Printer Friendly

Indian theories of knowledge compared with western theories and artificial intelligence.


The paper while surveying the Western and Indian epistemological theories, presents a comparison of the Nyaya (NN) treatment of knowledge with that of Artificial Intelligence (AI). Looking for newer definitions of knowledge with matching representation techniques has been the main concern of AI and Cognitive Science (Cog-Sci) researchers. Since most efforts in knowledge modeling have resulted in disappointments, there has been a shift in the major goals and approaches towards AI and Cog-Sci. On the goals side, there are now smaller objectives than high flying ones. On the approach side, there is a growing realization that traditional wisdom in ancient philosophies might have an answer to the complex problem. Therefore, the present paper is trying to put together the Indian views on knowledge in general and compare the NN techniques with those of the AI to see if the former can benefit the latter.


In the era of Fifth generation computing, there has been a tremendous effort to simulate intelligent behavior in machines. There are two ways in which the researchers are trying to solve this problem--by the grammar based computing in which the Knowledge of Language (KOL) is sought to be given to machines for Natural Language Understanding (NLU) or Processing (NLP), and by way of knowledge based computing in which the background/contextual or Real World Knowledge (RWK) is sought to be represented in machine for understanding real world situations and problem solving. There is a growing realization that for simulating intelligence in machines RWK representation is essential. However, so far no single framework applicable to all knowledge types, identities of the locus, object, and instrument of cognition could be evolved. Growing skepticism on these counts has led many AI researchers wonder whether

* knowledge can be represented

* computers can en/de-code knowledge in ordinary language

* computers can think and understand

* computers need knowledge to interact with people or other machines

In India, particularly in the Sanskrit language, there are definitive accounts of both the KOL and RWK. Panini's grammar, though definitely relevant for RWK., is actually an account of KOL which can serve as a model for other languages as well. The NN system provides one of the earliest accounts of RWK which is not very dissimilar to those of the AL philosophers. In this light, it is imperative to see what the NN system has to offer. Therefore, the paper attempts to juxtapose the Indian epistemological tradition, the NN in particular with that of the Artificial Intelligence


The western view on knowledge has centered round where knowledge is given (a priori--"from what is before") or acquired (posteriori--"from what is after"). While the former evolved into rationalistic mode of inquiry, the latter evolved into the empiricist mode. Plato argues that there is an awareness of absolute, universal ideas or forms, existing independent of any subject try in; to apprehend to them. Aristotle puts more emphasis on logical and empirical methods for gathering knowledge. He also accepts the view that knowledge is an apprehension of necessary and universal principles.

The early rationalists--Descartes, Spinoza and Leibniz--considered deductive reasoning based on axioms as the main source of knowledge. For the early empiricists--Francis Bacon and John Locke--the main source and Final test of knowledge was sense perception. While Berkeley agreed with Locke's views on sourcing knowledge to ideas, he rejected the separation of ideas from objects. Hume, an empiricist, disagreed with Berkeley's conclusion that knowledge can be sourced from ideas only. Hume divided all knowledge into two kinds (Distante 2003):

relations of ideas

* the knowledge found in mathematics and logic

* this knowledge is exact and certain but provides no information about the world.

matters of fact

* the knowledge derived from sense perception.

During renaissance and after the argument continued among the empiricists, rationalists, pragmatists and constructivists. According to the reflection-correspondence theory, of the empiricist school, knowledge results from a reflection of external objects to our brain or mind. According to them the meaning is what the words denote, and language, like a mirror, reflects things around us. A proposition is true if the relations between the images of things correspond to the relations between the things themselves. Wrong and distorted reflection (images which correspond to no real thing in the world) indicates falsity. It was Kant who worked to synthesize the contending arguments of the rationalists and empiricists. According to Kant, knowledge results from the organization of perceptual data on the basis of inborn cognitive structures or categories. Kantian categories include space, time, objects and causality. Kant distinguished between three types of knowledge (Distante 2003):

analytical a priori

* exact and certain, but also uninformative because it makes clear only what is contained in definitions

synthetic a posteriori

* conveys information about the world learned from experience

* is subject to the errors of the senses

synthetic a priori

* discovered by pure intuition and is both exact and certain

* it expresses the necessary conditions that the mind imposes on all objects of experience

In the aftermath of Kant, epistemological tug-of-war continued with scholars like Hegel and Comte. But with the American pragmatist school, the empiricism began to take a lead. In their view:

* knowledge consists of models that attempt to represent the environment in such a way as to maximally simplify problem-solving

* no model can ever hope to capture all relevant information, and even if such a complete model would exist, it would be too complicated to use in any practical way

* therefore, we must accept the parallel existence of different models, even though they may seem contradictory.

