Potential Barriers and Challenges in AI

17.2 Potential Barriers and Challenges in AI

In Sect. 15.2 we have introduced a psychological definition of intelligence as a set of abilities which allow one firstly, to adapt to a changing environment, and secondly,

a cognitive activity consisting of creating and operating abstract structures. After 27 Daniel Clement Dennett III—a professor of philosophy at Tufts University. His work concerns

philosophy of mind and philosophy of science. He was a student of G. Ryle and W.V.O. Quine. 28 For example, if a fish moves its fins, then it swims; if a thermometer senses that it is too cold,

then a thermostat turns up the heat. 29 Of course, according to Searle, Dennett does not make a distinction between as-if intentionality

and intrinsic intentionality. 30 Patricia Smith Churchland—a professor of philosophy at the University of California, San Diego

and University of Manitoba. Her work concerns philosophy of mind, neurophilosophy, and medical ethics.

31 Paul M. Churchland—a professor of philosophy at the University of California, San Diego and University of Manitoba. His work concerns philosophy of mind, neurophilosophy, and epistemology.

32 Hans Moravec—a researcher at Carnegie Mellon University. In 1980 he constructed a TV- equipped robot at Stanford University. He was a co-founder of Seegrid Corporation, which is a

company developing autonomous robots. 33 Raymond “Ray” Kurzweil—an inventor and a futurist. A specialist in computer recognition of

characters and speech.

17.2 Potential Barriers and Challenges in AI 241 the analysis of AI achievements made in Chap. 16 we can conclude that potential

barriers to AI development concern the second component of this definition. We will try to identify these barriers on the basis of the classification of cognitive operations introduced by St. Thomas Aquinas, because in our opinion it specifies the essence of generic cognitive processes adequately. Let us recall that he distinguished three acts of the intellect, namely concept comprehension, pronouncing a judgment, and reasoning.

Let us begin with the third cognitive operation, because the greatest achievements in AI have been obtained in this area. As we have discussed in Chap. 16 , reasoning is defined as proceeding from one proposition to another according to reliable rules of deduction. In the first half of the twentieth century sound theoretical foundations and effective methods of deductive reasoning were developed in mathematical logic. These methods are used in Artificial Intelligence successfully. Since a simulation of human reasoning should be performed according to logical principles, we use them for designing AI systems.

In the case of a simulation of concept comprehension research results are not so impressive. The standard Aristotelian approach to concept definition, which consists of giving its nearest genus and its specific difference, is used in formal sciences (e.g., in mathematics) successfully, but it is not so effective in other sciences, and it is usually inadequate in everyday life. There are two reasons for the difficulty of applying this approach in AI. Firstly, the Aristotelian rule is a very general principle. Therefore, defining an effective method (algorithm) on the basis of such a general principle is troublesome. Secondly, the Aristotelian creation of concepts by abstract- ing is based on the assumption of the existence of crisp categories. However, modern psycholinguistics claims that categories are of a fuzzy and radial nature, as we have

discussed when presenting Lakoff cognitive linguistics in Chap. 1 . Consequently, a process of abstracting is treated as an intrinsic intellectual process of comprehending the heart of the matter. However, modern science does not answer the question: How does a concept comprehension process, interpreted in such a way, proceed?

Sometimes, concept comprehension is considered equivalent to cluster analysis. This is a remarkable simplification of the problem. Let us notice that in the case of a concept, two its aspects are distinguished, namely its intension, which is the internal content, i.e., the set of properties that characterize objects falling within this concept, and its extension, which defines its range of applicability by designating objects falling within the concept. In cluster analysis there is a designer of the system, who has to define the feature space (the intensional aspect). The system only groups objects in clusters (the extensional aspect). We could talk about a system which comprehends concepts if it generates a feature space on the basis of observation of example objects.

The process of pronouncing a judgment is a generic cognitive process, which is little understood in psychology and philosophy. In order to discuss the possibility of a simulation of this process in AI we use the Kantian taxonomy of propositions, which

has been discussed in Chap. 15 . Simulation of a generation of analytic propositions a priori is performed in AI systems. Let us recall that such propositions concern knowl- edge already existing in our minds. In AI systems in the case of such propositions

242 17 Prospects of Artificial Intelligence we refer to a knowledge base directly, e.g., we find the correct part of a semantic

network, or we derive a required proposition with the resolution method. Synthetic propositions, which expand our knowledge, are divided into two groups. Synthetic propositions a posteriori are derived on the basis of experience gained. In AI such propositions are obtained by generalized learning, which has been discussed

in Sect. 16.8 . Unsupervised learning of neural networks and cluster analysis are the best examples of such learning. Although, as we have mentioned in Sect. 16.8 , there are a lot of open problems in this area, we use here the paradigm of inductive reasoning used successfully in empirical sciences.

