."
--> That is well said.
Solving a general system model (as in a typical feedback control problem) is a computational task that rarely can be reduced to rule-based logic. More often, system models are solved iteratively by means of successive approximations. Iteration -- looping through a repeated process of incremental refinement -- is an instance of recursion. Recursion is a powerful technique that lies at the heart of computer science. The result of a recursion cannot itself be expressed in the form of a rule.
--> That is true.
Rather it is the result of a process in which rules are applied repeatedly until the error between the current approximation and the perfect solution is small enough to be acceptable as a working solution. That is the solution is technically "false" but "satisfactory" from a practical point of view. The challenge in computer science is to write software that models the behavior of the system under study. Many real-world systems are inherently analog in nature, and involve renewal processes that make them also recursive in nature. Such systems cannot be expressed in terms of simple rules based on bivalent logic. Nonetheless they can be computed (at least approximately) by means of continuous functions approximated by floating point arithmetic, and by means of iterative approximations that converge to the desired solution. Recursion is a beast that can be tamed to produce processes that are stable in the limit. But taming a recursion requires a higher order method of reasoning than can be captured in a set of rules. That's why rule-based expert systems are so disappointing. They lack the ability to manage a recursion, and thus lack the faculty to solve problems which call for analog calculus and recursion techniques.
--> Nice text.
Rik Min