* the model which is to be chosen depends on the problems that are to be solved. The basic criterion is that the model should produce correct (or approximate) predictions (which may be tested) or problem-solutions, and be as simple as possible

The Constructivists took a radical view that knowledge is built up from scratch by the subject of knowledge. There are no "givens", neither objective empirical data or facts, nor inborn categories or cognitive structures. According to evolutionary epistemology, knowledge is constructed by the subject or group of subjects in order to adapt to their environment. Construction is an on-going process at different levels, biological as well as psychological or social.

In the modern times, the scholars like Husserl (phenomenology), Wittgenstein (logicalpositivism) and linguistic analysts approached epistemology in their own ways building the mainstream discussion.

At another level, idealists and realists were proposing two opposing schools of thought on epistemology. While the idealists attribute primary role to consciousness, or the immaterial mind, for the realists, mind-independent physical objects exist and can be known through the senses. The critical realists defined the knowledge of extra mental reality as three-way relationship between mind, object and content. For American neorealists, knowledge was relation between two objects (or the apprehension of such relations).

The philosophers like Kant, Spencer, Bergson, Alexander, and Dawes Hicks are said to be the founders of the Act Theory of Knowledge which explains knowledge in terms of activity. The behaviorists call it an activity of the body But the modern European philosophers like Moore and Broad have refuted this theory. Among the western philosophers, Berkeley is said to be the propounder of the Selfsubsistency Theory according to which knowledge is self subsists (Bijalwan 1987).


The ancient Indian thinkers had been quite conscious of the formidability of the problem. Their intellectual arguments, establishments, and disputations have helped build up a vast epistemological tradition (Singh & Jha 1994). The following outline will summarize the Indian position leading to the NN position (Bijalwan 1987):

Yogacara Buddhists

* they recognize vijaana (consciousness)

* according to them, world is built by consciousness which is self subsistent

* everything except vijaana is unreal

* sa=ga, Vasubandhu, Di=gnaga, Dharmakirti

* No objective world independent of perceiving mind

* Subject and object of cognition were different modes of alaya' (continuously changing stream of consciousness)


* knowledge is ultimate and reveals itself

* the Buddhist view is similar to the Vedanta position

* the Buddhists however, do not recognize the existence of a single intelligent abiding principle and admit only a chain of impressions

Madhyamika Buddhists and the Mimasakas

* refer to knowledge as an activity.

* knowledge is an existent fact that consists in the act of showing and leading to and object

* the Bhattas support the view that knowledge is an act of the soul.

* Kumarila also accepts knowledge as a dharma (property) of the soul


* according to the Naiyayikas, Gautama (Nyayasastra), Jayanta (Nyayamanjari), Gargeya (Tatvacintamani), and Visvanatha (Nyayasiddhantarmuktavali), knowledge is neither an act or relation but a guna, a (quality) of the self (Bijalwan 1987)

* Gautama (Nyayasastra)

* explains knowledge in terms of buddhi and refers to upalubdhi and jnana as its synonyms

* Jayanta (Nyayamanjari)

* uses this explanation as a counter-argument to the Samkhya view that knowledge is a mode of buddhi which transforms itself into the shape of the object that it cognizes.

* terms buddhi, jnana and upalabdhi represent different concepts. knowledge is perceived by manas (mind) as physical qualities are perceived by the sense organs.

* since the mind is an instrument to knowledge, and buddhi is a guna (quality) of the soul, the latter according to Jayanta is knowledge.


Knowledge representation (KR) and Information Retrieval (IR) are fundamental areas of AI research. From the AI point of view, knowledge can be defined as objects and attributes, or even an algorithm containing objects and procedures. The so-called "intelligent behavior" can be said to result from interaction between "data" and "processes" or "abilities" and "memory storage". According to Abrial (1974), a database is a model of the evolving physical world. The state of this model at a given instant represents the knowledge it has acquired from the world. But knowledge is more than static encoding of facts in interacting with the world. A basic premise of AI is that knowledge of something is the ability to form a mental model that accurately represents the thing as well as the actions that can be performed by it and on it. Then by testing actions on the model, persons (or robot) can predict what is likely to happen in the real world (Sowa 1984). Earlier, the dilemma with the AI people was whether to represent knowledge by following a declarative approach or a procedural approach. It has been now finally settled that a combination of a declarative as well as a procedural approach would be most suitable for knowledge representation. Other well known techniques for KR include--predicate calculus, semantic nets, frames, conceptual graphs, scripts, semantic primitives and neural nets which follow a wide range of strategies for KR. Still, reducing the enigma called knowledge to a definition or two would not be an end in itself. The entire process of human cognition would have to be understood and effective means would have to be evolved to represent knowledge (Hopcroft & Lipman 1979).


This section will present a classification of knowledge according to the NN system and its comparison with AI. The valid knowledge (yathartha) is termed pramana, and the invalid knowledge (ayathartha is termed apramana.

* pratyaksa (perception)

* anumana (inference)

* upamana (analogy)

* sabda (verbal testimony) and latter is sub-classified as

* samsaya (doubt)

* viparyaya (error)

* tarka (hypothetical argument)

* smrti (memory)

5.1. Comparison with AI

The following sub-sections define each of the above sub-classifications of invalid and valid knowledge and give their AI parallels--

5.1.1. apramana--the invalid knowledge

It is defined as false knowledge, doubtful or illusionary. The four types of invalid knowledge are defined and compared with AI as follows: smrti (memory)

* is impressions left by previous knowledge

* it is non valid because the object of knowledge is absent at the time of remembrance

* AI explanation

* memory as gaps in knowledge or probabilistic knowledge which must be corrected before processing starts and conclusion is arrived at

* such gaps or incomplete knowledge may be replaced by new knowledge [right arrow] non-monotonic knowledge (McDermott & Doyle 1980)

* static memory [right arrow] knowledge of object (WHAT)

* short term : memory in current focus

* long term: memory for future use

* procedural memory [right arrow] knowledge of behavior (HOW)

* short term

* long term samsaya (doubt)

* conflicting judgment on the precise character of an object of cognition (Bijalwan 1987)

* neither true or false, only wavering judgment

* AI explanation

* ambiguity resolution and bug handling come close to samsya tarka (hypothetical judgment)

* AI explanation

* Heuristics

* when reality is not known, truth is ascertained trying alternatives

* atmasraya (petito principi)

* if A [right arrow] A then A must be different from A

* anyonyasraya (dependent tarka)

* A[right arrow]B[right arrow]A

* cakraka (circular logic)

* recursion in AI

* A[right arrow]B[right arrow]C[right arrow]A

* anavastha (regressus ad infinitum or transitivity)

* A[right arrow]B[right arrow]C[right arrow]D ...

* tadanya-budhiat-prasanga (reduction ad absurdum)

* proves a proposition by negating the contradiction of the proposition

* A[right arrow]-( -A ) or + [right arrow] -(-) viparyaya (error)

* manifestation of a real object in another * positive misconception, incomplete knowledge, gaps in knowledge * AI explanation * can be compared to bugs and inconsistencies

5.1.2. pramana--the valid knowledge

It is defined as true knowledge of things which is free from doubts or illusion. There are four means to true knowledge--

* pratyaksa (perception)

* anumana (inference)

* upamana (analogy)

* sabda (verbal testimony) But not all the schools of Indian thought recognize all the four means to true knowledge.

* Caravakas accept

* pratyaksa (perception) only

* Vaisesikas and Buddhists accept

* pratyaksa (perception)

* anumana (inference)

* Samkhya & Yoga recognize

* pratyaksa (perception)

* anumana (inference)

* sabda (verbal testimony)

* Nyaya recognizes

* all four

Some other Indian schools use arthpatti, anupalabdhi, sa-bhava, aitihya as other means of knowledge.

