In the Beginning Was Information (30 page)

The gestural language of wolves:
A wolf is able to communicate fine shades of meaning by means of its eyes, lips, and tail. It can express 11 emotional states, including confidence, threat, uncertainty, depression, a defensive attitude, and subjugation, by different tail positions [F8]. This seems to be a complex system by means of which thousands of states could be encoded, but it lacks the decisive elements of a language, namely the ability to creatively combine single language elements to construct a sentence.

"Speaking birds": Parrots and budgies have the ability to faithfully imitate words and even sentences of human language. They simply reproduce what they hear. When a parrot says, "Polly wants a glass of water," this does not reflect an actual desire. Even when it says, "Good morning," it still is not a real greeting. There is no real meaning behind this "speech" of the bird; it does not speak English or German or even its own language when it makes these sounds. In contrast to the previous examples, this mimicry does not represent a communication system.

A2.3 Does "Artificial Intelligence" Exist?

The concept "artificial intelligence" — abbreviation AI — increasingly appears in catch phrases in many scientific journals as well as in popular literature [N1, S6, W1, and W7]. This falsely used catchword creates the impression that there will be no distinction between human intelligence and that of a computer in the future. The first computers were optimistically called electronic brains. Fortunately, this idea has been discarded, since there is no comparison between the human brain and a computer [G10]. Unfortunately, the American expression "artificial intelligence" was adopted uncritically, and caused more confusion than enlightenment. The question actually involves the following aspects:

– The objective is not (we hope!) to create systems which would artificially improve the intelligence of somebody with insufficient intellect, like one can employ an artificial kidney in the case of poor kidney performance.

– The word "intelligence" does not only refer to the intellect, but also includes the gathering, transmission, and evaluation of messages. An office specializing in intelligence does not deal with intellectual matters, but is concerned with information.

There is no fixed definition of AI, but the envisaged objective could be briefly described as follows: It involves a certain software technology and not, for example, the hardware of a new computer generation or some new kind of computer architecture. It does involve the development of programming systems for dealing with issues like the following:

1. Expert systems:
This involves information systems where questions and answers in some area of specialization can be stored in the form of if-then-else rules; new cases are incorporated as they occur. In medical diagnosis, in the search for mineral resources, and even in income tax consultations, the branching nature of the questions and the material requires special programming. An expert system essentially comprises a database containing single facts as well as the relevant rules and an inference mechanism consisting of a rule interpreter and an applicable strategy. Such systems can only be used in cases where knowledge can be formalized in rules, because programs cannot recognize semantic relationships and cannot link semantic elements.

2. Robotics:
Robots are the next stage of development of partly automated machines and they play an increasingly important role in manufacturing processes. They can gather data from their surroundings by means of sensors for pressure, heat, and light, for example, and evaluate these in terms of stored program values. Then the programmed movements are carried out by manipulative mechanisms. Such systems can achieve impressive effects, as illustrated by the organ-playing robot in Figure 5 (chapter 1). It should, however, be emphasized that robots can do nothing more than that for which they have been programmed; the ingenuity and foresight of the programmer when he models the relevant knowledge is of crucial importance.

Figure 40: "Good morning! I am the new medical director. Now please tell me how you’re doing" (sketched by Carsten Gitt).

Figure 41: Speechless (sketched by Carsten Gitt).

3. Image processing:
This involves the construction or reconstruction of images from data. Highly suitable areas of application are tomography (Greek
tome
= cut,
graphein
= write; in conjunction with x-rays) and the preparation of satellite images. The main concern is to recognize objects by contrasting shades of grey and by applying certain algorithms to pictures. Such images can then be erased, changed, or inserted elsewhere.

Image processing includes methods for the recognition, description, and comparison of sample objects.

4. Speech processing:
This includes:

– the conversion of spoken words into symbols independent of the person who speaks (the development of a "speech typewriter")

– translation from one language to another

– the ability to converse in natural language

In this area, the limitations of AI are particularly obvious, because it is in principle impossible to prepare software that can really understand and process meanings, as would be necessary for exercises requiring lingual discernment (e.g., translating, summarizing texts, and answering questions). Speech recognition and comprehension as well as the creative use of language are essential cognitive faculties of man, and is the prerogative of human intelligence.

Since the 1950s, the use of computers for natural language has been considered a challenge. The first efforts were directed to the automation of translation. For this purpose, a complete dictionary was stored, as well as the equivalents in the target language. Employing simple rules for word positioning and for the construction of sentences, the program then looked for associated concepts. However, all these efforts came to a dead end. The frequently quoted sentence, "The spirit is willing, but the flesh is weak," was translated into Russian and then translated back. The result was "The vodka is strong, but the steak is bad."

Joseph Weizenbaum of the Massachusetts Institute of Technology (Boston) wrote a program called ELIZA in which he ignored linguistic analyses, but employed a fixed answering schema thought to be sufficiently refined. The only result obtained was an understanding of language that was totally unrealistic. This program looked for certain words or word samples in the expressions of the input text, and selected suitable answers from a set of stored sentences or sample sentences. If, for example, a sentence containing the word "mother" is entered, then the system looks for a standard sentence in which this word occurs and responds with: "Tell me more about your mother." Most of the entered words are ignored by the program, but in spite of this, a very extensive background library is required for dealing with the large number of possible sentences that can be formulated by a person to give the impression of conversing with him.

Conclusion: The problem of enabling a computer to deal with natural language is still unsolved and will never be resolved. It is not possible to model a comprehension of semantic categories (e.g., metaphors, idioms, humor, and exaggerations), unmentioned intentions and convictions of the speaker, or emotions and motivations, for a computer. Human abilities like common sense, intelligence, and creativity cannot be simulated mechanically, because the intelligent use of language includes observation, thought, and actions. Even the best programs used for speech processing do not know what they are talking about, in the truest sense of this expression.

Computer programs which would really be able to imitate language comprehension and perform correct translations fail, as far as the following are concerned:

– Comprehension of meaning: A program cannot "understand" semantic relationships, neither can it link them with one another.

– Grammatical analysis: When translating text, a grammatical analysis is required initially. A program which analyzes some text without considering its meaning will not be able to analyze numerous sentences correctly.

– Language usage depends on its context: The meaning of a sentence cannot be found by adding the meanings of single words. What is more, the meaning of single words depends on the context in which they appear.

– Language employs background knowledge: Each and every sentence is rooted in a specific frame of reference and it can often only be understood in terms of a knowledge of this background.

– The richness of a language resides in poetic turns of speech and in its metaphors. These occur much more frequently in everyday speech than we might realize. There are countless sentences for which the meaning cannot be derived from the meanings of the component words.

– Languages are multivocal: In all languages, some words have more than one meaning (e.g., board as a noun can mean a wooden slab, or daily meals, and it also has more than one meaning as a verb), but this phenomenon is not restricted to words only. At all higher information levels (structural, semantic, pragmatic, and apobetic) there are uniquely personal components which cannot be accessed mechanically.

As we have seen, "artificial intelligence" is a misnomer, and, in addition to gradational differences, there are fundamental differences between human intelligence and AI:

– We should distinguish between data and knowledge, between algorithmically conditioned branching of a program and a deliberate decision, between sorting by comparison and associations, between an averaging of values and comprehension of meaning, between a formal decision tree and individual choice, between a sequence of computer operations and creative thought processes, and between the accumulation of data and a learning process. A computer can only do the former; herein lie its strengths, its areas of application, but also its limitations.

– AI could be regarded as a higher level of data processing, but definitely not as the beginning of independent thought on the part of computers. We err greatly when we believe that it would be possible to develop a system which will purportedly be something other than just a computing system. A chess-playing computer cannot think independently; it is no better or no worse than the strategy programmed into it by human intelligence.

– A machine cannot think independently; it can only algorithmically process information formulated and entered beforehand by one or more persons.

The AI question is discussed more fully in [G8]. Systems developed by "AI programmers" will become more meaningful in many areas, but in the light of our knowledge of the information concept, we should always keep in mind that no machine, however well programmed, will ever be able to generate creative information (see chapter 8), because this is fundamentally an intellectual process. We can now define [G8, p 41]:

Definition D7:
The distinctive characteristic of creative information is its novelty, i.e., "to think what nobody else has thought." This aspect can be described with concepts like inventiveness, creativity, originality, unconventionality, innovativeness, advancement, and abundance of ideas. Every piece of creative information represents an intellectual effort and is directly linked to an originator who is a person endowed with a free will and with cognitive abilities.

Appendix A3

Energy

A3.1 Energy, a Fundamental Quantity

The concept of energy (Greek
energeia
= activity) plays such a central role in all of physics as well as in the other natural sciences and in technology, that it is regarded as a fundamental entity, as information is shown to be. In contrast to information, energy belongs to the material world — the lowest level of Figure 14 (chapter 4). Energy appears in numerous forms, many of which can be converted into another form. Many physical processes fundamentally involve nothing else than the conversion of one form of energy into another. The most important forms of energy are: