6-Procedural.html

Pei Wang, 07/18/2013 09:34 am

Download (10.7 kB)

 
1
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
2
<HTML>
3
<HEAD>
4
        <META HTTP-EQUIV="CONTENT-TYPE" CONTENT="text/html; charset=utf-8">
5
        <TITLE>NARS Tutorial 6</TITLE>
6
        <META NAME="GENERATOR" CONTENT="OpenOffice.org 3.4  (Unix)">
7
        <META NAME="CREATED" CONTENT="0;0">
8
        <META NAME="CHANGED" CONTENT="20130718;16321900">
9
        <STYLE TYPE="text/css">
10
        <!--
11
                H3.ctl { font-family: "Arial Unicode MS" }
12
                H2.ctl { font-family: "Arial Unicode MS" }
13
        -->
14
        </STYLE>
15
</HEAD>
16
<BODY LANG="en-US" DIR="LTR">
17
<H3 CLASS="western" ALIGN=CENTER><A HREF="http://www.mindmakers.org/documents/13">NARS
18
Tutorial</A></H3>
19
<H1 ALIGN=CENTER>NAL-7 to NAL-9: Procedural Inference</H1>
20
<H2 CLASS="western">1. NAL-7: Temporal inference</H2>
21
<P>Events are statements with time-dependent truth-values.
22
Event-related terms and concepts have time-dependent meaning.
23
[Related problem: &quot;<A HREF="http://en.wikipedia.org/wiki/Grue_and_bleen">the
24
new riddle of induction</A>&quot;] 
25
</P>
26
<P>To &quot;reason about time&quot; and to &quot;reason in time&quot;
27
are different — reasoning takes time, and time changes. NARS gets a
28
&quot;personal sense of time&quot; using its internal working cycle
29
as a clock. The &quot;real-time experience&quot; of NARS measures the
30
time interval between perceived events. 
31
</P>
32
<P>Temporal relations are defined relatively between events. There
33
are two basic temporal relation: &quot;before–after&quot; and
34
&quot;when&quot;. Here NAL allows different granularity and accuracy
35
in representation. Temporal relations are combined with logical
36
connectors and copulas: two versions of conjunctions and 3 versions
37
of implication (equivalence). 
38
</P>
39
<P>Tense: event compared to &quot;now&quot;. Problem: &quot;now&quot;
40
is a moving reference. Tense is used only in external communication,
41
not in internal representation. The need to attach a time-stamp to
42
each sentence to remember is creation (perceiving or deriving time). 
43
</P>
44
<P>Multiple ways to represent temporal information: (1) built-in
45
temporal relations, (2) internal clock and tense, (3) acquired
46
temporal relations. 
47
</P>
48
<P>In IL-7, the logical and temporal relations are processed in
49
parallel by the inference rules. Then the inference rules are
50
extended into NAL-7 strong rules, and their reversions becomes NAL-7
51
weak rules. Tensed sentences are processed similarly. NARS also use
52
tensed judgment as evidence for eternal judgment.</P>
53
<H2 CLASS="western">2. NAL-8: Reasoning on operations and goals</H2>
54
<P>An operation is a &quot;realizable event&quot; or &quot;executable
55
term&quot;, with an <I>operator</I> (as a relation), plus some input
56
and output arguments (as a product). In this way, declarative
57
knowledge and procedural knowledge are unified using &quot;procedural
58
interpretation&quot;, as in <A HREF="http://en.wikipedia.org/wiki/Logic_programming">logic
59
programming</A>. 
60
</P>
61
<P>In programming languages, operations correspond to instructions,
62
statements, routines, procedures, functions, etc. In the definition
63
of an operation, an input argument can be an independent variable,
64
and an output argument is a dependent variable. When executed, all
65
input arguments are instantiated by constants, and the output
66
arguments are determined accordingly. 
67
</P>
68
<P>Knowledge about an operation can either be inheritance/similarity
69
or implication/equivalence. The inference rules are used as usual.
70
Since NAL does not attempt to completely specify the conditions and
71
consequences of an operation, it does not suffer from the <A HREF="http://en.wikipedia.org/wiki/Frame_problem">Frame
72
Problem</A>. 
73
</P>
74
<P>Compound operations can be formed as compound events, and executed
75
without much reasoning. With a given basic operation set, Narsese can
76
be used as a programming language, and reasoning carries out
77
planning, skill learning, self-programming, etc., like a logical
78
programming language with the inference engine as an interpreter.
