© V. Protasov

The new method of collective creativity in various areas of human activity is developed and tested. The classical genetic algorithm is put in its basis. The method allows the collective step by step to the best decision. Experimental measurement of intelligence of collective mind based on Isenk tests has shown the significant gain of IQ in comparison with the individual one.

Attempt to describe several results in the present paper is undertaken. This results recently were received optioned by the authors in more advents intellect construction by “less clever” components. Authors opinion the results maybe adequate of principles met systems transitions V. Turchin [1]. The people and elementary computers were used as separated components "mind" of a new kind. The considerable growth of "intelligence" such system was observed in comparison with "intelligence" separated components and in that and other cases. The same rules of interaction between components it were used unite these polar cases.

The given approach can appear interesting in a problem creation artificial intelligence. Method is applicable at designing symbiotic architectures from neuronets and neurocomputers, computers and their networks, group people, at use genetic rules working as a single unit. The authors for strengthening human intelligence [2] and team robots “group intelligence” [3] successfully applied these rules. The similar combination, probably, will result in construction of hierarchical networks with cascade strengthening of "intelligence" at each level, where the communications and rules of interaction between components one level in each hierarchical layer can be submitted by the same image.

Let's consider the effect amplification "intelligence" at the metasystem transition in the elementary case, when the elementary computers act as the components metasystem. In [3] for acceleration acceptance optimum decision at the choice purposes the group of robots, the increase competitive ability group in conditions counteraction contradictory group and at destruction or failures (refusals) separate robots and - or their calculators, a task was put to develop the effective algorithm use collective computing potential all robots.

The task of distribution aims in classical statement looks as following: there is m robots with calculators and n the aims, and each calculator has coordinates all robots and all purposes. It is necessary to construct such a set of pairs: the robot – is the aim, that on each robot at m < n or m = n there is one aim and at m> n on each aim there is one robot, and the sum S distances between pairs should be minimal.

The similar tasks, as experience shows, are enough effectively solved by a genetic method. An elementary, uneconomical on time method looks so. Each calculator belonging to the i-number robot, having all data on a position robots and purposes, builds a population possible decisions a task assignments and, following classical procedure, using the operators of crossover, mutation, estimation and selection, through some iterations finds the quasioptimal solution. As all calculators work by the uniform algorithm and with identical sets initial data, they come to the identical decisions without the interaction with each other and each robot chooses thus its aim. The advantage given method consists in autonomy work of each calculator, but this autonomy is a lack, as there are the obvious over- superfluity resources and at reasonable organization calculations the task can be solved in m time faster.

In [3] the problem of realization distributed calculations was put and solved, at which one supercalculator works as under construction of set of all calculators of group of robots. It was natural to require of such supercalculator the steady work at breakage of separate calculators and the reduction of time of reception decision.

The algorithm decision a task distribution aim in this case was the following: Each calculator with the use own generator of random numbers will create one variant decision of a task, and naturally, by the virtue of accident of a choice is far from optimum. All variants are different. The calculators, communicating among themselves casually, and number of connections j at each calculator are less than their common number m and is defined from model researches, exchange variants decision a task. Then each calculator crosses these individuals increasing thus popularization decisions twice. Further, following the procedure genetic algorithm, the calculator estimates the received posterity, rejects half worse individuals and it carries out the mutation in limits beforehand given probability, and again exchanges the best individuals. The given procedure will be carried out such number iterations, while the cumulative popularization decisions will not degenerate to the sole decision. This moment also is the moment termination of accounts. A result each robot becomes the owner best decision.

The demonstration program COLLINTROB was developed and tested by the given algorithm in the sphere Delphi and it has shown by the great number control accounts, that at an output on an identical degree accuracy quasioptimal genetic selection variants, the time accounts at the collective calculator, as well as was necessary to expect, but it was necessary to receive, is reduced proportionally to number calculators in system. At some experiments the intentional breakage separate calculators was simulated. The conclusion was made, that the accuracy decision a task has not suffered (i.e. was received the quasioptimal decision laying in a minus 5 % from optimum), but the time decision was increased proportionally.

M / (m-k), where k - number failed calculators.

The carried out development and researches have allowed to make the conclusion about the expediency of multisequencing of computing work at the decision optimization tasks in group of robots with the use a genetic method, as it results in the reduction time and the reception better results.

