McCarthy, J. (1995). What has AI in Common with Philosophy?. Proceedings of the 14th International Joint Conference on AI, Montreal, 2041-2042.

@InProceedings{McCarthy1995,
  author =        {John
L. McCarthy},
  title =         {What has AI in Common
with Philosophy?},
  booktitle =     {IJCAI},

year =   
     
{1995},
  pages =  
      
 {2041-2042},
  bibsource
 =     {DBLP,
 http://dblp.uni-trier.de/db/conf/ijcai/ijcai95.html#McCarthy95}
}

Author of the summary: Robert J. Lesurf, 2006, 2RJL2@qlink.queensu.ca

Cite this paper for:

• Philosophical positions which are compatible with intelligent programs.
• The viewing and interpretation of objects and situations by an AI system.
• Philosophies that aid in how intelligent systems can learn without having to experience.
• The pitfalls of philosophy with respect to AI.
• Determining which philosophical advances are most helpful construction of artificial intelligence systems.

Computer programs need to be equipped with a philosophy in order to reach human level intelligence. Such a philosophy includes a concept of what knowledge is and how may be obtained. [2041]

At present, there is no philosophical system sufficiently well defined to provide the foundation for a practical and intelligent computer program. [2041]

In most current AI work, the designer of the program constructs a philosophy in advance and builds a restricted representation of this into the system. [2041]

Hypothesis: philosophical attitudes that are required to be built into intelligent programs: [2041]

1. Science and common sense knowledge of the world must be accepted.
2. The concept of a “mind” has to be understood one feature at a time.
3. Beliefs and intentions are objects that can be formally represented.
4. It is permissible to use approximate concepts.

The philosophy of artificial intelligence can be viewed as its own academic subject area, with its own research methods aimed at clarifying the philosophical problems raised in AI. [2041]

A formal means of representing facts about the world must adequately represent actual available information, yet omit details that do not aid in decision making. [2042]

An attitude of free will and having to make choices must be built into intelligent systems. [2042]

Natural kinds of information must be described rather than defined in order to allow for approximate concepts to be used by the system. [2042]

There are four positions that can be taken with regard to objects or systems: [2042]

1. The physical structure of the system is treated.
2. The system is understood in terms of beliefs, goals, and intentions.
3. The system is understood in terms of its components.
4. Functional aspects of objects are viewed without regard to other properties.

It is sometimes necessary to assign quantifiable measurements to beliefs, hopes, promises, obligations, questions, answers, goals, and other reifications. [2042]

The use of counterfactuals is required to permit systems to learn without having to actually participate in an experience. [2042]

A system that relies only on the philosophy of relating actions to subsequent observations will severely limit what can be accomplished by the system. [2042]

The areas of AI which will benefit most from philosophical research include: [2042]

1. How to describe the relation between naming an occurrence and its corresponding suboccurences.
2. Establishing a useful level of responsiveness.
3. Uncovering what counterfactuals are useful for the system to learn intelligently.

Summary author's notes:

• The last two pages listed for the proceeding (pages 2043-2044) do not include information relevant to the paper.
• The author does not explicitly cite references for the article.