计算机能思考吗？图1专题10：“计算机是不是具有科学理性？”

Can Computers Think? The History and Status of the Debate - Map 1 of 7

问题10

Issue Area: Can computers reason scientifically?

问题域： 计算机是不是具有科学理性？

The link to the part of the map this discussion is about:

http://www.macrovu.com/CCTWeb/CCT1/CCTMap1Reason.html

1. Common start point by Alan Turing (omitted)

(Disputing 1) 115.Computers can't reason scientifically. Computers are unable to think and reason as human scientists do.

（反驳1）计算机不可能具有科学理性。计算机不能像人类科学家那样理性思考科学问题。

(Supporting 115) 116. Harry Collins, 1964 Scientific reasoning requires social agreement. Computers cannot reason scientifically because they are not members of society. Scientific laws and data do not follow from the application of an algorithm, but are developed through quasipolitical process of negotiation. (It seems social agreement here does indicate the term in the sense as it's used in social science; that may lead to a larger political/historical philosophical question whether (original) scientific discoveries and achievements (of a nation, lets say) should strongly relate to or depend on the extent of the social development or evolution; although to the translator, it seems the case according to the most apparent historical facts and thus it makes more sense when the question is interpreted this way -- translator)

（支持115）Harry Collins, 1964 科学理性要求社会契约。计算机不能进行科学理性思考因为它们不是社会成员。科学规律和数据并不是算法运用的结果，而是通过谈判的过程得以实现。

(Supporting 116) 117. Harry Collins, 1964The social test.The importance of socialization is demonstrated by the "social test", a variant of Turing test. In the test, a human control and a computer are both given a passage of "mucked up" English. Both the human control and the computer must correct all the errors and transliterate the passage into English. If a judge cannot tell which text was error-corrected by machine and which by the human control subject, then the machine passes this test for socialization. Note: For more on the Turing test, see Map 2.

（支持116）Harry Collins, 1964 社会测试。社会化的重要性可以通过“社会测试”来演示，它是图灵测试的一个变种。在这个测试中，人和计算机均被指派翻译一段被弄乱的英语文字。两者都被要求更正其中错误并将其重新翻译成正确的英语文本。如果一个评判人不能分辨哪个更正文本是由人类完成，哪个是有计算机完成的，那么机器就通过了这个社会化测试。注：了解更多关于图灵测试，见第二图。

(Supporting 115) 118. Carl Hempel, 1985 Computers can't introduce new terms or explanatory principles. A computer cannot be original because it cannot introduce new theoretical terms or principles. Computers' discoveries are limited to those that can be expressed using the program's fixed vocabulary and conceptual apparatus. Human discovery, by contrast, involves the introduction of new terms and principles that cannot be defined in terms of those previously available.

（支持115）Carl Hempel, 1985 计算机不能引入新的名词或解释性理论。计算机不具有原创性因为它不能引入新的理论名词和原理。计算机的发现都局限于那些能够被程序固定的词汇和概念工具所表述的东西。人类发现则包含了不能被已有概念所定义的新的名词和原理。

(Disputing 118) 119. Richard Scheines, 1988 Computers can introduce new terms. Computers can introduce new terms using automated principles of explanatory adequacy. This has been shown using a program that uses explanatory adequacy principles to introduce new terms in the domain of "causal models" -- a class of mathematical theories popular in social science.

（反驳118）Richard Scheines, 1988 计算机能引入新名词。计算机能通过适用自动化合理解释原理引入新的名词。这点已经通过一个能使用合理解释原理而在“因果模型”领域中产生新名词的程序而解释，而“因果模型”是一组在社会科学中受欢迎的数学理论。

(Supporting 115) 120. Carl Hempel, 1985 Computers can't adequately evaluate hypotheses. A computer model of scientific discovery would have to use a criterion of preference to choose between hypotheses that account for available data equally well. But criteria of preference tend to be imprecise and idiosyncratic, so it is unlikely that such a criterion could be implemented on a computer.

（支持115）Carl Hempel, 1985计算机不能恰当地评估假设。一个科学发现计算机模型必须适用一个优选准则从一组能对已有数据作出相同程度合理解释的假设中作出选择。不过这样的优选准则倾向于不精确并且比较怪异，因此这样的准则能在计算机上实现的可能性不大。

(Disputing 115) 121. Computers have already reasoned scientifically. Computer systems exist that have reasoned as scientists do, proposing explanatory hypotheses and choosing among them.

（反驳115）计算机已经能够进行科学理性思考。已经有计算机系统能像科学家那样思考，提出解释性假设并从中进行选择。

(Supporting 121) 122. Pat Langley, Hubert Simon, Gary Bradshaw, and Jan Zytkow, 1987 Implemented Model: BACON. A program for discovering laws from data by applying heuristics. BACON has discovered Kepler's law of planetary motion, Galileo's law of uniform acceleration, and Ohm's law of electrical resistance. Note: The history of BACON is complex and extends back into the 1960s.

（支持121）Pat Langley, Hubert Simon, Gary Bradshaw, and Jan Zytkow, 1987 实现模型：BACON。一个从已有数据中通过运用启发性方法发现规律的程序。BACON已经发现了行星运动的开普了定律，统一加速的伽利略定律，以及电阻的欧姆定律。注：BACON的发展史非常复杂并可以追溯到1960年代。

(Disputing 122) 123. Harry Collins, 1994 BACON only works when humans filters its data.Bacon only works through its interaction with scientists who filter its data and thereby predetermine its results. If humans did not constrain its data, it is doubtful that BACON would produce any original science.Supported by "The Front-End Assumption is Dubious, " Box 75.

（反驳122）Harry Collins, 1994 BACON只能在人类过滤其数据的情况下才能运作。BACON通过它和帮它筛选数据科学家的交互才能工作，因此这已经预先决定了其结果。如果人类不帮它限定数据，那么BACON是否仍旧能得出科学结论就很成问题了。被75框的“前端假设是可疑的”支持。

(Supporting 121) 124 B. G. Buchanan, D. H. Smith, W.C. White, R. Gritter, E. A. Feigenbaum, J. Ledergerg, and C. Djerassi, 1976Implemented Model: DENRAL. DENRAL is an expert system that analyzes and identifies chemical compounds by forming and testing hypotheses from experimental data. Meta-DENRAL, a component of DENRAL, has discovered how to synthesize previously unknown chemical compounds as well as entirely new rules of chemical analysis. It even has a publication to its credit.

（支持121）B. G. Buchanan, D. H. Smith, W.C. White, R. Gritter, E. A. Feigenbaum, J. Ledergerg, and C. Djerassi, 1976 已实现模型：DENRAL。DENRAL是一个专家系统，它通过从实验数据中生成和测试假设来分析和辨别化合物。Meta-DENRAL——一个DENRAL的组件——已经发现了如何合成以前未知的化合物甚至全新的化学分析法则。有一部刊物专门介绍之。

(End of First Map)

I have to come back to the preparation for my part of the application demonstration tmr morning...