79
NARS can also learn to program in another programming language by
80
using Narsese as the meta-language. 
81
</P>
82
<P>The execution of an atomic operation is normally not the duty of
83
the inference engine, but that of the host system (as the &quot;body&quot;)
84
or a peripheral device (as a &quot;tool&quot;). In the sensorimotor
85
interface of NARS allows operations to be registered in NARS, with
86
initial knowledge. In this way, NARS can serve as a mind of a robot
87
or an intelligent operating system controlling various application
88
software. 
89
</P>
90
<P>An &quot;NARS+&quot; is a NARS plus certain tools/organs. 
91
</P>
92
<P>Operations decide what the system <I>can do</I>, but not
93
necessarily what the system <I>will do</I>. A <I>goal</I> is an event
94
that the system desires to realize. 
95
</P>
96
<P>Traditional representations of &quot;goal&quot; in AI include
97
state and utility/fitness/reward function. Their common problem is to
98
represent a goal as certain, static, and complete. 
99
</P>
100
<P>AIKR implies that a system may have conflicting, competing, and
101
changing goals. To handle this situation, a relative &quot;degree of
102
desire&quot; is need. A desire-value is defined on every event,
103
indicating the extent it implies a (virtual) desired event.
104
Therefore, desire-value is a derivative of truth-value, and can be
105
processed accordingly. 
106
</P>
107
<P>Goal processing in NARS: 
108
</P>
109
<UL>
110
        <LI><P STYLE="margin-bottom: 0in">If a goal matches an existing
111
        judgment, it has been realized to the extent specified by the
112
        truth-value of the later. 
113
        </P>
114
        <LI><P STYLE="margin-bottom: 0in">If a goal is an operation, to
115
        realize it means to execute the corresponding procedure. 
116
        </P>
117
        <LI><P>Otherwise, to derive goals by using the inference rules
118
        backwards. 
119
        </P>
120
</UL>
121
<P>An input or derived goal is pre-processed via revision to modify
122
the desire-value of the corresponding event. 
123
</P>
124
<P>The decision-making rule check the plausibility of a highly
125
desired event, and turn it into a goal that is actually pursued. This
126
is where binary decision is needed, in spite of the multi-valued
127
truth-values, desire-values, and priority-values. Due to AIKR, goal
128
alienation becomes inevitable.</P>
129
<P>[Compared with traditional <A HREF="http://en.wikipedia.org/wiki/Decision_theory">decision
130
theory</A>, such as <A HREF="http://en.wikipedia.org/wiki/Expected_utility">the
131
expected utility hypothesis</A>, NARS does not assume given
132
probability and utility functions.]</P>
133
<P>[Compared with traditional AI techniques: 
134
</P>
135
<UL>
136
        <LI><P STYLE="margin-bottom: 0in"><A HREF="http://en.wikipedia.org/wiki/General_Problem_Solver">planning
137
        by search</A>, 
138
        </P>
139
        <LI><P STYLE="margin-bottom: 0in"><A HREF="http://en.wikipedia.org/wiki/Belief–desire–intention_software_model">BDI
140
        model</A>, 
141
        </P>
142
        <LI><P STYLE="margin-bottom: 0in"><A HREF="http://en.wikipedia.org/wiki/Production_system">Production
143
        system</A> and <A HREF="http://en.wikipedia.org/wiki/Behavior-based_robotics">reactive
144
        agents</A>, 
145
        </P>
146
        <LI><P STYLE="margin-bottom: 0in"><A HREF="http://en.wikipedia.org/wiki/Inductive_logic_programming">inductive
147
        logic programming</A>, 
148
        </P>
149
        <LI><P STYLE="margin-bottom: 0in"><A HREF="http://en.wikipedia.org/wiki/Reinforcement_learning">reinforcement
150
        learning</A>, 
151
        </P>
152
        <LI><P><A HREF="http://en.wikipedia.org/wiki/Genetic_programming">genetic
153
        programming</A>.] 
154
        </P>
155
</UL>
156
<H2 CLASS="western">3. NAL-9: Mental operations and goals</H2>
157
<P>NAL-9 allows certain internal operations of NARS to be controlled
158
by its own inference process. 
