关于人工智能科学之基础问题的哲学反思

the machine and the mind comuter was invented to take over certain mental labor.as computer becomes more and more capable ,a question raises naturally:"can a computer do everything that the mind does ?:"
structure neuroscience

Different Understandings :
Stucture (brain modeling)    ehavior (Turing Test)     Capability(expert systems)                   Function(cognitive facilities)        Principle (rationality&optimization)

the approach of NARSNARS follows the unified approach ,by using a single core technique (reasoning),and treating the other techniques as optional tools 

Reasoning System       Major components:
a formal language  a semantic theory  a set of inference rules   a memory structure
a control mechanism

Traditional Theories
Language and inference rules:first-order predicate calculus
Semantics:model theory
Memory:database or knowledge base

The Approach of NARS
NARS follows the unified approach,by using a single core thchnique (reasoning),and treating the other techniques as optional tools.

Different Techniquse:
(1)Logic --- to find the "laws of thought",and to use them as reasoning rules .
  Production System --- to represent knowledge as "condition-action"schemes.
  Probability --- to achieve the maximum expected utility in an uncertain environment.
Knowledge Base --- to store a huge amount of human knowledge in a computer-understandable
  form.
  Learning Algorithm --- to find patterns in the collected data .
  Robot --- to directly interact with the physical world
(3)Neural Network --- to train a network of simple units to get certain desired behaviors.
  Evolutionary System --- to generate winners in a population of solutions.
  Multi-agent System --- to depend on the cooperation of sub-systems to solve problems.

NARS as a Reasoning System
Major advantages of reasoning:
Separation between meta-level and object-level knowledge.
Domain-independent design.
Justifiability of the reasoning steps.

Summary
AGI is the attempt to return to the original and ultimate goal of AI.
The difference between AGI and mainstream AI is primarily in their research goals.
There is still no consensus the goal and technique  of AGI.

Reasoning System
Major components:   a formal language    a semantic theory     a set of inference rules
a memory structure   a control mechanism     The first three form a "logic".

Traditional Theories
Language and inference rules: first-order predicate calculus
Semantics: model theory
Memory:database or knowledge base 
Control:theory of computation and algorithm

Traditional Problems
Uncertainty: fuzzy concepts,changing meanings and truth-values,fallible results,conflicting
   evidence,nondeterministic inference process,...
   Semantic justification of non-deductive inference:induction,abduction,analogy,...
Counter-intuitive results:sorites paradox,implication paradox,confirmation paradox,Wason's
   selection task,...
   Computability and complexity:termination problem,combinatorial explosion,unanticipated
   problem,...

Proposed Solutions(以下很有可能只是一个问题)
non-monotonic logic,mental logic,paraconsistent logic,mental model,relevance logic,rcase-based reasoning,probabilistic logic,neural network,
inductive logic,genetic algorithm,temporal logic,heuristic algorithm,modal logic,situation calculus,learning algorithm
possible-world theory,...

Common Root of the Problems（重点：数理逻辑是智能逻辑的一个特例）
The traditional theories were developed in the study of the foundation of mathematics,while the problems appear outside mathematics.
The logic of mathematics is different from the logic of cognition and intelligence in their assumptions on the knowledge and resources.
一个系统它所拥有的知识和资源与它所要解决的问题之间的关系在AGI中很重要。
往往智能系统需要在知识不足和资源（例如：时间）有限的条件下工作。
Different Types of Systems
Pure-axiomatic: Systems's knowledge and resources are sufficient(with respect to the problems to be solved.)
Semi-axiomatic: some(but mot all)aspects of the knowledge and resources are sufficient.(或称：非经典逻辑)
Non-axiomatic:  the knowledge and resources of the system are insufficient.

What is NARS
NARS(Non-Axiomatic Reasoning System) is a reasoning system that is fully based on the Assumption of Insufficient Knowledge and Resources.
NARS is a finite, real time,open,and adaptive system.
NARS is different from the traditional systems in all major components.

Inheritance statement
s->p: there is an inheritance
relation fhrom term S to term P.
S is a specialization of P;
P is a generalization of S.

转载于:https://my.oschina.net/u/1861582/blog/285590