The following section discusses the means to valid knowledge and compares them with AI pratyaksa (perception)

* Five senses--sight, smell, taste, hearing, touch (mind [right arrow] sixth sense for cognition of feelings like pleasure, pain etc)

* six ways of perception for six kinds of objects

* substance, quality, universal quality, sound, sound-ness, nonexistence

* AI [right arrow] icon, schema and percept

* icon [right arrow] temporary record of a sensory input that the brain keeps

* schema [right arrow] blue print of the mental model

* percept [right arrow] assembling units which are to be assembled according to the schema to create a mental model

* Sowa (1984) calls the search mechanism "associative comparator" which must have the following characteristics:

* associative retrieval

* brain is like a computer, retrieves data by an address in the storage by matching the best pattern

* top-down match

* perception searches for percepts that match the overall pattern of an icon

* stimulus constancy

* stimuli from same external object are recognized as equivalent despite varying size, brightness etc

* distributed storage

* images are not stored at a specific point, but distributed anumana (inference)

* Gautama presupposes perception

* Jayanta gives five processes of inferring

* Perception of reason

* Remembrance of the universal concomitance

* Judgment that the subject of inference contains sense which, is concomitant with the object

* Knowledge of the consequence, and

* Judgment that the consequence is worthy of being accepted or rejected

* AI

* inference is very important and strong point of predicate calculus

* axioms [right arrow] theorems

* new facts can be deduced from axioms using rules of inference similar to ones given by Gautama

* Gautama gives 3 ways to infer:

* psrvavat (deduction)

* cause [right arrow] effect.

* cloud [right arrow] rain

* "Gauri eats grass" from "Gauri is a cow" and "all cows eat grass" --

(inst gauri cow)

(forall X (if (inst X cow) (food X grass)))

(food gauri grass)

* sesavat (abduction)

* process [right arrow] explanations

* effect [right arrow] cause

* river is swollen [right arrow] there could have been rain

* Abduction has the following paradigms

From: b

(if a b)

Infer: a

* unlike deduction, abduction is not a legal interference. It can lead to false conclusions--

from: (feels nervous Ira)

(forall (X) (if (is sick X)

(feels nervous X)))

infer: (is sick Ira)

* samanyatod (induction)

* infers consequent from the antecedent, which is neither cause nor effect

* while induction can take several forms, the most common is

from: (P a), (P b),...

infer: (forall (X) (P X))

* although this is not a sound inference, induction like abduction, is very useful in everyday life where it is more commonly known as learning

* for example, if we see a lot of leaves of green color, we might infer that all leaves are of green color

from: (if(inst leaf-1 leaf) (colour leaf_1 green))

(if(inst leaf-2 leaf) (colour leaf_2 green))

Infer: (forall (X) (if(inst X leaf) (colour X green)))

* Vacaspati divides inference into vita and avita (Bijalwan 1987). The former is based on universal agreement in presence. As an example, we can infer

from: whatever has smoke has fire the hill has smoke

infer: the hill has fire

* avita is based on universal agreement in absence

from: whatever is non-different from other objects has no

smell. Earth has smell

infer: Earth is different from others.

* Prasastapada has svarthanumana and pararthanumana as two types of inference.

* Svarthanumana

* the premises are known from our own experience

* pararthanumana

* they are discovered by one man and imparted into another through language.

* therefore, there is greater chance of error in pararthanumana which is based on words. svarthanumana is a mental process which is divided into data and samanyatodata by Prasastapada (Bijalwan 1987).