Unfortunately, we are still unable to simulate the process of generating math- ematical theorems, which is a fundamental process of mathematical development. According to I. Kant, such theorems correspond to synthetic propositions a priori. Let us analyze this problem in a more detailed way.

Mathematical theories are axiomatic-deductive systems. Firstly, basic notions and axioms 34 are defined. Then, a theory is developed by deductive reasoning, 35 which is based on the modus ponendo ponens rule. This rule is interpreted in the following way:

If the expression: I f A , t hen B is true

and the expression: A is true ,

then the expression: B is true as well .

When we develop axiomatic-deductive systems, we can apply this rule in two ways, namely as a progressive deduction or a regressive deduction [30]. In a progressive deduction we start from a true premise and we try to infer a conclusion. Thus, such a process is a kind of symbolic computing. This computing consists of manipulating symbolic expressions in order to generate new expressions.

The system Logic Theorist, which has been presented in Chap. 1 , is based on this method of deductive reasoning. On the other hand, in the case of a regressive deduction, we first formulate a conclusion and then we try to justify it via pointing out expressions of the system which can be used to derive this conclusion.

A remarkable expansion of axiomatic-deductive systems is obtained with the help of regressive deduction. Important research results in mathematics have been achieved in this way [30, 224]. Let us notice that formulating a conclusion, whose truth has not been proved at the moment of formulation is a crucial moment in this method. And again, modern science does not answer the question: How does the process of formulating such conclusions proceed? This phenomenon is usually described with such terms as insight, inspiration, or intuition [224]. Of course, such

a description does not allow us to define algorithms which simulate this cognitive process.

34 Axioms are propositions that are assumed to be true. 35 Concepts related to deductive reasoning are contained in Appendix F.2.

17.2 Potential Barriers and Challenges in AI 243 Two problems identified above, which are fundamental barriers to AI develop-

ment, result in more specific key problems that have been discussed in Chap. 16 . The fact that we do not know the mechanisms of concept comprehension results in difficulty with developing satisfactory methods of automatic generation of ontolo- gies in the area of knowledge representation and learning, automatic construction of abstract models of problems in the area of problem solving, and semantic analysis in Natural Language Processing.

The lack of models which describe the process of pronouncing a judgment is the main barrier in the areas of planning, automatic learning (the problem of formulating hypotheses), social intelligence, and creativity.

These barriers should not be used as an argument against the possibility of the development of intelligent systems in the future. They constitute, in the author’s opinion, the main challenge for research in Artificial Intelligence.

Dokumen yang terkait

Hubungan pH dan Viskositas Saliva terhadap Indeks DMF-T pada Siswa-siswi Sekolah Dasar Baletbaru I dan Baletbaru II Sukowono Jember (Relationship between Salivary pH and Viscosity to DMF-T Index of Pupils in Baletbaru I and Baletbaru II Elementary School)

0 46 5

Institutional Change and its Effect to Performance of Water Usage Assocition in Irrigation Water Managements

0 21 7

The Effectiveness of Computer-Assisted Language Learning in Teaching Past Tense to the Tenth Grade Students of SMAN 5 Tangerang Selatan

4 116 138

the Effectiveness of songs to increase students' vocabuloary at second grade students' of SMP Al Huda JAkarta

3 29 100

The effectiveness of classroom debate to improve students' speaking skilll (a quasi-experimental study at the elevent year student of SMAN 3 south Tangerang)

1 33 122

Kerjasama ASEAN-China melalui ASEAN-China cooperative response to dangerous drugs (ACCORD) dalam menanggulangi perdagangan di Segitiga Emas

2 36 164

The Effect of 95% Ethanol Extract of Javanese Long Pepper (Piper retrofractum Vahl.) to Total Cholesterol and Triglyceride Levels in Male Sprague Dawley Rats (Rattus novergicus) Administrated by High Fat Diet

2 21 50

Factors Related to Somatosensory Amplification of Patients with Epigas- tric Pain

0 0 15

The Concept and Value of the Teaching of Karma Yoga According to the Bhagavadgita Book

0 0 9

Pemanfaatan Permainan Tradisional sebagai Media Pembelajaran Anak Usia Dini untuk Mengembangkan Aspek Moral dan Bahasa Anak Utilization of Traditional Games as Media Learning Early Childhood to Develop Aspects of Moral and Language Children Irfan Haris

0 0 11