Mathesis Universalis No.4 - Autumn 1997
When using any part of this text - by Witold Marciszewski - refer, please, to the original URL.
Intelligent Beliefs - a Challenge to AI
Introduction to this Issue
"AI is a field of science and engineering concerned with the computational understanding of what is commonly called intelligent behavior, and with the creation of artifacts that exhibit such behavior." So Stuart C. Shapiro put in a nutshell the current AI notion in an entry of Encyclopedia of Artificial Intelligence (Second Edition, John Wiley & Sons, Inc., New York, 1992, 54-57).The author continues as follows (Sec. 1.2):"By `computational understanding' is meant a model that is expressed as a procedure that is [...] implementable on a computer." This definition appears in the context of the goal of Computational Philosophy which differs from Computational Psychology by the feature of not being restricted to the algorithms and data structures that the human mind actually does. The model to be implemented may happen to perform some tasks better than any people would. Furthermore, there are problem solving procedures called heuristics which fail to be algorithms and thereby are closer to the human cognitive activity as studied by philosophers; inasamuch as they are implementable on a computer they contribute to computational understanding of intelligent cognitive behaviour.The above statements (being paraphrases of Shapiro's text) deserve philosophers' attention for two reasons: (i) they hint at the typical philosophical entreprise of discovering cognitive procedures more rational than those being actually in use; (ii) they are written by the author who not only represents an extensive encyclopedical knowledge of AI but, moreover, belongs to leading figures in the team to carry out a huge project of AI research - SNePS, on semantical networks - whose headquarters are at the State University of New York in Buffalo.The research groups like that seem a proper addressee of philosophical questions concerning intelligent beliefs, to wit those to be answered with some heuristics in Computational Philosophy. Philosophical questions derive from the following design for Computational Psychology, as stated by Shapiro (op.cit.), provided the indicative mood gets replaced by what can be called a normative mood [as hinted below in brackets with a change of colour].
"For the goal of Computational Psychology [Philosophy] it is important that the algorithm [heuristic] expressed by the program be the same algorithm [heuristic] that people actually [should] use, and that the data structures used by the program be the same data structures [which should be] used by the human mind."Algorithms, heuristics and data structures which should be used by the human mind to attain intelligent BELIEFS, or (if you prefer other terms), rational or reasonable beliefs, is what absorbed philosophers since earliest times. Thus the criteria of intelligent belief and the methods to attain them, as proposed by philosophers, create a challenge to be faced by AI reserchers. They are expected to devise programs and heuristics to match both the psychological reality and the philosophical ideal.
In the search for a general criterion of resonabless, covering both scientific and other issues, a favoured criterion was that of clearness. According to some classics, as Descartes, it forms both the necessary and the sufficient condition of reasonably and unmistakenly accepting anything for true. According to other ones, clearnes of the ideas involved in a belief constitutes a necesssary condition of its reasonable acceptance; it is C.S. Peirce who is here chosen as a spokesman for that approach.
Three conceptions of clearness are compared in this issue.
- Descartes: intellectual clearness - known to mathematicians; these clearly "see" what is referred to by their axioms, as well as relations being grasped in their inferences.
- Peirce: empirical clearness - known to those who define theoretical concepts in terms of sensible properties; the latter grant their clearness to the former.
- James, William: practical clearness - known to the theorists of decision making. When a decision is a "leap in darkness" (as James puts it to express the lack of any cognitive reasons), but it is clear that the decision must be made to avoid a greater evil (to result from indecision), then we deal with that particular kind of clearness which may be called practical, or pragmatic.
Items 1 and 2 consist of selected quotations mixed with comments. Item 3 consists of the full text of James' pragmatist manifesto The Will to Believe, and a comment being a separate file.Obviously, even if clearness of the concepts involved is a necessary condition for an intelligent belief, it is hardly sufficient. Any intelligent belief should be non-trivial and well-founded, while clearness happens to be necessary to find a proper method of foundation; this is evident in the case of empirical clearness as necessary for testing hypotheses. As to mathematical propositions, Descartes' claim that it is also suffiient deserves a careful reflexion.Thus, there are numerous important cases in which clearness is a necessary prerequisite of intelligent belief. In all such cases the clearness is meant as an attribute of consciousness. And now, there is the question: whether the artifacts enjoying intelligent beliefs should be endowed with artificial consciousness, or will the AI theorists prove that consciousness is like a superfluous residual organ which at a higher level of evolution is to be dispensed with? This is a challenge to be met. Another one is concerned with the following feature of some beliefs.
The lack of cognitive clearness, and the necessity of relying on practical considerations alone is for James a tragic feature of our human existence which, anyway, we should face with our logic of decision. In this outlook, James is much like Pascal. However, there is a difference between them which proves vital for understanding how some intelligent beliefs can be produced.
There is no separate file concerning Pascal in this issue. Instead, the relevant texts of him are quoted and discussed within the item:
4. Intelligent Beliefs which Lack Clearness. On Pascal's Esprit de Finesse and Demon Procedures.Blaise Pascal (1623-1662), like other eminent minds of his time, was much impressed by the idea of automaton. Moreover, he was a great champion of that idea, as the first (practically) constructor of calculating machine. There is a visible trace of that in his philosophy, including the philosophy of belief, in particular, religious faith. Under the heading esprit de finesse he considered how the mind can handle immense complexity of information items, and suggested explanation in terms of non-conscious information processing, lacking the feature of clear consciousness (as postulated by Descartes).
His vision can be nowadays rendered in terms of what (in Unix and AI terminology) one calls "if-needed-procedures", or "demon procedures". Such procedures can result in intelligent beliefs though the believer is hardly aware of why he believes and which processes are responsible for the belief in question. Such a non-conscious processing seems to most efficiently handle the enormous informational complexity (in this sense, the lack of clearnes, being a price for the efficiency, has a constructive role). A reconstruction of such processes in artificial intelligent agents yields another challenge to AI research.
The third and the greatest challenge is as follows. With natural intelligent agents, the highest creativity they possess seems to derive from a highly involved interplay between the clear conscious beliefs and those owed to demons' activity, hence lacking the clearness of conscious processes. Such interactions should be reproduced in artificial intelligent agents, are they to match humans in creative powers. One may have philosophical reasons to believe or may have reasons to disbelieve in chances of such a project. Anyway, if engineers materialize it, this will end the philosophical dispute.
Do they? This does not seem a safe bet, but the question remains open.