Analyzing these experiments it is possible to make a conclusion about the increase "intelligence" collective calculators incorporated in the supercalculator by the exchanged connections and working on genetic rules, in comparison with the separate single calculator on the given class of tasks. In other words, in this case we see the metasystem transition resulting in the occurrence higher "mind", consisting from artificial components with a lower level of "intelligence".

The same rules organization interaction between separate components collective mind, this time, by people, were applied for self-organizing a new kind mind, as so-called genetic consultation (GC) [4].

Let's define this concept as follows: GC is a way organization the collective work people working on the united project with beforehand as the given purpose by rules, based on principles classical genetic algorithm and being formulated as precise or indistinct instructions organization individual work participants project and their interaction. It is possible to consider this symbiosis people working in computer networks by given rules, as a natural multi-agent system. Thus, each man, the participant project, is the intellectual agent this system. It is possible to say that GC is a way of organization ideal interface of inclusion man as the natural intellectual agent on the informational sphere (IS). There can be a collective mind, by the inclusion in IS through this interface of people, working under offered rules.

The method GC is ideal for collective development creative projects. For example a project with a designation aim, specifications, the cost of development is placed on a site collective creativity in a global or in a corporate network. There is an opportunity telework many experts above this project. The project is a field an attraction various useful ideas. The collective intelligence with the help a method genetic consultation generates new ideas, combines them and leaves in the project the best them, gives an expert estimation each variant and allows to estimate the contribution each participant as the generator ideas, or an expert, exposing them appropriate ratings.

The application of ratings will allow building hierarchical systems genetic consultation. Ratings can be defined for variants project, for offered ideas and participants by results of their work. In principle, in frameworks GC it is possible the formation of hierarchies from the participants projects by the development several initial alternative ideas and by the generation new ideas. There is a competition ratings not only ideas, but also people. If the participant’s rating regularly falls, he can leave for an other project (thus there is a natural selection not only variants, but also participants). It is possible to present one simple algorithms functioning collective intelligence, which was approved) by the author and his colleagues on a number of different tasks; both in a computer network, and in outmachine environment, as following:

1. One or several aim projects and its requirements are formulated. The structure participants and the way of their interaction are determined. The system of rules of generation of variants decisions (or their parts), the estimation, the ranking the, selection best variants, is formulated and given.

2. The skeleton project (under a skeleton is set is understood: the plan project, structure document is given). The skeleton project also can be received as the result collective creativity. The participants begin work. In the beginning they generate the first approximation (fill slots of skeleton completely or partially). According to the rules of interaction, they exchange copy decisions.

3. Each participant ranks variants at own discretion, carries out the operator natural selection, leaving some best variants.

4. Each participant combines from the variants, new (equivalents of crossover), brings in obvious changes (the equivalents genetic operator mutation.

5. These variants are sent to the respondents, and he accordingly receives other's best variants. When the popularization decisions will degenerate (many identical decisions) are sent, project is considered executed, and transition on item 3 is carried out.

According to the type project (soluble problem) the system rules can vary (there can be a variable structure participants, the amount and the list of respondents, the number left variants etc.). Such rules can be selected experimentally, under each type project including method collective creativity.

The method GC was approved at the decision following tasks. The groups’ students from 4 up to 20 men participated in experiments.

1. The collective creation photorobot.

The demonstration program PHOTOROBOT was developed on the basis offered above rules in sphere DELPHY-4 on which the plenty experiments on restoration subjective portraits was carried out. The received results have shown the reliability and the convergence offered method even when about half witnesses obviously lied in the estimations. Moreover, the program, taking into account the average estimations by the witnesses of quality variants, offered them, allows to find out the unfair witnesses and to reveal the best.

2. The multicriterion task an optimum choice assignments from n applicants on m vacant posts, where n > m.

The educational task on formation a list of staff of some conditional firm, made them was put to group students. By a method GC the students should reach the consolidated decision and to fill in vacancies in such way, that to satisfy the inconsistent requirements on the psychological compatibility and at the same time to achieve, that the made up team had a high rating and sufficient professionalism. The received results decision multicriter task on the assignments during experiments on several groups students have shown good convergence a method to quasioptimal decision a task on assignments, taking into account the tables preferences.

3.The formation an investment portfolio.

