DD201356A1 - PROGRAMMABLE LOGIC ARRANGEMENT FOR INTELLIGENT AUTOMATES AND ADAPTIVE CONTROLS WITH CUSTOMIZED INTELLECT - Google Patents

Abstract

The field of application encompasses a wide variety of applications for intelligent logic complexes with situation-based decision-making ability and adaptive capacity. The aim of the invention is the versatile, advantageous implementation of "learnable", associative assignment complexes for the purpose of implementing intelligent machines. The object of the invention is to show a programmable logic device for the realization of hierarchical logic complexes or data flow computers based on functional memories for use in adaptive systems with artificial intellect. The solution essentially consists of at least one allocator for at least one assignment level with fixed and / or variable logical functions for verifying invariants for binary and / or multivalued variables or for assigning output bits to input bits of the allocator, whose inputs and / or outputs at least one such coupling unit, connection unit and / or connection and connection unit are connected, which is integrable with the allocator and coupled to at least one programmable control complex, computer or processor. Fig. 9

Description

234948 8

Title of the invention

Programmable logic arrangement for intelligent automata and adaptive controls with artificial intellect.

Field of application of the invention

The field of application covers the most diverse areas of application for intelligent logic complexes with situation-dependent decision-making capability and adaptability. Areas of application are, for example, adaptive process, machine or robot controls, adaptive automata, hierarchical data flow computers, adaptive recognition systems, consulting units for the representation of knowledge (models) for situation-adapted decision-making in man-machine dialogue or adaptive systems with artificial intellect as models of the associative Memory in the brain for the representation of semantic networks.

The application of the invention for sensor signal processing adaptive controls of processes, production lines, machine tools and especially of robots of the third generation will in future have great economic importance for automation and rationalization.

Characteristic of the known technical solutions The theoretical and technical modeling of intelligent logical structures in accordance with previously investigated functional and structural principles of the central nervous system, above all of the memory in the brain, is a concern of the scientific community.

iaH0V.1981 * 9 ^r <2253

34948 8

discipline Artificial Intelligence.

In addition to adaptive allocators (B. B. the learning matrix of K.SLeinbuch), the perceptron of F. Rosenblatt, the cognitron of K. Fukushima and other associative object and speech recognition devices are to control intelligent behavior, especially solutions by computer programming with specially developed programming languages and, more recently, special computers (eg A. Newell and H. Simon production systems and LISP computers). A particular research direction is aimed at the development of hierarchical data flow controlled processors (data flow computers) based on functional memories.

For process, machine and robot control increasingly fixed or modifiable programs for task solving by modeling logical structures are being developed. The technical basis to form computers, also coupled with logic processors, using microprocessors and minicomputers. In DE-OS 29 32 394 (class G06F 15/46), an "intelligent, programmable process control arrangement" Texas Instruments is described. This is characterized by the coupling of a logical operation microprocessor (called Boolean processor) to a digital processor for arithmetic and statistical operations via a central memory unit. The Boolean processor is controlled by a program in the control-store. As a "bit processor", it mainly executes instructions for logical operations related to an accumulating bit PF (energy flow bit) and the addressed operand bit in an image register. It controls the process by means of an image register unit for input and output bits in accordance with the on / off states of sensors (sensors) or controllable devices arranged along the controlled process. It has 256 input bits and 256 output bits. Adaptive controls with artificial intellect and third-generation robots are not yet state of the art. Their logical structures possess artificial intellect

234948 8

(ie, "limited" intelligence) if they have algorithmic understanding and task-solving skills for goal-oriented behavior. The intended adaptability of technical controls aims at maintaining homeostasis (ultrastability), d. H. the automatic prevention and correction of disturbances to achieve stability at the best possible level. Flexible adaptability requires the acquisition of knowledge through structural learning.

Object of the invention

The aim of the invention is determined by experience in the technical implementation of a cybernetic model of adaptive systems with logical structure and artificial intellect developed by the author. To demonstrate adaptive controls with artificial intellect, in 1980 by programming a microprocessor with a screen and keyboard, a homeschool learner was simulated and tested to simulate drive-driven and motivational feeding behavior with multivalued variables for recognition, assessment and decision results. Its logical structure corresponds to that of an author-defined Intelligent Automata (see patents cited below).

The general aim of the invention is to demonstrate technical solutions for the realization of intelligent state machines with logical structures and artificial intellect, e.g. For adaptive process, machine or robot control. For this purpose, the modeling of causal networks of concepts is required as such semantic networks, which are characterized by inductive and deductive causal relationships between invariants for support information or elementary concepts. The aspired learning ability of the machine requires the presentation of extensible knowledge and knowledge acquisition. Technically modifiable logical structures in the sense of "functional storage" with processing and storage functions are modeled after the associative memory in the brain. '

A particular aim of the invention is the versatile, advantageous implementation of "learning ⁰ , associative assignment complexes for the purpose of implementing intelligent automatic machines with complex, situation-dependent decision-making ability and adaptivity in task solving to achieve intelligent behavior.

Prior to the invention, several patents have been published by the author to problems of the realization of intelligent machines since 1978. A problem to be solved is the economic implementation of many logical links with couplings for fixed and learnable relationships between invariants of binary or multivalued variables of the hierarchically coupled assignment levels or assignment complexes according to the invention. The problem is to be solved by the use of computers, in particular microcomputers / with the advantage of flexible programmability.

The time required for the many logical operations is to be reduced compared to the sequential and particularly bitwise operation of Boolean processors or computers by parallel processing by means of fast allocators in order to realize complicated "intelligent" controllers or machines for real-time applications, even with microprocessors of relatively low operating speeds.

Definable or programmable logic circuits, as freely as possible, are used freely as usable allocators in the sense of "functional memory". B »Gate arrays or PLAs, or even memory, eg. RAMs, EPROMs or EEROMs, and possibly "adaptive" allocators and associative memories.

Explanation of the essence of the invention

The object of the invention is to show a programmable logic device for the realization of hierarchical logic complexes or data flow computers based on functional memories for use in adaptive systems with artificial intellect. With the logic arrangement according to the invention are logical

⁵ 234948 8

Structures for causal conceptual networks (in general) and intelligent automata or "adaptive" assignment complexes (in particular) can be implemented or realized. You can z. B. be used for adaptive controls for sensor signal processing. The invention achieves the programmable predisposition of "structural learning" and "associating by experience" capabilities. The latter determine the flexibility of the adaptability of systems with logical structures and artificial intellect. They enable task resolution by means of correctable outside world models. The invention is based on the following theses: The structural storage of "learned" relationships is the prerequisite for associating predictions "from experience" for the purpose of appropriate decision-making. Learning (situation-adapted behavioral change) is attributed to structural and functional changes of the logical structure of the memory , Knowledge acquisition means to "capture" relationships during information processing in the subject by structurally building up or degrading causality relationships between verifiable invariants for support information or terms (see Effectiveness Changes of Synaptic Connections Between Neurons). This basic process of all knowledge-based forms of learning is called "structural learning" by the author. The memory, which is understood as an intelligent logical structure with learning and thinking ability and both memory and processing functions, can be modeled with a defined intelligent automaton. The latter possesses artificial intellect if he is capable of recognizing the essential and to solve the task through situation-adapted decision within such limits, which depend on the algorithmability of the task resolution to be simulated and the assessed learning ability.

The decisions of the intelligent automaton are on the one hand of recognized situation characteristics and on the other hand of situation-dependent, self-stabilization (homeostasis) serving

™ O **

4 8 8

Objectives of the automaton according to the "subjective situation" (eg motivations) determines which result from the consideration of associated predictions about situation characteristics of the outside world »The intelligent automaton has distributed or concentrated assessed, variable couplings, which the" learnable "( ie buildable and degradable) and represent the "learned" (ie built-up or degraded) relationships between verifiable binary or multi-valued invariant invariants. By means of fixed or changeable forward and backward couplings between logically connectable invariants for different abstract information representations (of any number of hierarchically couplable assignment levels) both "inductive" and "deductive" relationships can be implemented fixed, controlled or lerribar in the automaton. Such relationships of causality built up or broken down by structural learning form the basis for "associating with experience". The latter means generating such "associated" information as experiential predictions or expectations related to verified causal or key information by conditioning in at least one learning phase under the coincidence condition. What is needed are "predictions of the first kind", i. H. associated with recognized situation characteristics correlated "associated" information expected from the passively perceived external value and also "second type predictions", i. H. correlated with subjective decisions "associated" information, which are expected as possible consequences of the effects of the active subject on the outside world. The functional and structural principles of the intelligent automaton on the basis of assignment complexes according to the invention for recognizing, evaluating, deciding and controlling the actions were described by the inventor in 1978 in the following patents of the class G06F 15/18:

234948 θ

WP 145 436 Automaton with artificial intelligence WP 145 338 Attribution complex with properties similar to associative thinking

WP 140 927 Modular allocation unit with threshold logic WP 145 810 Method for associative coupling of signals

in a transmission channel WP 149 723 Intelligent automaton as a rough model of the brain without

Consciousness WPG06F / 224 629 Intelligent automat with adaptability

According to the invention the object is achieved by the features specified in the invention claim.

embodiment

1: Example of the hierarchical coupling of assignment levels ZN ₁ ,. .., ZN to form an assignment complex

Fig. 2: Example of a network with coupled allocation units ZE ₁₁ , ..., ZE for a complex allocation level or

J.J.W.

an assignment complex

Fig. 3: Example of an elementary assignment level with allocation units ZE., ..., ZE Fig. 4: Example of an allocation unit with at least one

logical function for the verification of invariants of at least one output variable (basic circuit) FIG. 5a: an example of an expandable logical structure of an assignment complex or intelligent automaton for modeling causal conceptual networks with fixed and variable couplings for causality relations FIG. 5a: shows the equivalent circuit diagram of FIG "adaptive" assignment complex 30

6 shows an example of the coarse structure of an intelligent automaton as a multistable system of coupled functional complexes

- 8 - β "5 AQ

cb's "^ Jj Sw ^ slk #

7 shows an example of a flowchart (Petri net) for the main processes and information presentations of an intelligent automaton

8: exemplary embodiment of an intelligent automaton with assignment complexes (block diagram)

Fig. 9: Circuit example of embodiments of a programmable logic device for the realization of hierarchical logic complexes and data flow computers for intelligent machines.

1 shows the formation of an assignment complex with a hierarchy of assignment levels ZISL, ..., ZN, starting from successive assignment levels (an image chain, see Figures 5a, 5b), by examples of additional forward and backward responses. With the Vorwärtskopplungeh 1 to each mapping higher levels of the hierarchy "inductive" causality relations between verifiable invariants (eg., Bits or groups of bits, characters or words) of binary and / or multi-valued variables of the allocation levels ZN _1, ..., ZN are defined , The feedbacks 2 (shown in phantom) to the same or lower levels of association define "deductive" causal relationships between invariants. The inductive and deductive relations are, according to the principle of causality and according to the principle of convergence and divergence of causality relations, principally possible from one invariant to any number of others and from any number of others to an invariant of a considered assignment level.

A network of causally correlated verifiable invariants is the model of a 1-way structure for representing information according to a semantic network. There can be at least one inductive and deductive causality relationship between two verifiable invariants. In addition, "self-sustained" relationships are possible from one invariant to the same invariant. The relationships of causality can disjoint each other,

234948 8

conjunctive and / or antivalent (alternative). They are the basic relations of the logical structure and define logical connections. The negation of invariants arises by including "zero invariants" of the same variable in logical operations. Due to deductive relationships through feedbacks at assignment levels or between multiple levels, repercussions for maintaining, confirming, or affirming the verification of invariants can be formed (eg, short-term storage, deduction, stabilization, and decision reconciliation).

An assignment complex with the level-related inputs 3 and outputs 4 for its input and output variables (see WP 145 338, class G06F 15/18) serves as a model of causal conceptual networks or semantic networks (see FIGS. 5a, 5b). The model can be divided into hierarchical display stages with variable configurations for displaying information. Per presentation level, an assignment level is defined. It forms verified invariant sets (vectors) with the invariants c at specific times τ. (ie the current values) of its N-valued ones

J ^s, .J

Output variables y. (j = 1, ..., n; s = 1, ..., N.) · These invariant sets are formations y € Y (g = 1, ..., N) of its output representation y = y, ie the configuration Y ₁ , ..., y of its output variable y .. The invariants C. € Y, become

YY ^s J

verified by logical operations or links (implementable with value assignments). The invariants included in logical associations of the assignment level c. 6 X. of its inputs are current values for N-valued variables x. (i = 1, ..., m; r = 1, ..., N.) · As invariant sets at certain times T, they form the formations x _f € X (f «1, ··, Ν _χ ) of the Input representation x = x ", ie the configuration x-, ..., χ of the input variable x., Of the assignment level. An assignment level implements the causal mapping <f: X - * Y and has the assignment function <p (x, z) = y in accordance with the result function of an abstract automaton in the sense of the automo-

234948 8

ZE ₁₁ / ··· / ZE- whose assignment level ZN _{2 is formed} with ^ ^E ? 1 '***' ZE _2v and its assignment level ZN with ZE., ..., ZE. In addition, however, the network may also be construed as a complex allocation level 26a (eg, as ZN of an assignment complex of Fig. 1). It has inputs 5 for the input representation χ and outputs 6 for the output representation y. The complex assignment level 26a leads to complicated logical operations for the verification of the invariants c. its starting point

J ^s variable y. by.

FIG. 3 shows the example of an elementary allocation level 26b with allocation units ZE ₁ ,. .., ZE shown. It generates output bits 10 for invariants c. the information representation y of its outputs 8 associated with the input bits 9 for invariants c. Examples of internal, retroactive couplings of the assignment level 26b, which lead from outputs 10 to 9 inputs of the allocation units ZE, ..., ZE are shown in dashed lines.

For each allocation unit ZE ₁ , ..., ZE 9 such invariants c. or c. which are to be included in the special logical operations of the allocation unit for invariant formation, so that according to the coding by at least one output bit 10 of the allocation unit, the verified invariant c. at least one output variable

y. of the assignment level 26b is generated in coded form.

4 shows the example of an allocation unit 25a, 25b with at least one logical function for verifying invariants of at least one output variable as a basic circuit. It has universally usable basic functions of the two-valued logic for verifying at least one output bit 20; 21; 23. The alternating sequence of logical conjunction 11; 13 and disjunction 12; 14 is advantageous for implementing a variety of logical operations. Instead of these basic functions

4 948 8

You can also use negated basic functions such as NAND and NOR and also additional negations. The logical links are z. B. implementable with logic gates integrated circuits. The allocation unit 25a, 25b is for generating a plurality of output bits 20; 23 with correspondingly many logical functions (eg, such as in FIG. The verifiable invariants are associated with one or more output bits 20; 21; 23 codable. Characteristic of the allocation unit 25a; 25b is an arbitrary number of inputs 15; 15a; 17; 17b for conjunctive relationships and inputs 16; 16a; 22 for disjoint relations, and in the case of inner couplings also internal inputs 18 and 19 for disjoint or conjunct inner relationships. The entrances 15a; 16a; 17a signals for negated invariants, z. B. by means of the inverters 24, are supplied. The inhibition is with negated invariants via inputs 15; 15a; 17; 17a, 19 implementable for conjunctive relationships. All inputs of the allocation unit are in principle with coupling units 67; 67a connectable to z. B. variable, adjustable, expressible or "learnable" couplings. Each output bit 20 of the allocation unit 25a may be used negated, e.g. By means of inverter 24. For internal retroactive couplings in the assignment level, the outputs 20a and 21a are provided.

For the purpose of disjunctively incorporating bits into logical operations of the allocation unit 25a; 25b, based on at least one output bit 20, are adjacent to the inputs 16; 16a; 18 nor inputs 22 of the optional unit 25b for the disjunctive coupling of associated information in transmission channels for the output bits, z. B. 20, by means of additional disjunctions 14 available. These disjunctive inclusion possibilities are used in association by "associate", i. H. Contentwise related, information with the equivalent, "associative" information of an output bit 20 are disjunctively linked. According to patent WP 145 810, class G06F, 15/18, associative coupling

234948 8

can be used. The logical disjunctions 14 can be implemented both by means of suitable hardware (eg, OR gate, NOR or wired or) or by software (eg with programmed logical operations). In the patent specification WP 140 927 the author describes an allocation unit with threshold logic and disjunctive linking units.

The allocation units of an allocation level 26a; 26b can be realized both for fixed and variable, controllable, interpretable or "learnable" logical functions for invariant formation. For example, the allocation unit 25a; 25b with their universal basic functions by selective use or wiring of their used and redundant inputs 15; 16; 17; 18; 19; 22 can be used with specially provided bits to define the desired logical functions for various applications with specifiable functions. The realization of arbitrary allocation units or allocators is possible in a known manner with the following technical means:

logic elements (eg gates, threshold elements (see WP 140 927), switches, contacts or gates), fixed, programmable or "adaptive" logic circuits (eg PLAs or gate arrays), allocators (e.g. ROMs, PROMs, EPROMs, EEROMs or decoders), classifiers (eg identification devices or learning matrices) and / or read / write or associative memories (eg RAMs, associative memories or associative coupling matrices, see WP 145 810) ).

In addition, possibly for the allocation units internal or external coupling units 67; 67a for prepared (67a) or destructible "latent" (67) connections and relationships.

With storing connection units 93; 94; 96; 97; 95; 99, as shown in FIG. 9, the control and / or alteration of logical and / or assignment functions of the allocation levels and the establishment or reduction of relationships or couplings between bits, lines, signals, states, data or invariants are performed. They can be program-controlled or

34948 8

autonomously (eg, through structural learning) change their coupling, guiding, ability and / or storage properties or their effectiveness. The latter may be controllable, adjustable, switchable, variable and / or expressible (eg by destroying or making connections).

To implement the structural learning, such coupling or connection units 67; 93; 94; 96; 97; 95; 99 can be used, which by conditioning in learning phases autonomously either build up or degrade their coupling or connection properties (eg under the coincidence condition). The coupling units 67; 67a are connected to connection units 93; 94; 95; 96; 97, v 99 can be integrated or connectable or may be identical to them. Technical means for their realization are shown below for experience memory.

Each assignment level 26a; 26b of an assignment complex serves to implement the following basic processes of logical thinking (see Fig. 1):

- Analysis of the formations x "of the input representation χ by logical operations or operations for the verification of individual invariants c ·,

JS

- Synthesis of formations y of the initial representation y by verification of these invariant sets with individually verified invariants c. for y as display elements (items).

J ^s 9 '

Definition: As information (lat .: in formation) we mean the semantic content of a formation of several invariants (invariant hunt, vector) or in particular a verified invariant (signal value or signal value set, eg character or word). Information can be coded by means of different formations or invariants and represented with different abstraction lines. Decisive for the understanding or decoding of the information is the agreed code corresponding to the common term or invariant supply of information transmitter and receiver.

- Abstraction of formations x _f by assigning only one formation y to several x _{f of the} same class (classification for the purpose

234948 θ

Informationsreduktiori, d. H. Reduction of the redundancy of representations χ = x "by formation of y),

- generalization of formations y as the abstracted essence of classified formations x ",

- acquisition of knowledge through structural learning, ie changing the logical structure of the "adaptive" assignment level by building up or dismantling its "learnable" relationships of causality,

Induction and deduction (logical closure) by associating information due to disjoint inductive or deductive causality relationships of the assignment level, which may be fixed or "learned" (e.g., for associating predictions "from experience").

"Learning" (ie latent, structurally prepared and trainable) relationships (eg, c "6" c or c & c) of a logical structure that can be expanded and modified are the basic prerequisites for structural learning , which is a conditioning-dependent change of the structure Structural learning means the structural storage of acquired knowledge as "experiences" through "learning" relationships between invariants for support information resulting from one or more conditioning (during the "learning phase") during information processing By transforming informational representations by means of the logical structure Knowledge acquisition through structural learning is the prerequisite for the flexible adaptability of an intelligent logical structure by "associating with experience", ie empirical prediction of associated information the construction or reduction of causality relationships under certain internal conditions during conditioning. For the conditioning of structural learning, the condition of coincidence through the temporal contiguity (ie the quasi- coincident coincidence ) of signal events of verified invariants for correlatable information or concepts applies. The coincidence condition can also be defined neurobiologically for changes in the effectiveness of synaptic connections between neurons by the coincidence of post- and presynaptic signal events.

- 234948 8

(see more advanced conjunction theory according to Marr (1969) and experiments by v. Baumgarten).

The construction or dismantling of causality relationships is either determined or stochastically implemented. Practical criteria are threshold values e or cf for the conditional probability (or relative or counting frequency) of the event of a verified invariant C ₂ for "associable" information, due to the quasi-simultaneously verified invariant c, for causal or "key information": P (C ₂ Ic ₁ )> £ * 0 for the relationship building and P (C ₂ Ic ₁ ) - cf * 0, e / & t for the dismantling of relationships. The invariant C ₁ characterizes the key information for the verification of at least one correlated with her "associated" information, which z. B. represented by the invariant C ₂ . The construction of the causality relationship between C ₁ and C ₂ is possible in the learning phase by one-off (iffc = 0) or multiple conditioning (eg according to the predisposition). The result is the conditional assignment c- - * c ~.

The relationship C ₁ O "c" is structurally stored, ζ. Β in an experiential memory With an invariant C ₁ for a key information, several verifiable invariants C ₂ , ..., c can be associated with "learnable" inductive or deductive Causality relationships are correlated (see Associative Memory for Conditional Assignments) By eliminating "learned" or degradable causality relationships, forgetting or relearning is implemented, forgetting can be time or age related, eg, determined by time constants for the validity of relationships in cases where there is no or insufficient affirmation of established and / or degradable causal relationships due to lack of or too little conditioning, the reduction of a relationship generally follows from falling below the threshold £ for the conditional probability (or relative or count frequency) of the actual event of the associate information in coincidence with the correlated key information. The "relearning" by the construction or dismantling of antivalent relationships (eg for several

234948 θ

Invariant of a variable) has the consequence that in the construction of a buildable "latent" relationship forcibly reduces the relationship with this relationship antivalent relationship or "forgotten" is. The implementation of the relationship or degradation of relationships by changing, making or destroying of connections (associations) z. B. with modifiable couplings, relationships, states, efficiencies, conductivities, memory properties or assignments between invariants possible. For at least one or more disjoint learnable and / or learned relationships at least one Erfahrungsspeieher of acquired knowledge or acquired knowledge is provided which implements the relationships. The experience memory models the associative memory of the logical structure in the sense of a "functional memory". It is implemented or realized with individually formable memories and / or coupling or connection units. For example, associative memories, read / write memory, RAMs, stored lists for conditional allocations, "adaptive" allocators, associative coupling matrices (as in WP G06F / 145 810) and / or experience-dependent controlled or variable coupling units 67; 67a or storing connection units 93; 94; 95; 96; 97; 99, feasible by z. As registers, flip-flops, memory cells, coupling elements, variable resistors, switches, diodes, gating circuits, electrical, optical, magnetic, physical, biological or chemical routes, cells or tracks with variable conductivity, connection, effectiveness and / or storage properties, which can be controlled, adjusted, programmed, trained or determined according to conditioning during structural learning. .

