Home | Publications | CS Home

Knowledge Considered Harmful

---

Shaul Markovitch and Paul Scott. Knowledge Considered Harmful. Technical Report 030788, The Center for Machine Intelligence, Ann Arbor, MI, 1988.

---

Abstract

In this paper we show that in some circumstances correct knowledge can be harmful. We identify three types of knowledge which impair either problem solving performance or the quality of solutions. Irrelevant knowledge and harmfully redundant knowledge have deleterious effect on performance while sociopathic knowledge leads to poorer solutions. Harmful redundancy and sociopathy have some important features in common. Each is a property of an entire knowledge base rather than of a particular piece of knowledge. Consequently procedures to minimize them are inherently complex since any such algorithm must consider all possible subsets of the knowledge base. Both of them impose limits on the quality of a knowledge base that may be acquired or learned by purely incremental methods.

---

Keywords: Utility Problem, Forgetting

---

Secondary Keywords:

---

Online version:

---

Bibtex entry:

 @techreport{Markovitch:1988:KCH,
  Author = {Shaul Markovitch and Paul Scott},
  Title = {Knowledge Considered Harmful},
  Year = {1988},
  Number = {030788},
  Type = {Technical Report},
  Institution = {The Center for Machine Intelligence, Ann Arbor, MI},
  Url = {http://www.cs.technion.ac.il/~shaulm/papers/pdf/Markovitch-Scott-KCH-1988.pdf},
  Keywords = {Utility Problem, Forgetting},
  Secondary-keywords = {Harmful Knowledge, Redundant Knowledge},
  Abstract = {
    In this paper we show that in some circumstances correct knowledge
    can be harmful. We identify three types of knowledge which impair
    either problem solving performance or the quality of solutions.
    Irrelevant knowledge and harmfully redundant knowledge have
    deleterious effect on performance while sociopathic knowledge
    leads to poorer solutions. Harmful redundancy and sociopathy have
    some important features in common. Each is a property of an entire
    knowledge base rather than of a particular piece of knowledge.
    Consequently procedures to minimize them are inherently complex
    since any such algorithm must consider all possible subsets of the
    knowledge base. Both of them impose limits on the quality of a
    knowledge base that may be acquired or learned by purely
    incremental methods.
  }

  }