159
</P>
160
<P>The selection of operations is made by balancing flexibility,
161
efficiency, and stability. A candidate list is given in Section 13.1
162
of the textbook.</P>
163
<P>After the full implementation of NAL-9, the control mechanism of
164
NARS will be a combination of some &quot;autonomic&quot; processes
165
and some &quot;voluntary&quot; processes.</P>
166
<P>Some mental operations will correspond to <I>feeling</I> and
167
<I>emotion</I>, which are based on evaluation of various factors from
168
the system's viewpoint (as extended desire-values) and the operations
169
that are invoked as quick responses to the evaluation results.
170
Emotion is necessary for a complicated intelligent system. 
171
</P>
172
<P>The system's internal experience corresponding to the mental
173
operations are on the conceptual-level, and it forms the system's
174
consciousness. 
175
</P>
176
<P>Like its &quot;external&quot; experience, the &quot;internal&quot;
177
experience of NARS is also subjective, discontinuous, and incomplete.
178
Since these two types of experience are produced by different
179
operations, there is an &quot;<A HREF="http://en.wikipedia.org/wiki/Explanatory_gap">explanation
180
gap</A>.&quot; 
181
</P>
182
<H2 CLASS="western">4. Summary</H2>
183
<P>NARS provides a unified mechanism to reproduce various cognitive
184
functions. 
185
</P>
186
<P>For example, some cognitive functions can be uniformly represented
187
as question-answering: 
188
</P>
189
<UL>
190
        <LI><P STYLE="margin-bottom: 0in"><TT>? --&gt; T</TT> : searching 
191
        </P>
192
        <LI><P STYLE="margin-bottom: 0in"><TT>T --&gt; ?</TT> : recognition 
193
        </P>
194
        <LI><P STYLE="margin-bottom: 0in"><TT>T &lt;-&gt;?</TT> : naming 
195
        </P>
196
        <LI><P STYLE="margin-bottom: 0in"><TT>? ==&gt; T</TT> : explanation 
197
        </P>
198
        <LI><P STYLE="margin-bottom: 0in"><TT>T ==&gt; ?</TT> : prediction 
199
        </P>
200
        <LI><P><TT>T &lt;=&gt;?</TT> : summarizing 
201
        </P>
202
</UL>
203
<P>The relation between functions and inference types is not
204
one-to-one, but many-to-many. In NARS, causal relation is an acquired
205
notion and has domain-dependent meaning. There are also different
206
types of &quot;explanation&quot;. [Temporal induction and <A HREF="http://en.wikipedia.org/wiki/Learning#Classical_conditioning">classical
207
conditioning</A> are closely related.]</P>
208
<P>[The following association is proposed by <A HREF="http://plato.stanford.edu/entries/peirce/#dia">Peirce</A>:
209
</P>
210
<UL>
211
        <LI><P STYLE="margin-bottom: 0in">deduction : demonstration 
212
        </P>
213
        <LI><P STYLE="margin-bottom: 0in">induction : generalization 
214
        </P>
215
        <LI><P>abduction : explanation 
216
        </P>
217
</UL>
218
<P>However, in general a cognitive function is provided in multiple
219
ways by the cooperation of various inference rules.]</P>
220
<P>Similarly, many traditionally separated processes are absorbed
221
into the reasoning framework:</P>
222
<UL>
223
        <LI><P STYLE="margin-bottom: 0in">perception as action, 
224
        </P>
225
        <LI><P STYLE="margin-bottom: 0in">reaction and planning, 
226
        </P>
227
        <LI><P>incremental skill learning. 
228
        </P>
229
</UL>
230
<P>Compared with the various <A HREF="http://bicasociety.org/cogarch/architectures.htm">cognitive
231
architectures</A>, the unified approach followed by NARS has the
232
advantages of consistency and simplicity.</P>
233
<HR>
234
<H2 CLASS="western">Reference</H2>
235
<UL>
236
        <LI><P STYLE="margin-bottom: 0in"><A HREF="NAL-Wang.pdf"><I>Non-Axiomatic
237
        Logic: A Model of Intelligent Reasoning</I></A>, Ch. 11-14 
238
        </P>
239
        <LI><P><A HREF="http://code.google.com/p/open-nars/source/browse/trunk/nars-dist/Examples-1.3.3/Example-NAL7-abridged.txt"><I>Temporal
240
        Examples</I></A>, <A HREF="http://code.google.com/p/open-nars/wiki/ProceduralExamples"><I>Procedural
241
        Examples</I></A> 
242
        </P>
243
</UL>
244
</BODY>
245
</HTML>