However, an experience memory can also be modeled with stored condition bits (eg, flag bits) or ternary signals of at least one "experience-dependent" or "learnable" program. This implements conditional assignments, conditionally variably stored relationships between bits, data or invariants, and / or conditionally varying logical functions or

34948 8

Associations according to the programmable conditioning in at least one learning phase, eg by conditional value assignments.

When implementing or realizing assignment levels and / or assignment complexes, the "learnable", i. H. be built up or degraded, causality relationships both at the assignment level and outside prepared, predisposed, trained, stored and / or programmed. The conditions for building or breaking learnable relationships through structural learning within and beyond a defined allocation level 26a; 26b are assessed either fixed or variable and / or can be defined depending on the situation, eg. B. by programs or situation-dependent signals.

Due to the existence of learnable relationships, ie the ability to learn structurally, "associating with experience" is possible. The Asso ziieren is defined as the process of verification of at least one invariant for such "associated" Information, which is related to the verification of at least one invariant for a causative or key information causally related. By key information, several conditionally associated "associated" information can be generated when the latter has been associated (ie correlated) with it, e.g. For example, when they existed with her at least once. The "associated" information is readable from an associative memory. They can be stored by means of invariants in memory cells of an experience memory and can be generated or addressed with invariants for key information. The learned relationships between invariants or signals for key information and associated information are structurally, z. For example, by manufactured couplings, connections or state changes for the generation or verification of invariants for associated information, stored, for example, in "adaptive" allocators, internal or external experience storage or by condition bits for "experience-dependent" programs.

234948 8

FIGS. 5a, 5b show the basic representation of the extensible modifiable logical structure of an assignment complex with possible fixed and variable couplings for fixed or "learnable" relationships between invariants for support information or concepts for modeling causal concepts or semantic networks.

A causal network of terms is syntactically definable as a directed network with interrelated terms or support information. It can be represented with a context or structure graph whose nodes invariants for support information or elementary concepts and whose edges define the causality relations as basic relations of the logical structure. Causal conceptual nets can be modeled as a hierarchy of successive layers or presentation levels. These include invariants for support information, for which certain degrees of abstraction and generalization (corresponding to the representation levels) apply. The invariants of a modeled conceptual network represent support information or elementary concepts and are correlated via inductive and deductive causality relations. The representation levels for a conceptual network are modeled by hierarchically couplable assignment levels 26a; 26b; 27; 28; 29 with allocation units (eg 25a, 25b) implemented or realized. Instead of the arbitrary number of assignment levels 27; 28; 29 "adaptive" allocators or assignment complexes (eg, Fig. 1, Fig. 2 or 30) are used.

The hierarchical chaining (cascading) of assignment levels with abstraction, assuming the identity of the output representation of one level with the input representation of the next assignment level / represents an imaging chain. The level of abstraction of the imaging chain increases in the direction from the lowest to the highest assignment level. An image chain with the individual maps of the assignment levels (^ ₁ , A - * B, <f ₂ : B- ^ C, ..., Cp ^: 3 -> K implements the mapping ψ « _β ^ _β ~ ^ l ° ^ 2 ° * * *

234948 θ

The logical functions of the mapping levels of an imaging chain are extended to form additional mapping complexes by predisposing or constructable additional disjunctive forward and / or backward couplings between verifiable invariants of the mapping levels. As a result, invariants of the same or different assignment levels are additionally included in logical links fixed or variable or can be included by structural learning. Such author-defined assignment complexes have capabilities for inductive and deductive association analogous to logical reasoning (see WP 145338). An assignment complex with only fixed, immutable couplings for causality relationships is called a "rigid" assignment complex. In contrast, a "non-rigid" assignment complex with fixed and "learnable" relationships is an adaptive assignment complex. The latter has a modifiable logical structure with the ability to learn structurally and associate from experience. It is a rough model of associative memory in the brain with short-term and long-term storage. Its structure corresponds to a data flow controlled hierarchical logic complex on the basis of functional memory as a new type of processor.

An adaptive assignment complex is implemented by couplings 47; 48; 49; 50 for assessed fixed or changeable (learnable) inductive and / or deductive causality relations between invariants in principle of all assignment levels 27; 28, 29. Particularly characteristic are formable "latent" couplings 43; 44; 44 "; 45; 45 '; 46; 46' for" learnable "inductive and / or deductive relationships which, for better technical implementation, may be grouped together in at least one store of experience 31, 32, 33, 34. The relationships that can be built or degraded are "learnable" relationships between verifiable invariants, in principle, of all output variables of the assignment levels 27, 28, 29 of the hierarchy can be embodied or can be set to latent.

For structurally storing the "learnable" and "learned" inductive or deductive causality relationships both in and out of an allocation unit of the assignment levels 27; 28;

234948 8

as well as in at least one experience memory 31; 32; 33; 34 can z. B. experience-dependent controllable coupling units 67; 67a, "adaptive" allocators, variable allocations (eg in memories), conditionally variable invariant logic functions, read / write memories (eg RAMs), memory units (eg flip-flops or registers), Associative memories or associative coupling matrices (see WP 145 810) are used (see above). Structural learning makes each invariant causally related to one another for "associative" information and for causal or key information, which on the other hand can be "associated" information (ie correlatable). The correlatable invariants can have both the same and different assignment levels 27; 28; 29 belong. Oeder experience memory 31; 32; 33; 34 serves for the correlation by means of couplings for invariants of the "associative" information 43a; 44a; 45a; 46a and couplings for such invariants of the causal or key information 43; 44; 45; 46; which can also be generated associatively. Assuming prepared latent "learnable" relationships, "learned" relationships are formed and stored through structural learning. Established relationships in an experiential memory are implementable, for example, through established connections or stored invariants that apply to associated information that is conditionally or causally associated with key information so that it can be called, verified, or addressed by the latter.

The "associate from experience" means that with key information couplings, ζ. B. 45b; 45c; 45d; 45e, due to learned relationships in at least one experience store.31; 32; 33; 34, experience has shown that at least one such "associated" information is generated, which has been correlated with at least one current key information by structural learning and via one of the couplings 43 '; 44 '; 45 '; 46' is transmitted. The invariants for associated information are identical or equivalent to equivalent invariants for associative information. They are disjunctively interlinkable with each other.

2349A8 8

Associated information is transferred to the transmission channels 47a; 48a; 49a; 50a for the synonymous associable information coupled and / or in disjunctive links, z. B. 35; 36; 37; 38, for the verification of invariants for associable information (see WP 145 810). The disjunctive inclusion of associated information in logic operations is both via disjoint inputs 44 ";45"; 46 ";16;18; 22 of at least one assignment level 27; 28; 29 as well as by means of logical disjunctions or disjunctive inclusion units, eg 25b for 35; 36; 37; 38, in connection with inputs and outputs 47a; 48a; 49a; 50a of the assignment levels feasible (see WP 145 810 and WP 140 927.) The verified invariants for causal or key information (eg couplings 45b; 45c; 45d; 45e for output variables of the assignment level 28) may also be direct , without separate experiential or associative memories, for causing or generating invariants for "associated" input or output information of the allocation levels (eg, couplings 44 ";45"; 46 ";43"; 44 "; 45 '; 46 ';48a;49a;50a;47;48;49; 50). Such association is by means of "adaptive" allocators or classifiers (eg learning matrices) for at least one "learning" assignment level 27; 28; 29 or programmatically by "experience-dependent" conditional assignments implementable. \ The inductive or deductive associating or closing is implemented by means of more prone or "trained" inductive or deductive causality relationships, starting from invariants key information of at least one output (eg. 45) of at least one assignment levels 27; 28; 29 of the hierarchy of verifiable invariants for associated information of higher assignment levels (eg 46 '; 46 "), equal and / or lower assignment levels (eg 45';45"; 44 '; 44 ";43'). The prerequisite for associating is that associative information (eg, via couplings 44a, 45a, 46a) has been related to the key information.

-234948 8

For the purpose of preventing (inhibiting) associated information or associating it with experience and / or for the purpose of controlling structural learning by promoting or preventing the build-up of causality relationships, eg. B. by means of acting on key information signals for evaluation results or motivations, at least one transmission channel correlatable information by logical conjunctions, gates, valves or gates 39; 39 '; 39 "; 40; 40 '; 41; 41'; 42; 42 'with certain control inputs 39a; 39'a; 39" a; 40a; 40'a; 41a; 41'a; 42a; 42'a either blockable or unblockable. The controlling effect on associable information, e.g. B. the channels 44a; 45a, is also possible. The "adaptive", expandable logic structure of FIG. 5a applies in principle to an adaptive assignment complex 30 with any number of assignment levels, which is shown in FIG. 5b with its equivalent circuit diagram.

Fig. 5b is an abstract, generalized view of Fig. 5a. The adaptive assignment complex 30 can have the structure of a defined intelligent automaton or a functional complex of such an automaton. He has the inputs 47; 47a; 43c; ...; 43e; 40a; ...; 42'a and the outputs 48; 49; , , . 51, which with any number of assignment levels 26a; 26b and / or assignment complexes 30 can be coupled. FIGS. 5a, 5b show the possibility of associative coupling of experience signals 43 'from an experience memory 31 into the transmission channel 47a for input information of the assignment complex. By means of retroactive couplings 44b; ...; 46b for key information, input learned information can be verified through associative predictions through learned deductive relationships, whereby analogies can be implemented by supplementing items "from memory". On the other hand, by means of feedforward couplings 43c; ...; 43e key information of inputs 43 via inductive relations to in principle all assignment levels (27; 28; 29) transferable. They can produce associated abstract information that implements inductive reasoning, knee-jerk associations, direct abstractions, and reflexes.

234948 8

The outputs 48; 49; ...; 51 of the adaptive allocation complex include different assignment levels 27; 28; 29 and may represent level-related associated information.

FIG. 6 shows an example of the coarse structure of the author-defined Intelligent Automata as a hierarchical system, which can be realized by data-flow-controlled logic complexes, processors or computers based on functional memory. The illustrated coarse structure can be decomposed into the following functional complexes: recognition complex 54, evaluation complex 55, decision complex 56, action complex 57 and association complexes 58; 59. This presentation is based on the structural and functional principles of the intelligent automaton as defined by the author in patents WP 145 436, WP 149 723 and WP 224 629. The intelligent automaton and its functional complexes can be realized with hierarchical assignment levels or assignment complexes. The modifiability of its logical structure through structural learning is achieved with forward and backward feedback for "learnable" and "learned" causality relationships. They are distributed over several assignment levels or in at least one separate experience memory of the association complexes 58; 59 can be implemented in order to simplify the implementation. The association complex 58 generates by inductively (58a) and deductively (58b) associating from experience caused by afferent signals 60a; 65a for key information, afferent signals 63a; 63b for associated information, which are called predictions of the first kind . These have the character of hypotheses or prejudices and allow the refinement and experiential supplementation of outside world information of the passive-perceptual automaton as in the understanding in the subject by associative thinking (for example, analogies and associative understanding related).

The association complex 59 creates afferents by deductive association from experience caused by feedback signals 61 for the "efective design" of the decision complex 56

234 948 8

Associated information signals 64 called second type predictions . They have the character of predictions or expectations, and serve in decision-making as assessable, likely consequences of the machine's active impact on the outside world according to the current e-design. The causal relationships between invariants for efference drafts and conditionally assigned invariants for associated recognition results as consequences, stored in the association complex 59, represent an inner model of the outside world on which the unconfirmed efference drafts are tested and evaluated for their "experience"effect; before they are confirmed as a decision. The means of association complexes 58; 59 generated signals 63a; 63b; 64 for associated information as predictions are disjunctively coupled into the transmission channels of afferent equivalent signals for recognition results (afferences) or associable information (analogous to afference synthesis according to PK Anochin).

The recognition signal signals 6Od are evaluated in the evaluation complex 55 with signals 63a; 64 for associated predictions and signals 61b; 62d for internal conditions (eg subjective situation) with respect to the situation-dependent objective of the machine (controllable by the signals 68) compared and evaluated. The evaluation result signals 62b operate on the decision complex 56 so that the interference design is determined with the best possible consequence by decision matching.

Structural learning, d. H. the construction or dismantling of the relationships of causality in at least one association complex 58; 59, with signals for decision and / or evaluation results 61e; 62e; 62f; 62g; 62h, z. By logical conjunctions or gates 61'e; 62'e; 62'f; 62'g; 62'h in the transmission channels for associated (63a), associable (60c) or key information (60a; 61a; 65a), preventable (i.e., inhibition), or deliverable by, e.g. B. the transmission of the key information is blocked or unblocked. As well

234948 8

by inhibiting influences on the key information, associating it with experience is preventable. In addition, inhibiting the transmission of associated information signals 63a; 63b; 64 possible, z. By gate 61'e. These control possibilities can be implemented analogously to axo-axonal connections for presynaptic inhibition in the cerebral cortex.

The Intelligent Automaton (see Fig. 6) is a multi- stable system with several hierarchical levels with respect to the control of the action complex 57. A multi-stable system (defined by WR Ashby) is an "almost collapsible" system (according to A. Newell), ie one off stable subsystems existing overall system in which the subsystems are temporarily independent of each other or interconnected by a subfunction. The individual subsystems within different hierarchical levels are connected to the controlling centers of the higher levels. They are part of broader subsystems of the overall system. The higher level can control or regulate the underlying level by signals. The functions of the levels are based on a dialectical unity of autonomy and hierarchy. The subsystems can independently perform stabilization and adaptation operations if they are granted this autonomy from the higher level.

The structure of the intelligent automaton can in principle be divided into the following hierarchy levels or levels:

A first level subsystem for unconditional and "conditional reflexes" of first inductive predictive type recognition formed by the recognition complex 54 having a learning inductive relationship association complex 58a, the action complex 57 and the couplings 60b; 6 NC; 63a; 65; 65a; 66; 69 /

analogous: low-developed nervous system with reflex arc and conditional assignments

Second subsystem for "conditional reactions" by decision and recognition with inductive and deductive predictions of the first kind,

formed by extension of the first subsystem, especially around the decision complex 56, the association complex 58b for

23A 94 8 8

learnable deductive relationships and the additional couplings 60; 61; 61c; 61 d; 61 e; 60a; 63b; 65b,

analogous: low-developed brain for associative recognition and experience-dependent reaction

3. Subsystem of the third level for "conditional actions" by deciding according to the evaluation of the detected situation with inductive and deductive predictions of the first kind, formed by extension of the classification (55) (for the determination of the subjective position and control of the elementary subsystems) and the additional couplings 6Od; 61b; 62; 62a; 62b; 62c; 62d; 62f; 62g analogous to: more advanced brain with limbic system for generating situation-dependent reference variables for drive-controlled, motivational or emotional recognition, decision-making and action for dynamic stabilization (homeostasis)

4. Overall system of the fourth level for "conditional consideration of decisions" by incorporating second type predictions in the evaluation for optimal decision making, formed by extension of the third subsystem, especially around the associative complex 59 for learnable deductive relationships and the additional couplings 60c; 61a; 62e; 64, analogous: sophisticated brain for controlling decision ·

alternative to the inner model of the outside world. As elementary subsystems of the multistable automaton, the hierarchically successive and network-coupled functional complexes 54; 55; 56; 57 for the basic processes of recognizing, evaluating, deciding and acting (eg executing behavioral programs). For autonomous stabilization of the subsystems, the feedbacks are 6Oe; 61c; 61d; 62d; 69 and 53a provided for special inputs of the functional complexes. The subsystems will then receive ultrastability (homeostasis) if they are stabilizable at different levels. They can be modeled by intelligent automata or assignment complexes 30 with fixed and / or learnable causality relations between assignment levels.

234948 8

The functional complexes as elementary subsystems of the intelligent automaton are comparable with anatomically, structurally and functionally similar brain sections. The recognition complex 54 with the association complexes 58; 59 is a rough model of the end brain (olfactory bulb and cerebrum) with associative cortex. The evaluation complex 55 and decision complex 56 in conjunction with the association complex 59 are comparable to the limbic system (broader sense) in the diencephalon and midbrain in conjunction with the prefrontal cortex for associating expectations. The action complex 57 controls relatively independently motor operations or behavioral programs analogous to the cerebellum of the hindbrain.

The subsystems of the nested hierarchical levels can be classified into a development series which corresponds to quantitative and qualitative stages of brain development (encephalization). Systems of the third and above all the fourth level show a qualitative leap through their evaluation complex 55. They have the ability to learn, solve, and understand (as defined by the author in IVP 145 436).

The subsystems of the Intelligent Automata defined above provide the following (nested classified) forms of learning based on structural learning and corresponding to psychologically defined learning modes:

1. conditional reflex - through inductive association in recognition

in the first level system,

2. Conditional reaction - through inductive and deductive association

in recognizing and deciding in the second level system,

3. conditional action - by means of drive-controlled, motivational

recognizing, assessing and deciding on the basis of predictions of the first kind in the third-level system,

4. conditional consideration of decisions - through control

decision alternatives, depending on second type predictions due to modeling in the fourth level system.

234948 8

The psychological effect law (according to Thorndike) for the so-called "instrumental learning" (effect is means of learning) according to the method trial and error (trial and error) applies analogously for the intelligent automaton. The effect-related predictions by associating from experience are included in logical operations both before the action (when deciding) and after the action (as compared to the actual effect).

In the following, the functional complexes of the intelligent automaton are described in their interactions.

The sensory input signals 52 in the recognition complex 54 are first suppressed, decoded, translated, analyzed, abstracted and invariants for input terms in the recognition complex in response to feedback signals for evaluation results 62a (eg identification motivation), recognition results 6Oe, decision results 61d and efference copies 69 54 assigned (see Fig. 7). The recognition is enhanced by inductive and / or deductive association by means of fixed and "learnable" causality relationships in the recognition complex 54 and / or association complex 58. With inductive and / or deductive relationships for predictions of the first kind, disturbances (irrelevance) or "unsharp" formulated input information can be corrected or specified by the experience of the automaton (eg analogous conclusions). Through abstraction and generalization, the redundancy of the input information is greatly reduced. With this information reduction, the reduction of the required channel capacity for the transmission of the recognition results (afferences) is achieved. A special side of cognition is the subjective assessment of selected situation features. Judgments about the situation are formed by the logical inclusion of associated information. The afferent signals 60; 60a; 60b; 60c; 6Od; 6Oe for recognition results are synthesized by association from experience with inductive and deductive predictions of the first and second kind (63a, 63b, 64) (compare: Afochem's Anochin Synthesis). The associated information with a higher degree of abstraction (63a; 64) and with a lower degree of abstraction (63b) are in transmission.

234948 8

channels for afferents can be coupled in a disjunctive manner (see WP 145 81C WP 149 723).

Afferent signals 65a; 65b of the low-level abstraction detection complex 54 are connected to afferent signals 60a; 60b higher levels of abstraction inductively and / or deductively correlated. The forward and / or feedback (eg, probability bindings) of the "learned" causality relationships for first type predictions become structural either in the recognition complex 54 or separately in at least one inductive relationship experiential memory 58a and / or at least in a deductible experiential memory 58b saved.

The afferent signals 60c of the recognition complex 54 are provided with decision decision results signals 61a of the decision complex 56, influenced by signals 62e; 62h of the evaluation complex 55, so that corresponding "learned" couplings (eg probability bindings) are structurally stored for predictions of the second kind in the recognition complex 54 or separately in at least one drift memory 59 for deductive relationships.

The evaluation complex 55 evaluates the afferent signals 6Od of the recognition complex 54 and compares them with the associated signals 64 of the second type prediction association complex 59 in response to signals (possibly) being passed to its inputs 61b; 62d; 68 of the decision-making complex 56 and / or valuation complex 55 or of changeable targets (eg plan or controllable objective of the machine). The evaluation results signals 62 of the evaluation complex 55 characterize the "subjective location" of the machine. They are classifiable into "drive" (conation), "motivation" and "affect" (emotion). These determine situation-dependent objectives, which (for example as current reference variables) are used to control in principle all functional complexes 54; 55; 56; 57; 58; 59 serve according to the adaptability. Through these signals, drive-driven and motivational behavior is achieved to ensure the stability of the machine at the best possible level (ultra-stability). The signals 62a; 62b; Steer 62c

2349-48 8

in the sense of situation-dependent reference variables the recognition, decision-making and acting according to the subjective position of the automaton. They define attention (62a), decision-making, acknowledgment or denial (62b), and possibly affects (62c). Special signals 62e; 62f; 62g; 62h for evaluation results may prevent or facilitate knowledge acquisition through structural learning and also associating experience for first and / or second type predictions by influencing key information (60a; 61a; 65a) or associative information (60c).

The assessment complex 56 recognizes the current task (analogous to the problem recognition), depending on the afferent signals for recognition results 60 sent to it and the signals 62b for evaluation results (eg decision motivation) as well as possibly signals 61c for own decision results. He devises a solution to the problem and generates signals 61a; 61b as preliminary result of the decision, called "efference draft". The evaluation compendium 55 examines (criticizes) the interference design in comparison with the recognized and assessed external and internal situation (represented by the afferent signals 6Od) by direct and / or indirect inclusion of the signals 61a; 61b in its links. The indirect inclusion of the signals 67a for the interference design in the evaluation is made by the correlated predictions of the second kind (predictions) if at least one experience memory 59 exists for an inner model of the outside world. The criticism of the efference design is represented by signals 62b which act on the decision complex 56 via its inputs either as confirmation (affirmation) or rejection (inhibition) with respect to the current efference design. This interaction between decision complex 56 and valuation complex 55 serves the purpose of decision-making . It is defined as the consideration of the efference design that, in terms of its expected experience of consequences (in the case of correlated predictions of the second kind), in comparison to the recognized and assessed external and internal situation (afferents) under the internal conditions

- 234948 8

(Signals 62; 62d) will be most advantageous or optimal for the machine in terms of reducing its need, subject pressure reduction, or achieving current objectives. If the e-design of the evaluation complex 55 is not confirmed but rejected, the decision-making complex 56 must change the proposed solution to the problem. He will then search for a new solution (e-design) if the task has been reset by the evaluation complex 55 by means of the signals 62b. The resulting new efference design will again be fed with feedback signals; 61b to the evaluation complex 55 for review. The affirmative signals 62b of the evaluation complex 55 corroborate the interference design to be considered an accepted decision result (ie, a "decision") for instructing the action complex with corresponding signals 61. According to the interference principle, signals 61d corresponding to the confirmed interference design and / or signals 69 are fed back from the integration complex 57 to the detection complex 54 as an "efference copy". They serve as reference information (references) to ensure the correct recognition of the reafferences in relation to the action of the machine (eg the action-related adjustment of the sensors). "Reafferences" are those afferences that are perceived as a result of decisions as consequences or effects of the action of the machine. They are recognition results as consequences of the confirmed efference design. After the decision has been adjusted, the "settled" signals 61a for the "confirmed e-design" are transmitted to the association complex 59. If they are not suppressed with inhibitory signals 62e in front of the evaluation complex 55, they cause the association of second type predictions as "expectations", which are transmitted with the signals 64 to the evaluation complex 55. Expectations are the predicates about experience-based consequences of the situation-dependent decision of the automaton or its action. They influence the evaluation results 62, which determine the subjective situation during the action (eg decision motivation). The expectation (64) is compared with recognition results (6Od) in the evaluation complex 55.

234 9 48 8

If, after the decision, the detected situation changes, the changed recognition results 60c are related as reafferences with the confirmed efference design 61a. If reafferences (60c) and expectations (64) coincide, the coincidence condition for confirmation of structurally stored causality relations between the verified Effegenzentwurf (61a) and such a conditionally assigned to it, associated afferents (64) Fulfill t _{t /} where the verified associable afferents ( 60c). The learned relationships for predictions of the second kind with respect to the verified efference plot 61a are confirmed. From the evaluation complex 55, the decision motive 62b is not changed in this case. The decision complex 56 and the action complex 57 may, however, be caused by special signals 60; 66 for afferences or reafferences, autonomous (in terms of the first and second subsystems) to control additional reflexes or reactions. In this case, no new task is formulated with the signals 62b of the evaluation complex 55 for the decision complex 56. The evaluation of the situation in comparison with the expectation leads to the strengthening or damping of the emotional state of a subject. The evaluation complex 55 may, due to the success of the decision, ie, the positive evaluation of the effect (reafference), generate such emotional evaluation results 62 (eg, affects) which actions of satisfaction (eg, utterances) through autonomous control of the action complex 57 and the improved subjective attitude (motivation) of the signals 62a; 62b; 62c; 62d; 62e; 62f; Express 62g (eg to increase learning and attention). If the reafferences (6Od) do not or only partially agree with the expectation 64, changed evaluation results 62, e.g. B. new task and changed objectives, the consequence, which act on the other functional complexes. Reafcences 60c, recognized as the "new-found" effect of the action, are correlatable to the confirmed e-design 61a by structural learning in the association complex 59, unless the efference-word 61a is replaced by a new task for the decision-making complex

- 234948 8

56 n is changed and the coincidence condition is fulfilled.

The structural learning of "redefined" relationships of causality or the affirmation of already learned relationships, eg. In the experience memories 58 and / or 59, is particularly encouraged when an acceptable or positively assessed effect has occurred, which leads to the affirmation of the decision. This is the case if the evaluation of the response ratios 60d (possibly in comparison to the expectation 64) does not lead to a deterioration and possibly to an improvement of the subjective situation (eg through need satisfaction or success experience), so that a positive evaluation result 62b results the decision complex 56 acts so that the last confirmed e-reference design 61; 61a continues to apply or is made valid again. This learning principle based on the evaluation of the effect presupposes the existence of a valuation complex 55. It applies to systems of the third and fourth hierarchical level of the intelligent automaton. These have decision-making ability for promising actions through effect-related predictions of the first and / or second kind. This principle is in line with the "effect law" for "instrumental learning" (ie the evaluation of the effect is the learning tool) defined by the psychologist Thorndike and has been demonstrated in animal experiments. Thorndike presented the universal learning method "trial and error". It is based on the evaluation of the effect of actions of the subject. It leads to structural learning of relationships of causality for predictions of the first and especially second kind, ie the acquisition of knowledge about the relationships within and (from the subject) to the outside world.

The intelligent automaton with an inner model of the outside world may "consider" the decision for an action with the most advantageous effect to be expected, after having "learned" the effects of various decisions. He has flexible adaptability, if he by situation-adapted

234948 8

Deciding to solve the problems posed by the evaluation complex 55 on the basis of accumulated experience so that its homeostasis is maintained, even if the external world conditions change so that inner models of the outside world, ie learned relationships for first and / or second-type predictions, corrected or need to be retrained. The adaptability is affected by the setting of the machine, ie its general readiness to adapt to certain situations; to react in a predetermined and usually practiced way. Insufficient learning ability and lack of "learnable" causal relationships of the logical structure imply an inflexible attitude and, as in the case of one-sided training, lead to out-of-order decisions with the effect of rigidity over untrained changes of the outside world. Rigidity can be alleviated or eliminated by increasing the ability to learn and to gain versatile training (ie, learning many relationships for predictions) to improve and correct the inner-world model.

The action complex 57 executes the behavioral programs, which by the logical combination of its input signals 61; 62c; 66; 53a be determined. It can control and regulate special processes with autonomous routines and generates the control signals 53 for a plurality of effectors or control devices, depending on feedback signals 53a, z. above sea level. of effectors. Special signals 66 from sensors and or from the detection complex 54 can trigger knee-jerk reactions (similar to the so-called "reflex arc") in the action complex 57. The behavioral programs are of signals 61; 62c can be triggered or influenced for decision results (ie decisions) and evaluation results (that is, Ir. Affects).

The implementation of the functions of the intelligent automaton is advantageous with programmable computers, e.g. Microcomputers, or logic or assignment complexes possible, each for one or more functional complexes 54; 55; 56; 57; 58; 59 (eg multiprocessor or multicomputer system as in WP 224 629). The simulation of the machine with

- * »- 234948 8

Computer programs are advantageously possible with programming languages for logical and list operations.

Fig. 7 shows an ex-flow chart in the manner of a Petri net for the main data-flow controlled processes (as transitions) of an intelligent machine with designated information representations on squares (represented by circles). The temporal succession of the main processes implemented by the functional complexes can be chosen such that as many logical operations as possible run parallel in time. The verified invariants of transformed information representations are stored because of the sequential processing on special memory locations, which serve to couple the processes. The transformation of the information presentations can be controlled step by step, cyclically and iteratively. The information processing and storage for implementing the main processes of the intelligent automatic machine are carried out with control methods and technical means of computer technology and microelectronics in a known manner. ,

Fig. 8 shows an embodiment of an intelligent automaton (see Figures 6 and 7) with "rigid" or "adaptive" assignment complexes 70; 71; 72 as in FIGS. 5a, 5b, omitting the action complex 57 (compare WP 149 723 and WP 224 629). The inputs 74 of the machine (eg, for receptor or sensor signals) are connected to inputs 75 of a detection complex 70. , Outputs 76 of the afferent detection complex 70 are via special couplings 76b; 76c; 76d; 76e; 76f are connected to at least one experiential or associative memory 73 as well as to inputs 81 of a decision complex 72 and also to inputs 79 of a rating complex 71. The experience memory 73 stores the learned relationships between the invariant for key information of the outputs. 82b and / or 76c (possibly also 80) and the correlated invariants for associable afferent information of the outputs 76b of the recognition complex 70. The structural learning by building or reducing causality relations between correlatable invariants of the outputs of allocation units or

234948 8

Allocation levels of the machine can be determined both by means of the experience memory 73 and in "adaptive" assignment complexes 70; 71; 72 are implemented. The outputs 76 '; 76 "of the associated information memory 73 as afferent first and / or second type predictions are either coupled (in case 76") to disjoint inputs 79 of the evaluation complex 71 or (in case 76 ") to corresponding outputs 76d of the recognition complex 70 via logical Disjunctions 78 (according to the method of associative coupling, patented by WP 145 810).

The outputs 80 of the evaluation complex 71 are coupled via special couplings 80c (eg, for recognition motivations) to inputs 75 of the recognition complex 70 and via couplings 80d (eg, for decision motions) to inputs 81 of the decision complex 72 and possibly via couplings 80a (e.g. B. for affects) connected to inputs of an action complex. Possibly, specific outputs 80b of the evaluation complex 71 for controlling structural learning and / or association may be learned from experience with logical conjunctions, gates or gates (76'b; 76'c; 82'b) to affect the signals 76b; 76c; 82b for key information or associative information of the experience memory 73 are coupled. The outputs 82 of the decision complex 72 are connected to inputs for key information of the experience memory 73 via special couplings 82b for decision results (i.e., referencing schemes). Via possible couplings 82c; 82d, certain decision results (efference copies) for influencing the recognition and / or evaluation can be fed back to inputs 75 of the recognition complex 70 or to inputs 79 of the evaluation complex 71 so that special reference signals are available for the decision-dependent evaluation of the reafference based on the reafference principle.

Decision decision signals 82e may suppress information associated with logical conjunctions 77. From outputs 82a; 80a; 76a of the machine become situation dependent signals according to the decisions, evaluations and / or

234948 8

Judging or recognition results of the intelligent machine generated. The efference signals of the outputs 82a of the decision complex 72 are verified to initiate actions or behavioral programs (eg, by means of an action complex 57 for controlling effectors) in accordance with decisions made by weighing eference decisions (82b) with evaluation results (82b). 8Od) according to the associated predictions of the first and / or second kind (76 ¹ ). The coupling of the evaluation and decision-making complex 71; 72 allows, by associating with experience, the finding of optimal decision results of the outputs 82a for the detected situation of the inputs 74 both before and after the machine initiated action (see the description of FIG. 6 and in WP 149 723 and WP 224) 629).

The intelligent automaton according to FIG. 8 has the structure of at least one "adaptive" assignment complex according to FIGS. 5a, 5b. The implementation of "learning" assignment complexes 30; 70; 71; 72 (see Figures 1, 2, 5a, 5b, and 8) is advantageously possible by means of computers (eg, with microprocessors) and experiential or associative memory as described above. The logical functions of the assignment levels or assignment complexes can be stored in programmable lists for the verification of the invariants by value assignments for variables. Such "assignment lists" contain the logical conditions for the associations of invariants corresponding to the inductive and deductive relationships. They can be expanded or changed through structural learning. Programs for interpreting the assignment lists implement the assignment complexes. List-oriented languages (eg LISP) are applicable. The interpretation method was successfully used in the development of an inventor-designed realization example "Learning Homeostat" in 1980. With a microprocessor-controlled display device, multivalued variables with invariants for different recognition results (external and internal situation as well as judgments), evaluation results (drive,

234948 8

Motivation and emotion) and decision results for seven behavioral programs, depending on five classes of input terms. In order to demonstrate an adaptive control with artificial intellect, drive-driven and motivational decision-making was simulated according to a modeled causal network with 53 support information. The developed learning homeostat shows ultra-stability due to its adaptivity with subjective rating, structural learning, and associating predictions.

The sequential execution of the logical operations according to the structurally determined logical and assignment functions of the assignment levels requires a relatively high amount of time. This multiplies by cyclic repetitions of the assignment level operations for the purpose of successive or iterative implementation of the effect of feedback in the hierarchy of allocation levels. This sequential implementation may be too time-consuming for "smart" controllers in real-time operation, especially because of the relatively low processing speeds of microprocessors.

FIG. 9 shows a circuit example for embodiment variants of the programmable logic arrangement according to the invention for implementing hierarchical logic complexes and data flow cancellation on the busses of functional memories, in particular for the implementation or implementation of intelligent automata whose functional complexes 54; 55; 56; 57; 58; 59, "adaptive" assignment complexes 30; 70; 71; 72 and assignment levels 26a; 26b; 27; 28; 29. In order to reduce the time required for logical operations by bit-parallel assignments, at least one allocator 91; 92; 105; 106 for implementing at least one assignment level 26a; 26b used. This is with storing connection units 93; 94; 96; 97 or with at least one connection and connection unit 95; 96; 97; 99 for coupling to at least one programmable control complex, computer or processor 87. The allocator can then be coupled directly to the interface or bus 98 of at least one processor 87 or control complex if, like the sequencer 106, it has at least: an integrated connection and connection unit 95; 96; 97; 99 with the

234 9 48 8

, - 40 -

required interface 98 has.

Line connection unit 93; 94; 96; 97 or connection and connection unit 95; 96; 97; 99 is used to control, setting, influencing and / or connection of the allocator 91; 92; 105 to the interface 98 analogous to an input / output device. It can, for example, by means of registers, flip-flops, memory cells, activatable signal generator, coupling units 67; 67a, gate; Circuits, gates and / or circuits or cells with variable connection, conductivity, effectiveness and / or storage properties can be realized. It is with at least one allocator 91; 92; 105 integrable on a chip or within a unit. They may include means or circuits for the temporally parallel or serial providing and / or for writing and reading bits, signals or invariants i'ür or of at least one associated allocator 91; 92; 105 own. It can also be equipped or connected with coding or decoding devices.

The flexible auslebare structure of the programmable logic device of Figure 9 allows the connection of multiple allocators 91; 92; 105; 106 various embodiments to the interface of at least one processor 87th

The allocators 91; 92; 105; 106 have the logical functions of the above-defined assignment levels 26a; 26b (see Figures 1, 2, 3, 4) and have the character of functional memories. They will with; above-indicated technical means for allocation units ZE ₁₁ , ..., ZE; ZE ₁ , ..., ZE; 25a; 25b and possibly internal and / or external coupling units 67; 67a realized (see Fig. 4).

At least one allocator 91; 92; 105; 106 for at least one assignment level 26a; 26b; 27; 28; 29 is used to form at least one "adaptive" assignment complex 30, function complex 55; 56; 57; 58; 59; 70; 71; 72 or intelligent machines with programmable functions controlled by at least one processor 78, computer or control complex. Instead of an allocator 91; 92; 105; 106 is a network or hierarchy of coupled allocation units 25a; 25b or allocators or assignments

234948 8

levels 27; 28; 29 can be used (see 26a, 3o). To realize a hierarchical multistable system with nested subsystems, eg. As an intelligent machine, or a hierarchy of erf indungsgernäßen logic arrangements, z. As a data-flow computer, is the use of at least one computer (e.g., microcomputer) or programmable logic array (configurable as assignment complex 30, complex allocation level 26a and / or intelligent automaton) instead of at least one allocator 91; 92; 105; 106 possible. In FIG. 9, the inputs 101 of the allocator 91 with associated outputs of the connection unit 93 and the outputs 102 of the allocator 91 with associated inputs of the connection unit 94 are coupled. By integrating these devices, the "integrated" allocator 105 (e.g., on a chip) can be formed. It has parallel or serial transmission channels 107; 108 for input bits 107 and output bits 108 and also for control and / or addressing signals. For the purpose of realizing the interface 98 for at least one control complex, computer or processor 87, the connection unit 95 is connected to the "integrated" allocator 105. It provides the interface control for data input and data output with respect to the mapper 105. The link units 93; 94 and the terminal unit 95 can be integrated into a connection and termination unit 99 (eg on a chip), the latter and the allocator 91 being combined, the allocator 106 (eg on a chip) with interface 98 for direct connection to at least one processor 87. In Fig. 9, a second associate 92 connected to interface 98; 106, which shows another embodiment of the programmable logic device. The inputs 103 and outputs 104 of the allocator 92 are provided with separate connection and terminal units 96; 97, which can be integrated with the allocator 92 so that an allocator 106 (eg, on a chip) with single or dual interface 98 results for connection to at least one processor 87. For interface simplification, a line interface can be realized according to the bus principle. The summary of the data inputs 109 with the data

234948 8

cjängen 110 of the integrated allocator 106, z. B. by bidirectional lines is advantageously feasible. In Fig. 9, the line connection for a possible line interface with the connection 111 is shown. \

The control of at least one allocator 91; 92; 105; 106 for implementing an assignment level 26a; 26b or allocator complex 30 is performed via interface 98 by at least one processor 87, computer or programmable control complex. To the entrances 101; 103; 107; 109 of a selected or addressed allocator 91; 92; 105; 106 true and / or negated BILs, UaLen or signals for invariants of binary or multi-valued voices are provided via the interface 98. About the connection units 93; 94 or connecting and connecting units 95; 96; 97; 99, such bits or data to the inputs 101; the allocator 91; 92, which serve as information representations for the programmed processes and logical operations or assignments, depending on defined couplings for "learned" causality relations between verifiable invariants The input bits of the inputs 101, 103 of an allocator 91, 92 are (usefully quantized in bit groups) time-serially or in parallel in storing connection units 93, 96 (eg with registers) writable.The provision of the "input bits" of the inputs 101, 103 is performed by at least one program, eg Program memory 88 of the computer, ao controlled that fixed and / or changeable or "learnable" (ie buildable and / or degradable) couplings for inductive and / or deductive relationships between the invariants for SLützinformationen or terms hierarchical assignment levels of at least one assignment complex, model or Causal term network. The input provided with sbits represent or characterize sensory, received, transformed, calculated, associated, read, afferent and / or efferent information representations. They are placed in the logical operations or links of the implemented assignment level 26a; 26b included.

234948 8

The "output bits" of the outputs 102; 104 of the used allocator 91; 92 are functions of the logically linked input bits ιοί; 103 of the allocator 91; 92 and possibly depending on internal states of the allocator 91; 92 (analogous to an abstract automaton). They represent verified invariants of the output variables of at least one implemenut-ttc_concentration level 26a; 26b.

üie providing the input bits and the further processing, transmission, transcoding and / or storage of the output bits of at least one allocator 91; 92; 105; 106 are at least one programmable control complex, computer or processor 87, z. Microprocessor, according to the designed fixed and / or variable logical or assignment functions of at least one implemented assignment level 26a; 26b; 27; 28; 29 controlled. The verified invariant output bits of at least one implemented allocation level are provided by at least one control complex, computer or processor 87 from the selected or addressed allocator 91; 92; 105; 106 queried and / or read. They are used to represent transformed, transmitted, afferent, efferent, correlatable, associable and / or associated information representations for assignment results, e.g. B. recognition, Uewertungs- or decision results, written in coded form in at least one data memory 89 and / or experience memory program dependent. This stored data can again selectively, z. By interpreting programmed allocation lists, to read corresponding bits, signals or invariants for manipulating the inputs 101; 103; 107, 109 of the same or at least one other allocator 91; 92; 105; 106 and / or to further process, transmit, output or use as control information.

In the embodiment of Fig. 9, an input / output unit 85 is provided with sensory inputs 83 and outputs 84 for effector signals. It serves to couple peripheral units.

234 948 8

Line Possibly Applicable Interactive Link Unit 86 connects to the interface 100 for other processors, computers, or systems that can be coupled to the programmed logic device. It can contain spaces that are available from several sides to · the ^v transferring data within a multiprocessor or multi-computer system. It is used for interaction with the programmable logic device, eg. B. if the latter is a subsystem or functional complexes of an intelligent machine, which is realized with multiple processors. For the purpose of liberalizing the logical or mapping functions of at least one implemented mapping level a 26a; 26b of the logic arrangement according to the invention, the use of at least one multifunctional, universal or "adaptive" allocator 91; 92; 105; 106 is provided, which is equipped with such variable, redundant, synthesizable, selectable or formable logical functions or links, which by controlled influencing selected inputs 9; 22; 101; 103; 107, 109 of the allocator 91; 92; 105; Can be selected, set, changed, programmed, initiated and / or controlled with signals for true or negated bits or invariants. For example, allocation units 25a; 25b can be used with universal or variably varied basic logic functions (eg with redundant inputs 15, 16, 17, 22), whose functions for specific time intervals are coordinated with the correspondingly programmed input bits of the allocation units 25a; 25b are defined or validated.

By repeated use of a program-dependent controlled, functionally changeable allocator 91; 92; 105; 106 are several different assignment levels 26a; 26b or "adaptive" allocators with logic functions or conditional assignments defined for specific time intervals implemetable.An example of a multifunctional allocation unit 25a; 25b is shown in FIG. The logical functions for verifying the invariants are traceable to the basic functions of the logical (Boolean) algebra. In Fig. 4 is

234948 8

the alternating sequence of conjunction 11; 13 and disjunction 12; 14 for universal use in conjunction with negations 24 for input signals 15a; 16a; 17a and output signals 20 are shown. The required for a given time interval logical functions are program dependent by controlled influencing the inputs 9; 22 of the multifunctional allocation unit 25a; 25b with selected true and / or negated bits for special inputs 15; 16; 17; 18; 19; 22 adjustable ° or formable. With this program-dependent VQrUFKIo "" tion of at least one logical function of an allocation unit 25a; 25b, the implementation of various allocation levels and / or structural learning is feasible.

For at least one allocator 91; 92; 105; 106, an "adaptive" allocator having the ability to learn structurally can be used. He can with at least one internal or external coupling unit 67; 67a or storing connection unit 93; 94; 96; 97, which independently or externally controlled, according to the conditioning, can produce special couplings for constructible, "learnable" relationships between (eg quasi - verifiable) invariants with a fulfilled learning condition and / or degradable or "learned" relationships in the case of Learning to relearn (eg in the case of antivalent relationships) or in cases of insufficient or too seldom affirmation or conditioning.

The constructible or degradable relationships of causality are prerequisites for the implementation of structural learning and association from experience. In designing or programming an "adaptive" assignment complex or intelligent automaton, learnable relationships are conceivable. For these causality relationships, buildable or degradable couplings are within and / or outside at least one implemented or realized assignment level 26a; 26b; 27; 28; 29; 91; 92 assessable, formable and / or programmable, eg. B. by means of the above-characterized experience memory 31; 32; 33; 34; 73; 90, associative memory 58; 59; 90, "adaptive" allocator 25a; 25b

, ₄₆ . 234948. 8

with coupling units 67; 67a, storing connection units 93; 94; 95; 96; 97; 99 and / or interpretable conditional assignment lists in memories 88; 89; 90 (as functional memory). ,,,

The conditions for opening or breaking relationships or links in structural learning or for controlling selected inputs 7; 9; 22; 101; 103; 107; 109 or outputs 8; 10; 102; 104; 108; 110 at least one allocator 91; 92; 105; 106 with true or negated bits or invariants, which z. B. in logical functions or links at least one assignment level 26a; 26b may be either fixed or variable "predisposed" and / or be situation-dependent definable, z. B. with programs oder'Signalen. The above-described structural or functional change in structural learning is according to the conceived conditioning in the learning phase, z. B. by computer programs, with respect to correlatable invariants for sensory, transformed, associable, associated, afferent, efferent and / or received information representations performed. By conditioning inductive and / or deductive relationships of causality with couplings, connections or associations can be constructed or broken down between bits or invariants of these information representations.

Associating from experience, d. H. the generation of predictions of the first and / or second type, as a prerequisite for adaptive recognition and decision (analogous to associative thinking) is a technically simulatable process according to the concept of the intelligent logical structure of the automaton. For structurally storing "learned" causality relationships, distributed and / or concentrated experiential memories (58; 59; 73; 90) or "experiential" programs related to conditionally verifiable invariants of at least one implemented assignment level 26a; 26b; 27; 28; 29 of the programmable logic device can be used. The associative coupling of information (see WP 145 810) when associating is used for at least one associate 25a; 25b, 26a; 26b; 27; 28; 29; 30; 70; 71; 91; 92; 105; 106 with at least one source 43 '; 44 '; 45 '; 46 '; 63a; 63b; 64; 76 'for

234948 8

associated information (Erfahrungs Β, experience memory 31; 32; 33; 34; 58; 59; 73; 88; 89; 90 or coupling units 67a and / or connection units 93; 94; 96; 97; 99) via at least one of its inputs 5; 7; 9; 22; 44 ", 45"; 46 "/ 101; 103; 107; 109 and / or by disjunctive linkage 14; 36; 37; 38; 78 with at least one of its outputs 6; 8; 10; 20; 21; 23; 48a; 49a; 50a; 76d; 102, 104, 108, 110 are hardware-based and / or software-based, such associated information is included in its logical operations or assignments _: which are inductively or deductively correlated to verified key information A source of key information can have at least one output an assigner 6; 8; 10; 20; 21; 23; 44; 45; 46; 76; 102; 104; 108; 110 and / or another source, eg for input signals 47 a; ,

The inventively programmable logic array is both with multiple allocators 91; 92; 105; 106 for at least one assignment level 26a; 26b and with a plurality of processors 87 for the control of the allocators for performing simultaneous operations ausrüs tbar.

The timing of the processes shown in Fig. 7 in the intelligent machine is controlled by means of at least one computer program for the logic device. The causal relations of the logical structures modeled by means of logic arrangement and program (for causal conceptual networks) can be defined by logical functions of the verifiable invariants. The latter can be analytically expressed using basic functions of Boolean algebra. The bivalent logic for the verification of invariants for binary and multivalued variables applies. Associated information can be included in the logical functions as predictions of the first and / or second type in accordance with the structurally stored couplings for learned relationships of causality.

The implementation of the programmable verification of invariants for the output variables of the assignment levels 26a; 26b; 27; 28; 29 can be autonomous from at least one allocator 91; 92; 105; 106 and programmatically performed

234948 8

become. The Impact of Feedback on Deductive Relations 2; 45b; 45c; 45d are by repeatedly driving allocators 91; 92; 105; 106 with signals or invariants corresponding to the information representations to be transformed for its inputs 101; 103; 107; 109 implementable.

The advantageous programmability of the logic array, e.g. by means of special programming languages (eg LISP for list interpretation) permitted by suitably implemented allocators 91; 92; 105; 106 with selectable logical functions (eg, multifunctional or universal allocators), the flexible technical modeling of intelligent logical structures for a variety of applications, and especially the implementation of algorithmic intelligent processes of task solving. At least one processor 87 or computer, which is part of the inventive logic arrangement, can be used independently or in connection with the control of at least one allocator 91; 92; 105; 106 also own operations, z. Arithmetic, statistical, associative, processing, input and / or output operations. It is associated with at least one other processor, computer, associative memory (see WP 224 629), input-output devices, D / A converters, A / D converters, sensors, effectors, interactive or input / output units 85; 86 and / or at least one other logic arrangement according to the invention can be coupled.

Claims (10)

- «- 23Λ 948 invention claim 1. Programmable logic device for the realization of hierarchical logic complexes and data flow computers based on functional memory for use in adaptive systems with artificial intellect, especially for the implementation or implementation of defined intelligent machines, their functional complexes, "learning" assignment complexes and assignment levels (as defined in this patent and in WP 145 436, WP 145 338, WP 149 723 and WP 224 629) and for adaptive process, machine or robot controls for the sensor signal processing which recognize programmable skills for structural learning and situational dependency and / or having asserting associated predictions from experience, characterized by a) at least one allocator (91; 92; 105; 106), "% for at least one assignment level (26a; 26b; 27; 28; 29), having fixed and / or variable logical functions for verifying invariants for binary and / or multivalued variables or for assigning output bits to input bits of the allocator whose inputs (101, 103, 107, 109) and / or outputs (102, 104, 108, 110) are connected to at least one such coupling unit (67, 67a ), Connection unit (93; 94) and / or connection and connection unit (95; 96; 97; 99) which can be integrated with the allocator (91; 92; 105; 106) and with at least one programmable control complex, computer or Processor (87) and / or satisfy the interface conditions for the coupling (98) of the allocator (106) with at least one processor (87), computer or control complex, b) at least one processor (87), computer or control complex having at least one allocator as in a), a program memory (88), data memory (89) and / or experience memory (90) and possibly at least one input / output unit (85 ) and / or interactive coupling unit (86), and for providing such true and / or negated bits or data 234948 θ for inputs (101; 103; 107; 109) of at least one allocator (91; 92; 105; 106) as under a); which represent invariants or values of binary or polyvalent variables for representing sensory, read / transformed, calculated / associated, associable, afferent and / or efferent information, for the purpose that fixed, variable and / or "learnable", d. H. constructible and / or degradable couplings for inductive and / or deductive relationships between the invariants for support information or concepts of modeled semantic networks are ρrogranimabhängig implemented, c) at least one processor (87), computer or control complex, which may be identical to b) and which is used for the purpose of transmitting, transcoding, processing and / or storing such true and / or negated bits or data of outputs (102, 104 108; 110) of at least one allocator (91; 92; 105; 106) as in a) which contains verified binary or multivalued variable invariants for representing transformed, afferent, efferent, associated, associative and / or key information of the assignment results , z. B. recognition, evaluation or decision results, and which reads the output bits of at least one allocator (91; 92; "105; 106) and in.codierter form in at least one data memory (89) and / or experience memory (90) program dependent and possibly places these bits in causal relationship with other correlatable bits according to conditioning for structural learning, and also by selectively storing stored data, providing true or negated bits, signals or invariants for selected inputs (101, 103, 107; 109) at least one allocator (91; 92; 105; 106) as under a) for the program-dependent implementation of fixed and / or variable logic or assignment functions of at least one assignment level (26a; 26b; 27; 28; 29) as under b) can control 234948 8 d) possibly at least one allocator (91; 92; 105; 106) inserted instead of a) which is equipped as a multifunctional, universal or "adaptive" allocator with such redundant, variable, synthesizable, selectable and / or formable logical functions; which is program-dependently set, selected, changed, programmed by controlled influencing of selected inputs (9; 22; 101; 103; 107; 109) of the allocator (91; 92; 105; 106) with signals for true and / or negated bits or invariants , can be initiated and / or controlled so that by repeatedly using such an allocator (91; 92; 105; 106) several different assignment levels (26a; 26b; 27; 28; 29) or "learning" allocators are defined at specific time intervals logical functions or conditional assignments can be implemented, e) possibly having at least one "adaptive" allocator (91; 92; 105; 106) substituted for a) having structural learning capability and at least one internal and / or external coupling unit (67; 67a) or storing connection unit (93; 94; 96; 97) in accordance with the conditioning, independently or externally controlled, can establish effective couplings for "learnable" latent or constructible relationships between verified invariants when the learning condition is fulfilled in the learning phase and / or "learned" resp can destroy or "forget" degradable relationships during relearning or in cases of inadequate or too seldom affirmation or conditioning by dismantling couplings, f) possibly at least one assignment level (26a; 26b) or an assignment complex (30; 70; 71; 72) implemented with at least one allocation unit (25a; 25b) or at least one allocator (91; 92; 105; 106) as under a ), d) and e) and can be embodied as a network or hierarchy of allocation units, for the purpose of forming at least one model of semantic or conceptual networks, a programmable intelligent automaton, "adaptive" assignment complex (30) and / or function complex (54; 55; 56; 57; 59; 70; 71; 72; 73) of an intelligent machine controlled with at least 23Λ9Λ8 8 a processor (87), computer or control complex as in b) and c), g) possibly at least one programmable logic device according to the invention, designed as assignment level (26a; 26b; 27; 28; 29), assignment complex (26a; 30), function complex (54; 55; 56; 57; 58; 59; 70; 71; 72 73) and / or intelligent automaton which is used instead of at least one allocator (91; 92; 105; 106) as in a) and for hierarchical nesting of inventive logic arrangements of at least one processor (87), computer or control complex as in b) and c) is controllable, h) possibly at least one assignment complex, an assignment level, an allocator, a programmable logic device according to the invention, a processor and / or computer, used instead of at least one allocator (91; 92; 105; 106) as under a ) and possibly formed as a recognition complex (54; 70), evaluation complex (55; 71), decision complex (56; 72), action complex (57) and / or association complex (58; 59; 73), which is used for implementation tion of the structural and functional principles of at least one intelligent automaton of at least one processor (87), computer or control complex as in b) and c) is controllable. 2. Programmable logic arrangement according to item 1, characterized by at least one allocator (91; 92; 105; 106) for at least one assignment level (26a; 26b) consisting of at least one allocation unit (ZE--, ..., ZE; , ..., ^Z E _U ; 25a; 25b) whose outputs (10) can be coupled to inputs (9) of the same and / or other allocation units and to whose inputs (5; 7; 101; 103; 107; 109) and / or whose outputs (6; 8; 102; 104; 108; 110) program-dependent selected bits are time-parallel or serial transferable. 234948 θ 3. Programmable logic arrangement according to item 1 and 2, characterized in that at least one allocator (91; 92; 105; 106) or an allocation unit (ZE ₁₁ , ..., ZE, ZE ₁ , ..., ZE _and 25a, 25b) with logic elements, gates and / or threshold elements, fixed, programmable or "adaptive" logic circuits, allocators, Classifiers and / or read / write or associative memories can be realized and that possibly internal or external coupling units (67, 67a) and / or storing connection units (93; 94; 96; 97; 95; 99) are provided, which for control or modification of logical and / or assignment functions of the allocator or the structure or degradation of relationships or couplings between bits, lines, signals, states, data or invariants either program-dependent, controlled or autonomous, z. For example, by conditioning for structural learning, their coupling, conductivity and / or storage properties or can change their effectiveness. Programmable logic arrangement according to items 1 and 3, characterized by at least one allocation unit (25a; 25b) having at least one such logical function for invariant formation, which is implemented with basic functions of the Boolean algebra, eg. Conjunction (11), disjunction (12), conjunction (13) and possibly additional disjunction (14), controllable by means of conjunctive inputs (15; 15a; 17; 17a), e.g. 17a) for negated bits or invariants, and / or disjunctive inputs (16; 16a; 22) and possibly internal disjoint inputs (18) and / or internal convolute inputs (19), and further characterized in that their outputs (20; 20a) may be connected to inverters (24), internal inputs (18; 19) and / or additional disjunctions (14) or disjunctive inclusion units (25b), and for purposes of determination or changing at least one logical function of the allocation unit (25a; 25b) controllable, adjustable, switchable, programmable, variable, expressible and / or experience-dependent adjustable 234948 8 Coupling units (67, 67a) and / or connection units (93, 94, 96, 97, 99) which can be used with selectable, addressable and / or special inputs (15, 15a, 16, 16a, 17, 17a, 18, 19 22) both inside and outside the allocation unit (25a; 25b) and which are integrated, connected or identical to connection units (93; 94; 95; 96; 97; 99). 5. Programmable logic arrangement according to item 1 and 4, characterized - That uls requirement for structural learning or for the formation of at least one conditional assignment, depending on at least one programmable or fixed Lerribonditionung, z. B. on the condition of coincidence or temporal Kontiquität verified invariants or signal events for mutually obtainable information, at least one controllable or "learnable" coupling to implement buildable and / or degradable causal relationships between correct. lable bits, data, addresses, signals, states and / or invariants for variables within and / or outside at least one implemented assignment complex (30; 70; 71; 72; 91), assignment levels (26a; 26b; 27; 28; 29; 91; 92) and / or an allocation unit (25a; 25b; 91), designed, prepared, assessed, designed and / or programmed, e.g. By means of distributed and / or concentrated experience memories (31; 32; 33; 34; 73; 90), associative memories (58; 59; 90), "adaptive" allocators (25a; 25b) with coupling units (67; 67a), storing Connection units (93; 94; 95; 96; 97; 99) and / or interpretable conditional assignment lists in memories (88; 89; 90) and - that the conditions for the construction or dismantling of relationships or couplings, d, h. for structural learning or for controlled manipulation of selected inputs (7; 9; 22; 101; 103; 107; 109) or outputs (8; 10; 102; 104; 108; 110) of at least one assigner (91; 92; 105; 106 ) with true or negated bits or invariants, which z. B. in logical functions or links at least one assignment level 234948 8 (26a; 26b), either fixed or variable and / or definable depending on the situation, eg. B. with programs or signals. 6. Programmable logic device according to item 1 and 5, characterized by at least one "learned", constructed or degraded coupling, connection, association, conditional assignment or causality relationship between bits, data, addresses, signals or invariants of variables as display forms for coded information of at least one source of key information (45; 45b; 45c; 45d; 45e; 60a; 61a; 65a; 76c; 82b) and from at least one source of associative information (43 '; 44'; 45 '; 46 "). ; 63a; 63b; 64; 76 'and 43a; 44a; 45a; 46a; 60b; 76b) which is stored by structural change (structural learning) implementable with - at least one experience memory (31; 32; 33; 34; 58; 59; 73; 88; 89; 90) realized with e.g. Associative memories, read / write memories (89, 90), RAMs, stored lists, "adaptive" allocators (25a; 25b; 91; 92), associative coupling matrices (as in WP G06F / 145 810) and / or distributed or concentrated coupling units (67; 67a) or storing connection units (93; 94; 95; 96; 97; 99), which for the purpose of building or degradation of at least one causal relationship can be conditioned and z. For example, the following technical means can be realized: registers, flip-flops, memory cells, activatable signal generators, artificial synapses, coupling elements, variable resistors or semiconductors, switches, ¹ diodes, logical conjunctions, gates, gates or valves, electrical, optical, magnetic , physical, chemical or biological routes, pathways or cells with variable conductivity, binding, excitation, transmission, reception and / or storage characteristics whose functions or effects are controlled, adjusted, programmed, designed or determined according to learning conditions can be and / or with condition bits (eg, flag bits) or ternary signals for at least one "experience-dependent" or "learnable" program for implementing conditional ones 234948 8 Associations, conditionally variably stored relationships between bits, data, addresses or invariants, and / or conditionally varying logical functions or links, which is destructible (i.e., "forgetful") or recoverable (i.e., re-learnable), e.g. Depending on time constants or falling below or reaching or exceeding thresholds for the conditional probability or relative frequency of signal events for associated information in the condition of coincidence by temporal contiguity or quasi-simultaneous verification of such bits, data, signals or invariants for support information or terms that are related (ie, correlated) with each other or correlatable, or depending on the build-up or degradation of an antivalent (ie, alternative) relational relationship. 7. Programmable logic arrangement according to item 1 and 6, characterized in that for the program-dependent, associative or "learned" inclusion of bits, data, signals or invariants in logical functions or links at least one implemented assignment level (26a; 26b) Such conjunctive inputs ( 15; 15a; 17; 17a; 19) and / or disjoint inputs (16; 18; 22) of at least one allocation unit (25a; 25b) or of an allocator (91; 92; 105; 106) which are denoted by true or negated ones Bits are influenced according to fixed, variable and / or constructed, inductive or deductive causal relationships. A programmable logic device according to items 1 and 7, characterized in that it is capable of "associating by experience" and in that at least one allocator (91; 92; 105; 106) or an allocation unit (25a; 25b) has such disjoint inputs (16 ; 18; 22; 44 "; 45"; 46 "; 44 '; 45'; 46 ') which are used for the disjunctive inclusion or associative coupling of bits, signals or invariants for associated information of specific sources (43'; 44 '). ; 45 "; 46 '; 63a; 63b; 64; 76") into the 234948 8 logical functions, links or assignments, and / or in transmission channels for outputs (20; 21; 23; 44a; 48a; 60c; 6Od; 76d; 102; 104; 108; 110) of the assigner (91; 92; 105; 106) , z. By means of at least one unit for disjunctive inclusion (25b), with logical disjunctions (14; 35; 36; 37; 38; 78) and / or associative coupling natrices (eg for 32 and 36; 58) as in WP G06F / 145 810. 9. Programmable logic arrangement according to item 1 and 8, characterized in that at least one allocator (91; -92; 105; 106), an allocation unit (25a; 25b), an assignment level (26a; 26b; 27; 28; 29) and or an association complex (26a; 30; 70; 71) having at least one associated information source (43 "; 44"; 45 '; 46 "; 63a; 63b; 64; 76'), eg, experience memory (31 ; 32; 33; 34; 58; 59; 73; 88; 89; 90) or coupling units (67a) and / or connection units (93; 94; 96; 97; 99) via at least one of its inputs (5; 7 ; 9; 22; 44 "; 45"; 46 "; 101; 103; 107; 109) and / or by disjunctive links (14; 36; 37; 38; 78) with at least one of its outputs (6; 8; 10 ; 20; 21; 23; 48a; 49a; 50a; 76d; 102; 104; 108; 110) is hardware and / or software coupled so that such associated information is included in its logical operations or associations and / or its output bits (20; 48a; 49a; 50a; 76d) are disjunctively linked, w which are correlated inductively or deductively with verified key information, at least one of its influenced outputs (6; 8th; 10; 20; 21; 23; 44; 45; 46; 76; 102; 104; 108; 110) and / or its inputs (43; 47a) may act as a source of key information (45b; 45c; 45d; 45e; 60a; 61a; 65a and 43b; 43c; 43d; 43e; 76c; 82b), respectively. Programmable logic arrangement according to items 1 and 9, characterized in that for the purpose of preventing (inhibiting) associated information or association from experience and for the purpose of controllability of structural learning by promoting or preventing the buildup or dismantling of Causality relationships, eg. B. by means of signals 234948 8 for valuation results, at least one transmission channel of correlatable information, e.g. For key or associative information, through logical conjunctions, gates, valves or gates (39; 39 '; 39 "; 40; 40'; 41; 41 '; 42; 42'; 61'e; 62'e; 62 "h; 62 ', g; 76 * b; 77; 82'b) is blockable or unblockable. Programmable logic arrangement according to item 1, characterized in that the control complex associated with at least one allocator (91; 92; 105; 106) contains at least one processor (87) or computer and that possibly at least one input / output unit (85; ), the situation-dependent input signals and signals of transformed, transmitted, calculated, afferent, efferent, associative; or associated informational presentations transmits, transmits or receives and / or stores, and possibly with at least one D / A converter, A / D converter, sensor, probe, receptor, effector, actuator, at least one visual or phonetic input or output device , Object or image recognition means and / or interactive unit (86) for coupling to at least one other system. 12.Programmable logic arrangement according to item 1, characterized in that at least one processor (87) or computer in connection with the control of at least one allocator (91; 9.2; 105; 106) in the implementation or realization of at least one assignment level (26a; 26b ), Assignment complex (30) and / or Intelligent Automata own operations, eg. Arithmetic, statistical, associative, processing, storage, input and / or output operations, and / or with at least one other processor, computer, associative memory, input / output devices, and / or at least one other inventive logic arrangement, for. B. by means of a unit (86) for interactive coupling, may be connected. 234948 8 Programmable logic arrangement according to item 1, characterized in that it is designed and programmed to be a technical means for implementing at least one such intelligent automaton which, in accordance with its basic processes of recognizing and judging, applying and deciding, in connection with associative predictions by structural learning, in the following functional complexes can be disassembled and in principle, the ^expansion] approximately enabled logical structure of at least one "adaptive" association complex (26a; 30) has: - Detection complex (54; 70) with inputs (52) of the machine for z. Sensory signals, having inputs (62a; 63a; 63b; 6Oe; 61d; 69) for signals from the evaluation complex (55), association complex (58), recognition complex (54), decision complex (56) and / or action complex (57) and with outputs (60; 60a; 60b; 60c; 6Od; 6Oe; 65; 65a; 65b) for generated recognition results (efferences), judgments, associators and key information for associative propositions of the first kind, - evaluation complex (55; 71) with inputs (6Od; 61b; 62d; 64; 68) for signals from the recognition complex (54), decision complex (56), evaluation complex (55), association complex (59) and / or target or schedule specifications and with outputs (62a; 62b; 62c; 62d; 62e; 62f; 62g) for generated evaluation results, e.g. Eg for impulses, motivations, affects or emotions and possibly for preventing structural learning and / or association from experience, Decision circuit (56; 72) having inputs (60; 62b; 61c) for signals from the detection complex (54), evaluation complex (55) and / or decision complex (56) and outputs (61; 61a; 61b; 61c; 6ld; 6Ie ) for generated decision results and z, B. key information (61a) for associative recognition results as predictions of the second kind, control signals (6Ie) for structural learning and / or association from experience and feedback efference copies (61b; 61c; 6Id), - possibly action complex (57) with inputs (61; 62c; 66; 234948 8 53a) for decision results, evaluation results, recognition results and / or feedbacks from effectors and with outputs (53, 69) for generated efferent signals (efs), e.g. Possibly for feedback control of effectors, and possibly feedback efference replicas (69), possibly at least one association complex (58; 59) or experiential memory (73) having inputs (6Gb; 60c; 65b) for associative afferents or recognition results, with inputs (60a 61a, 65a) for afferent and / or efferent key information and with outputs (63a, 63b, 64) for associated afferent information as predictions of the first and / or second type correlated to verified key information. For this .. JL ... pages drawings