* Udyotakara gives kevalanvayin (K), kevalavyatirekin (KV), and amayavyatirekin (AV) as three types of inference. If a middle term is only positively related to the major term it is called K. Here. The reason exists in the subject and similar instances, and is devoid of dissimilar instances. For example--we can infer "pot is nameable" from "pot is knowable" and "all knowable things are nameable"--

(property X nameable)))

(if (inst pot knowable)

(property pot nameable))

(property pot nameable)

In the case of KV, the middle term is negatively related to the major term. Reason exists in the subject, but not in dissimilar instances. For example--

from: (no (not P) is (M))

S is M

infer: S is P

When the middle is positively as well as negatively related to the major term, and reason exists in the entire subject, and in all similar instances but does not exist in dissimilar instances, the case is that of AV, as "sound is non-eternal because it is produced like a jar". The inference of Fire from smoke is also of this kind. Jayanta does not however, accept this division of inference in to K, KV and AV by Udyotakara (Bijalwan 1987). upamana (analogy)

* upamana is the means by which we gain the knowledge of a previously unknown object on the basis of its similarity to another object previously well known

* Mimansists and the Vedantists also accept analogy as an independent source of knowledge

* AI

* analogy is too slippery to be seen as just another road to induction

* often, analogy matches are so slipshod that they merely suggest solutions to the problem of knowledge acquisition in general, and for inductive learning in particular

* analogy is useful because it serves as initial framework for developing the new concepts sabda (verbal testimony)

* Leaving aside the Carvakas, Vaisesikas and the Buddhists, all the other systems of Indian philosophy accept sabda as a distinct source of knowledge, but they also differ with one another with respect to its nature forms and a number of other aspects.

* Jayanta (Nyayamanjari)

* contains a long discussion on this problem

* justices the acceptance of sabda as a distinct means of knowledge

* gives serious thought to the philosophy of language

* tries to explain the views of his predecessors and evaluates

* the arguments of prominent scholars

* his account of sabda is as much relevant to grammar, rhetoric, and linguistics as it is to logic

* convincingly proves that words do not exist before their production

* also refutes the sphota theory of grammarians and argues that the relationships between words and meaning is conventional

* AI

* sabda can be compared to the concept of formal string and language.

* language is a set of strings made from some alphabet

if a fixed alphabet = E, then the language formed by

E is called [E.sup.*]

ifE= {a} then

[E.sup.*] = {e. a. aa. aaa. ...}


The main goal of this paper was to compare and contrast the NN system with the AI system. As has been established, the NN system has many parallels in terms of treating valid and invalid forms of knowledge with those of the AI which uses other techniques as well besides logic and inferencing. These techniques, for example, semantic networks, conceptual graphs, frames, rules based systems and procedures' suggest practical ways to represent knowledge unlike the NN and even the western schools which put too much emphasis on Fixing the identity of the locus of knowledge acquisition. The latter probably is not so relevant for AI at present. Nonetheless, there is a need to further delve deep into the some of the Indian systems including the NN system and others like Buddhist logic to arrive at suitable models of human cognition which can be useful for AI.


1. Distante, Patrick. 2003, History of Western Philosophy, Summary Outline-Epistemology. Available online: <>

2. Bijalwan, C. D. 1987. Indian Theory of Knowledge (based upon Jayanta's NyOya MaDjarx). New Delhi: Heritage Publishers.

3. Charniak, Eugene & McDermott, Drew. 1985. Introduction to Artificial Intelligence. Massachusetts, California, England, Canada: Addison-Wesley Publishing Company

4. McDermott, Drew & Doyle, Jon. 1980. Non Monotonic Logic I, Artificial Intelligence, 13.

5. Sowa, J. F. 1984. Conceptual Structures: Information Processing in Mind and Machine. London: Addison-Wesley Publishing Company.

6. Singh, G. V. & Jha, G. N. 1994. Indian theory of knowledge: An artificial intelligence perspective. Paper Presented at the national seminar on Inference Mechanisms in Shastras and Computer Science, ASR, Melcote, 1994

7. Hopcroft, John E., & Lipman, Jeffery D. 1979. Introduction to Automata Theory, Languages and Computation. Addison-Wesley Publishing Company Inc.



Jawaharlal Nehru University, New Delhi
COPYRIGHT 2007 Bahri Publications
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2007 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Jha, Girish Nath
Publication:Creative Forum
Geographic Code:9INDI
Date:Jan 1, 2007
Previous Article:The fable as narrative in the Indian tradition.
Next Article:Flavours of flowers: a rasaesthetics of Karnad's "Flowers".

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters