DISTINCTIVE FEATURES OF INTELLIGENCE
Ph.D. in Technical Michael Kozlov
Expert of Institute integration and professional adaptation, Netanya (Israel)
E-mail: 19mike19k@gmail.com tel.(972)527 052 460
There is nothing in the mind that is not first in the senses, exceptthe mind.
Leibniz
SUMMARY:
The article examines features inherent to a beings endowed with intelligence and formulated the basic necessary properties of intelligence that enables the assessment and classification of the analyzed technical systems as an intellectual and helping in the construction of complex technical systems. The presence of the subject-oriented emotional matrices at intelligent agents allows us to speak about the subjective emotional intelligence.
KEY WORDS: Artificial Intelligence, intelligent agent, matrix of artificial emotions, subject-oriented quantities of information.
INTRODUCTION
Complicated technical systems, working in conditions of uncertainty, are trying to endow by certain artificial intelligence (AI) to ensure the optimal functioning of a number of parameters. Such as adaptation to the environment, robustness and security.
In order to more clearly understand the principles of construction of different systems with AI, such as intelligent agents [1] or a system of Data Mining [2], need to know, what the necessary properties must have an intelligent system.
Given elaborated the evolution to perfection in living organisms mechanism to respond adequately to the situation and the relative newness endowing the elements of intelligence a technical systems usually use to create the AI simulated of intelligence beings. Therefore, consider how suitable the existing definitions of a living intelligence to construct systems with AI.
From the existing set of definitions of intelligence can distinguish the following:
“ Human intelligence, mental quality that consists of the abilities to learn from experience, adapt to new situations, understand and handle abstract concepts, and use knowledge to manipulate one’s environment” [3];
“ … the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, quickly to learn and learn from experience” [4];
“ Intelligence is the resultant of the processes of acquiring, storing in memory, retrieving,
combining, comparing, and using in new contexts information and conceptual skills; it is an abstraction” [5].
A prominent researcher in the field of AI Marvin Lee Minsky [6] considers the concept of intelligence as something relative, allowing to the subject to solve complex problems that are not yet understood.
Reflecting the views of many researchers presented definitions cover the basic properties of intelligence. However, they are not a rosary enough and and not very suitable for the formation of the requirements to be met by the system with AI. Proceeding from this, we formulate the necessary basic properties of intelligence allowing to classify the system as a intellectual and use them in the construction and evaluation of complex technical systems.
FEATURES OF INTELLIGENT SYSTEMS
Studies of brain functioning showed the important role of emotions in human decision-making [7]. Emotions in humans focused on the survival of the individual and the species. The state of emotional matrix of the subject will be determined by the set of potential opportunities, the environment and its objectives [8]. In society the possibilities and the roles of individual subjects are different and therefore will have different setting their emotional matrices. Possessing by emotional settings, people complement each other in solving problems, improving thus the reliability at the decision-making.
The signals coming from the external environment to the living organism, endowed with intelligence, pass emotional filtering and evaluation, forming patterns of subjective information. Such perception by consciousness of the subject (apperception) [9] occurs taking into account the previously received and the ordered the knowledge and skills, as well as data about the internal state of the organism and its resources.
On the basis of the subject-oriented apperception produced decision, which may appear in an external response. Such a reaction can be unpredictable. So we can talk about the subjective emotional intelligence and estimate given in [10] the humorous definition: «Intelligence — this is what we have, and there is no at other people.»
On the necessity of emotion for systems with AI Marvin Minsky [6] writes “The question is not whether intelligent machines can have
emotions, but whether machines can be intelligent without any emotions”. In accordance with this, in [8] considered the advisability endowing each intelligent agent (IA) in multi-agent systems by individual adaptable matrix of artificial emotions, depending on the agent’s target role in a team. On the basis of data research the centers of emotions in humans, conducted by Antonio Damasio [11] and other cognitive psychologists, in [8] for basic artificial emotions are encouraged to use the six analogues of emotions: fear, pleasure, disgust, suffering, admiration, anxiety.
Emotions are important in making decision at a rapid, automatic, unconscious level. And from the adequacy of emotions the situation and richness of their tones will depend the optimal response. Each of emotions plays a special role. Thus fear and pleasure can be attributed to the dynamic fast emotions with respect to the external environment. Formation disgust to anything can be delayed in time and made on the basis of extracting information from temporary memory. Suffering related with the assessment of the internal state, but in a society on its base can be formed an external emotion compassion for someone. Admiration is a external emotion that stimulates imitation and is an important factor in learning. On its basis, can be formed an emotion of dominance. If several repetitions of negative stimuli generated anxiety. This emotion transforms organism into a state of high alert and, at the same time, significantly limits the number of options when making decision (MD).
In the presence at IA an individual matrix of artificial emotions, constructed on the basis of the base matrix, with accumulation the experience of interaction with the environment and with other subjects his individual matrix will gradually to pass buildup and evolution. Such a matrix of emotions will help to ensure IA by necessary energy resources, to avoid dangerous situations, to stimulate the achievement of the objectives, monitor the status of individual parts and the software. Emotion «admiration» by stimulating imitation, in combination with the «pleasure» provides rapid learning to new skills. Imitation emotions of anxiety can be used when creating technical systems, working conditions which suggest the occurrence of situations in which necessity of operative, faster response allows to reduce range of analysis for MD and accordingly to reduce the reliability of the evaluation of the expected results.
Using a matrix of emotions greatly expand and improve the possibility of IA and allow to operate successfully under conditions of uncertainty. Resorting to the judgment of Goethe about whole and its parts [12], an individual matrix of emotions can be compared with the soul of a living subject, with withdrawal of which disappears the connection of its parts.
Knowing the individual matrices of emotions can to predict the response of the subject to external stimuli. Such works are carried out. For example, using the computational cognitive architecture CLARION [13] the cognitive psychologists conduct studies of human behavior in various situations, including the behavior of the group of people without resorting to time-consuming and not always permitted experiences a direct on the living entities, and using computer simulations. This procedure can give even the best result due to a wide range of metered multifactor impacts in short period of time.
To simplify a solved problem in such studies can use the methods of reduction of a system of differential equations based on Tikhonov’s theorem [14]. Their use allows, with the existing differences in time scales at various factors, for the certain time interval reduce the number of variables by converting slowly changing factors in the constant parameters.
Note, however, is not always the prediction of response of emotional intelligence, which depends on the internal state and surrounding conditions of physical and social environment, possibly using statistical regression models.
In living beings, with the ability to generate and store knowledges, quantities of information supplied to them with patterns of signals will depend on the knowledge already available in their minds about similar patterns. To estimate received quantities of information I we use the expression given in [15]:
I = H1 – H2, (1)
where H1 — a priori uncertainty of knowledge on this issue; H2 — a posteriori uncertainty of knowledge, after receiving the signal.
With increasing knowledge about received patterns the quantities of information will tend to zero, which corresponds to the negentropy principle of information [16].
For evaluate the information received can use and the concept of the value of information [17], considered as the increment of the probability of reaching the target as a result of use of this information. Meaning value of information V is given by:
V = log2P1 – log2P0 , (2)
where P1 — probability of achieving the goal, after receiving information; P0 — probability of achieving the goal before receiving information.
In humans, as in higher animals, signal processing takes place in parallel, on an unconscious level, related with work of implicit (unperceived) memory and on a conscious level, related with work of explicit (of conscious) memory. Response to input signal comes from the regions of the brain associated with these levels, by the two neural pathways in the amygdala of the limbic system, where formed signal of reaction [18]. Information processing at an unconscious level occurs much faster and the difference between implicit and explicit responses may be more than 500 ms [19]. According to some observations, in higher animals, with faster implicit reaction a explicit response can be much longer in time than in humans.
If the brain does not take a decision on the need for an explicit response, then control signal is generated based on the implicit faster reaction. In experiments with conflicting stimuli were registered using by the subjects the implicit skill in situations where conflict stimuli was up to 25%, and when their number was 75% the subjects began to use only explicit knowledge [20]. Most of the time, human behavior occurs in an automatic unconscious level.
Data from the short-term memory (STM) come to a temporary explicit memory. After forming there the stable models of subjective knowledge (MSK), they will be recorded in long-term memory (LTM). Due to the complexity of such knowledge they have already worn the subconscious nature, do not can be verbalized [21] and can be regarded as skills. Given the complexity of of awareness the data stored in the LTM, and the speed of access to it, in [22] LTM attributed to the store of implicit memory. The duration of the formation of MSK up to transfer them to the LTM can be quite large, it reaches at person several years [23].
The figure shows the formation the models of subjective decisions (MSD) from the incoming signal input patterns (IP).
From the input patterns on the upper path shown in the figure, are formed MSK that are used to generate decision. With the accumulation of knowledge will be a transition of explicit knowledge in implicit skills and so formed the models of subjective skills (MSS). In the future for similar patterns decision-making procedure will be carried out by the lower much more rapid way, based on MSS. So is carried out the evolutionary chain of information perception as the Unconscious — Conscious — Subconscious. Here, in particular, we can speak of the primacy of intelligence with respect to developed at its basis of the newly acquired instinct.
Based on the foregoing, intelligent systems, for which is important operativeness and situational reliability in the development and decision-making, can operate follows. When receive a signal the formation and decision-making is done by two ways at the same time on implicitly and explicitly. At the MD system the first comes by the fastest implicit way one of the options stereotypical solutions and at high reliability of the assessment of the expected results in relation to the signal in this embodiment is formed the signal of response. When low estimation of reliability the result implicit way is blocked and used variants of solutions going by a longer an explicit way. If again evaluation of the reliability result is insufficient, then turned on the mechanism of experimentation based on samples and evaluation of results of experiments. With its help iteratively produced an adequate solution to the situation, using both explicit and implicit mechanisms.
When analyzing intelligent systems should take into account as the number of patterns simultaneously processed, well as the processing speed. George Miller has defined limit of information throughput of STM a person by number 7 ± 2 chunks («magic number 7 ± 2») [24], where the chunk a semantic image (pattern), for example, a letter or a whole phrase. Based on further research Herbert Simon [25] volume of STM was determined in the range from five to seven chunks. The duration of storage of the pattern in the STM is about 12c and playing time 35ms. [26]. Similar restrictions on simultaneously processed patterns have and higher animals. Research by visually counting numbers of items in the visual field showed that such a calculation is very fast and accurate for numbers of items from one to four, but slowly and with errors for greater numbers of items [27]. For process of fast determine numbers of items in [28] was introduced the concept of «subitizing», and for a range of fast counted objects «subitizing range.» For numbers of items that fall within the subitizing range, time spent on counts of items is in the range of 40-100 ms on item, and for the numbers of items on the numbers of more than four — 250-350 ms.
For IA limits on information processing will be determined as their technical capabilities well as solved problems. At the same possible decreasing the processing speed by increasing the number of patterns simultaneously processed, and vice versa. Experiments showed by Robert Sternberg [26] this within certain limits is inherent to man.
Short-term (working) memory involved not only in the processing of input information, but also to further streamline formed knowledge and decision-making. Such a procedure hampered, with the active human response to external signals because of the limited capacity of the STM. However, nature has found the perfect solution to improve efficiency of the intellectual apparatus for the implementation of transcendental apperception by Kant [29]. During sleep, or in a period of prolonged rest in the absence of continuously incoming stimuli to which to pay attention, the brain goes into default mode [30], in which there is intensive processing and ordering previously received information. Formation of new, more complex MSK is made within a several hundred low-frequency cycles of the default mode network by internal the feedback circuits coming from various sections of memory on the STM. Each time at entry MSK in STM produced their new emotional assessment, performing as if their new awareness. In the process of this generation MSK are increased the weights models of knowledge more reliable for the subject and other MSK are suppressed [22]. Consolidation MSK will occur in STM is primarily on base of data stored in the temporary memory, with connectivity a data of LTM and constant innate memory.
To consolidate MSK requires emotional assessment of event, the default mode of brain function, and the time on the formation of MSK in this state. This may explain the results of investigations emotionally shocking effects on humans [31], which showed selective and retroactive enhancement of memory after 6 hours and more about event what was paired with shock.
BASIC FEATURES OF INTELLIGENCE
Based on the above, we formulate for living and artificial subjects necessary features, possessing which they can be attributed to the having intelligence.
Can be attributed to intelligent the beings or artificial self-organizing adaptive systems with the following properties:
- subjectively to perceive incoming signals, based on an assessment of environmental and their resource capabilities, producing of selection, sampling and subject-oriented interpretation of incoming information;
- with each new arrival of the same type of pattern decrease the subject-oriented quantities of information necessary for decision-making to respond to such a pattern;
- to carry out purposeful behavior by simulation analyzed processes, estimation of probability of options development of events and generating an acceptable response for a subject.
Subjective perception (apperception) incoming information related to its emotional assessment plays a crucial role in adaptive selection necessary data for the subject and significantly reduces the volume of their processing. For technical systems such reaction to the input patterns can be considered as anti-spam filtering of information, unimportant or interfere with work this system. Such subject-oriented selection can promote information security IA.
On the basis of the first and second properties of intelligence, using the expression (1), the subject-oriented quantities of information E, extracted from the input signal i +1 can be written as
E(i+1) = N(i) – N(i+1) , (3)
where N(i) — the uncertainty accumulated of the subjective knowledge about the object before admission the signal i+1; N(i+1) — a posteriori uncertainty of the subjective knowledge about the object, after receiving the signal.
Value N(i + 1) can be expressed as a functional of signal processing of the observed object in the form of
N(i+1) = S(i+1)*H, (4)
where S(i + 1) — a dynamic operator of perception information by the subject (the operator of apperception); H — value of the initial information uncertainty inherent in the model of the object. Scale uncertainty H will depend on the information value of the observed object for the subject and its technical capabilities.
The operator of apperception S plays a crucial role in the formation of subjective knowledge. It is dynamic and depends as on the available knowledge well as the state of the emotional matrix a subject. And for different subjects, due to their actions, coloration of received subjective knowledge about the object can vary greatly and even be both positive and negative.
Subject-oriented quantities of information E as the accumulation of subjective knowledge about the object will decrease and, at constant operator of apperception S, with each receipt of a signal from an object the value of E will be closer to zero.
It is interesting to note that one subject, with its inherent operator of apperception S, may superficially perceive information and another with operator of apperception S, with a maximum in the band of interest of the subject, become an expert after some time, has accumulated huge amount of knowledge in a particular area. However, in both cases at multiple perceptions of these subjects of such patterns may be the same final value of the subject-oriented quantities of information E. Thus, the operator of apperception S displays the subjectivity of intelligence.
The third required property provides to intelligent systems the ability to predict developments of events and make optimal decisions. This feature possibly present in the so-called Swarm Intelligence [32], characterizing the collective behavior of representatives of fauna as ants, bees and birds. And we can talk about the presence of intelligence in the association, which is conventionally called of a Flock Emotional Intelligent Agents [33]. This flock of is a hierarchical structure, on the lower level which a lot of IA possessing all features of intelligence, and on the top level is a emergent collective intelligence.
The presence of these three properties allows intelligent systems are optimally adapt to the environment and evolutionally develop. These characteristics of intelligence corresponds to the presence its at many living organisms and limits the class of systems with AI.
CONCLUSIONS
Formulated three main properties of intelligence enables the assessment and classification of the analyzed systems as intelligent. The presence at intelligent agents a subject-oriented matrices of emotions makes it possible to talk about the subjective emotional intelligence. Considered features of endowed with intelligence of living beings, allow aid in the construction of complex technical systems.
REFERENCES
1. Poole D.L., Mackworth A. K., Goebel R. Computational intelligence: a logical approach. Oxford University Press. Oxford, UK, 1998.
2. Witten, I. H. Data mining : practical machine learning tools and techniques.—3rd ed. / Ian H. Witten, Eibe Frank, Mark A. Hall — Morgan Kaufmann, 2011.
3. Robert J. Sternberg. Human intelligence. Encyclopaedia Britannica.
4. Gottfredson Linda S. (1997). «Mainstream Science on Intelligence (editorial)». Intelligence 24 (1): Pp. 13–23.
5. Humphreys L. G. (1979). «The construct of general intelligence». Intelligence 3 (2): Pp. 105–120.
6. Minsky M. The Society of Mind. Simon and Schuster. 1988.
7. Damasio A. Descartes’ Error: Emotion, Reason, and the Human Brain. N.Y. Avon Books. 1994.
8. Kozlov M. Artificial emotions in autonomous multi-agent systems. Collection of articles 19th International Scientific conference «Creative scientists searches Israel and the world today» Ashkelon. International Centre for Research and practice of creativity. 2013. Pp. 64 — 73.
9. Leibniz: Selections. Edited by Philip P. Wiener. New York: Charles Scribner’s Sons, 1982.
10. Solozhenkin V.V. Psychological foundations of medical practice: the textbook for students of higher educational institutions. — Publishing House: Academic Project. M., 2003.
11. A. Damasio. Self Comes to Mind: Constructing the Conscious Brain. Pantheon. 2010.
12. Goethe. Faust. Oxford University Press, USA, 1998.
13. Sun R. (2009). Motivational representations within a computational cognitive architecture. Cognitive Computation, 1(1), Pp. 91–103.
14. A.N. Tikhonov A.B. Vasil’eva, A.G. Sveshnikov. Differential equations — M .: Nauka. 1998.
15. Ornatskii P.P. Theoretical Foundations of Information and measuring equipment. Kiev: Vishcha School, 1983.
16. Brillouin L. Science and information theory. Academic Press Inc. Publishers. New York. 1956.
17. Kharkevich A.A. The value of information // Problems of Cybernetics. — 1960 — Vol. 4.
18. Kandel E. R. In Search of Memory: The Emergence of a New Science of Mind. N. Y.: W. W. Norton & Company, 2007.
19. Shelepin J.E. Localization opponenting decision-making mechanisms in the frontal cortex / J.E. Shelepin, V.A. Fokin, A.K. Harauzov, N. Foreman, S.V. Pronin, O.A. Vakhrameeva, V.N. Chihman // The fifth international conference on cognitive science. Kaliningrad, Russia — June 18–24, 2012 — V. 2 — Pp. 841—842.
20. Moroshkina N.V. The interaction of implicit and explicit knowledge in the process of learning: what kind of knowledge is more important? / N.V. Moroshkina, I.I. Ivanchov // The fifth international conference on cognitive science. Kaliningrad, Russia — June 18–24, 2012 — V. 2 — Pp. 540—542.
21. Kihlstrom J. The Cognitive Unconscious. Science. 1987. – V. 237. – Pp. 1445–1452.
22. Kozlov M. Computer — brain memory model and decision-making for complex systems. Journal Zhitomir State Technological University — 2013. — №2(65). — Pp. 95-105.
23. R.F. Schmidt and G. Thews (eds.), ed. (1989). Human Physiology (second, completely revised edition). Berlin, Heidelberg, New York: Springer-Verlag.
24. George A. Miller. The Magical Number Seven, Plus or Minus Two // The Psychological Review, 1956, vol. 63.- Pp. 81—97.
25. Simon H.A. How big is a chunk. Science. 1974. N 183 — Pp. 482—488.
26. Solso R.L. Cognitive psychology. — 6th ed. — Boston : Allyn and Bacon, 2001.
27. Piazza M., Mechelli A., Butterworth B., Price C.J. Are subitizing and counting implemented as separate or functionally overlapping processes? NeuroImage 15, 435–446 (2002).
28. Kaufman E. L., Lord M. W., Reese T., Volkmann J. 1949. The discrimination of visual number. Am. J. Psychol. 62: 498–525.
29. Kant Immanuel. The critique of pure reason / edited [and translated]by Paul Guyer, Allen W. Wood. Cambridge. Universiy Press, I998.
30. Marcus E. Raichle, Abraham Z. Snyder. A default mode of brain function: A brief history of an evolving idea / NeuroImage 37 (2007). —Pp. 1083— 1090.
31. Dunsmoor J.E., Murty V.P., Davachi L., Phelps E.A. Emotional learning selectively and retroactively strengthens memories for related events. Nature. 2015 Apr 16; 520 (7547). Pp. 345-348.
32. Bullnheimer B., Hartl R.F., Strauss C. A. New Rank-based Version of the Ant System: A Computational Study. Central European Journal for Operations Research and Economics, 1(7), 1999. Pp. 25 – 38.
33. Kozlov M. Artificial emotions in collective artificial intelligence systems. Journal Zhitomir State Technological University — 2012. — №3(62). — Pp. 99-106
Иллюстрация: 5psy.ru