In classical statement this task is formulated as follows. There are some banks with the various interest rates dependent on the sum, brought in by the client, and on term contribution. It is necessary to construct such a program of distribution investments on banks, in order to maximize the gain capital for the given period time.

After the acquaintance with the initial data and the aim project the students separate from each other, using the last experience and intuition, make the first variants decision a task: thus the partial, incomplete decision was supposed. Further, with the use algorithm GC, described above, at collective work after the small amount iterations the popularization decisions degenerated to the sole decision. As showed comparative the account this.

Tasks on computer, the received decision was quasioptimal and was in 5 % from optimum.

4. The direct measurement IQ group students working on a method GC, with the use known tests on Isenk [5 |.

The measurement IQ collective, supplied with the method collective creativity was carried out under the following scheme. It was offered to pass a testing under the verbal test on Isenk, to a group of students, consisting of 8 men, individually. The disorder in individual IQ from 100 up to 170 numbers was received. The group students was broken into subgroups - "leaders" with 1Q from 120 up to 170 numbers (average number 143) and " backwards " with IQ from 100 up to 120 numbers (average number 112). Further, from each subgroup: GC was formed, and a rule interaction and new tests were given to all participants. The opportunity writing off and spying was excluded. According to a technique on Isenk on the decision test, as well as in the first case, 30 minutes were allocated. For this time of group students had time to carry out till three iterations collective coordination points view according to a technique and rules GC, stated above. The following results were received. IQ collective mind group of "leaders" has made 215 units. For the sake justice it is necessary to note, that the technique on Isenk is not designed for the exact measurement IQ at such below-cutoff meaning factors, but it is necessary to say, that 45-50 tasks were solved correctly. The factor intelligence group has made 180 numbers (33 correct answers), by exceeding IQ the strongest leaders. Both in that and in the other case the result excess collective IQ above average individual on 70 numbers was fixed! Certainly, by results one experiment still it is impossible to make far-reaching conclusions and generalizations about the force a method, but the author is sure in reproducibility similar results further and other researchers. In opinion author, it is enough rules, stated in the present work, of functioning GC, for starting and checking serviceability a method in various areas creativity work.

Experiments on the use a method GC in the following areas human creativity also were carried out: a collective spelling verses, music, drawing up psychological portraits, an assembly simple computer programs, a choice best continuation a chess party, the task direct-sales representative, the drawing persons and the abstract diagram.

The experiments have confirmed, that factor of intelligence collective mind is higher; than the intelligence separate participant and at work collective mind thre is diffusion knowledge from the best participants to rests. The method allows to rank the participants under the contribution to a creative product, gave them objective estimations by a natural way. Time decision test tasks, it is enough simple for the first experiments, as a rule, did not exceed two hours (on the test on Isenk 30 minutes). The students with enthusiasm took part in these experiments, and it was interesting to them to be a part of collective mind exceeding their opportunities individually.

The received effect amplification of intelligence at metasystem transitions and self-organizing new, more powerful mind both in case artificial, and in case natural components allows to make a conclusion about some kind of universality applicability genetic rules at construction various kinds mind. Also it is possible to hope for an opportunity application given method for increase intelligence at association other carriers, for example neuronets and neurocomputers. And this is a question, which remains open and waits for the researcher.

Literature

1. Turchin V.Ph. A phenomenon of a science. The cybernetic approach to evolution. M.: SCIENCE, 1993, 295pg.

2. Protasov V.I. Generation of new knowledge by network human-machine intelligence. Statement of a problem. // Neurocomputers). Development and application. -M. 2001, N 7-8, p.94-103.

3. Protasov V.I., Vitiska N.I., Shelhkova L.A. Use of collective intelligence of group of robots for acceleration of acceptance expedient decisions. Materials of scientific youth school “Intellectual robotic system” (PRS-2001) ". October 1-6, Gelendjik 2001. pg. 187-189.

4. Protasov V.I., Vitiska N.I., Shustov E.V. The decision of complex problems by a method genetic consilium. The theses reports international conference " Intellectual multiple-processor system (IMS-2001) ". October 1-6, Gelendjik 2001. pg. 53-55.

5. Tests IQ. Techniques of definition of factors of intellectual development. Rostov-on Don: "Phenix", 2001, pg.128.

Back