N
ANOBIOLOGY, Volume 2, Number 2, 1993, pp. 99-107
On the Information Processing Capabilities of the Brain:
Shifting the Paradigm
SIMON Y. BERKOVICH
Department of Electrical Engineering and Computer Science
The George Washington University
Washington D.C. 20052, USA
Explanation of the performance of the human brain does not seem to be possible without shifting the conventional paradigm beyond the framework of the established scientific view on the physical world. This problem is treated on the basis of a cellular automaton model which represents the material formations as traveling wave solutions. Besides that the distributed cellular automaton mechanism creates fast propagating diffusional activities which may be involved in information processing by the holographic principles. A surprising implication of this possibility is that cognitive information processing can be organized outside of the brain using the cellular automaton background of the physical world as a holographic medium.
INTRODUCTION
Progress in understanding the anatomical elements of the brain elucidates in the first place its sensory input/output operations. But little is known about the cognitive information processing between input and output which represents the essence of the machinery of the brain. Apparently, a unified theory of sensory and cognitive processes based on the development of neural network paradigm (see [1]) can be primarily successful in explanation of the input/output operations. As to the internal processing in the brain, it requires enormously more computational power which cannot be achieved without employing some new fast "hardware". In the work [2] it has been suggested that the additional "hardware" subserving the neural networks activities comes form the cytoskeleton, highly ordered constructions of filamentous protein polymers in the interiors of the neurons. Information processing with this construction can be organized by means of intraneuronal cytoskeletal elements such as microtubules involving physical mechanism of conformational motions of the polymers. This type of information processing provides an essential enhancement to pure neural networks.
However, understanding the highest cognitive functions of the brain is apparently related to a drastic paradigm shift in its organization as a whole. Being the most sophisticated constituent of Nature, the human brain could require such a mechanism that its realization may inadvertently lead to a modification of the entire view of the universe. A generality of this kind should raise a substantial amount of critique, greater than that which is normally associated with a paradigm shift [3]. However, trying to understand the brain by incremental adjustments of the existing paradigms may be condemned to Sisyphus labor. So, what can modern science say about the nature of mind far beyond the ancient philosophers?
An approach to a radical change in the paradigm of the organization of the human brain and its place in the physical universe has been described in the book [4]. Starting with the concepts of a holographic mechanism of information processing in the brain by Karl Pribram and the interpretation of quantum mechanics by David Bohm this book develops a global integrated view on the transcendental phenomena of Nature. A convincing foundation for the presented view would be obtained if possible principles for realization of the "holographic universe" were at least outlined. Otherwise, such a paradigm appears as an unsubstantiated variation of popular metaphysical speculations.
This paper advocates a hypothetical cellular automaton construction which may provide a possible "hardware" implementation of the holographic universe concepts. Although strange at the first sight the emerged paradigm illuminates a conceivable direction where a coherent explanation of the brain functionality may be found.
CHARACTERIZATION OF THE BRAIN PERFORMANCE
Since the brain is assumed to be a computing machine its organization is subject to a concrete engineering analysis. A sound technical design has an objective to meet system's specifications with an economical solution. In the case of the brain this implies finding a single operational mechanism for all its enormous capabilities. Those include tremendous processing power, virtually unlimited and indestructible memory, striking fault-tolerance and reliability, diversity of functions under a versatile combination of centralized and decentralized control. Here is given a brief review of the very well known facts regarding the basic performance characteristics of the brain.
Processing Power
The processing power of the system of neurons in the brain can be roughly evaluated by the number of events which may occur in this system per second. The number of neurons is about 1010 and their switching time is about 10-2 sec, so the number of events per second is about 1012 . This figure is comparable with the number of operations per second in massively parallel computer systems approaching the teraop barrier. Thus, the information processing power of the system of neurons does not drastically exceed
that available through modern microelectronic technology. In the expanded construction suggested in [2] the number of binary events per second may reach 1023 to 1025. However, as in all massively parallel systems a problem arises whether a substantial portion of this estimated raw computational power can be effectually utilized.
Memory Capacity
The capacity of the long term human memory is virtually unlimited. According to von Neumann [5], estimated by the amount of information which can be transferred to a human brain during its lifetime, the lower bound of this capacity is about 2.8× 1020 bits. To be stored in the brain of about 103 cm3 this requires density of informational storage about at least 3× 1017 bits/cm-3. The time of content-addressable retrieval is rather short and essentially independent from the amount of stored information. Once recorded, information in the brain is supposed to be retained permanently. Thus, images don't fade with time and can be easily recognized over decades.
Reliability and Fault-tolerance
Failures of individual neurons are due to physico-chemical processes similar to those in discrete electronic components, so both are characterized by approximately the same failure rate, about 10-7 1/hour. This implies that in 100 years about 10% of the neurons fail in arbitrary places of the brain. There are also many others internal and external causes of damage to the neurons. Despite all of this, the human brain, at least for its highest cognitive functions and mental abilities, does not seem to exhibit a noticeable degradation with normal aging or even with some extensive destructions. Conventional fault-tolerance techniques with fault detection and isolation are not applicable for realization of such a flexible structure.
Algorithmic Effectiveness
The human brain carries out with ease a great variety of algorithms having quite different computational schemes, for example, visual perception, speech recognition, translation from one language into another, playing chess, making decisions in uncertain situations, reasoning on philosophical matters etc. Parallel processing arrangements, if efficient, are highly specialized. It is very unlikely that the mechanism of the brain is composed of partial contributions of different computational schemes. For this reason, neural network schemes being efficient for a narrow class of recognition problems, might not be involved in the major algorithmic operations of the brain. The algorithmic universality of the brain has to be ensured by a universal robust computational model.
Interrupt Control
The capability of interrupt is vitally important for information processing system dealing simultaneously with a number of different tasks, especially in a real-time environment. Exercising an interrupt involves a number of miscellaneous tasks: monitoring the requests, identifying the cause, gracefully stopping one process and saving its status for a future resumption, starting a new process and so on. Managing many jobs concurrently, human brain apparently handles the requests for interrupts very effectively wherever they come from, external sources or internal impulses. Consider, for example, such an ubiquitous situation of a person driving a car, listening to a radio, talking to passengers etc; at the same time, the brain of this person must process information regarding unexpected emergencies, physiological functions of the body and much more. The fact that most of this information processing is done "automatically" implies great flexibility in combining decentralized and centralized control. The centralized control for the brain means execution of global operations which are based on the state of the memory as a whole rather than solely on the locally acquired signals. The property of consciousness is crucially determined by the efficiency of integration of decentralized and centralized control.
DESIGN ANALYSIS
As soon as we adhere to the primitive assumption that the brain is just an ultra high performance computer to understand the brain is to reveal the cardinal principles underlying its structural design. The speed of electrochemical processes in the brain is obviously very low. Therefore, to uncover the design of the brain one would turn attention towards seemingly powerful principles of massively parallel computing. However, the perceived potential advantages of parallel computational schemes are fragile and fade away confronting numerous algorithmic and structural impediments. Decades of intensive research definitely show that the sustained productivity of the developed massively parallel multiprocessor systems can just marginally exceed that of a powerful uniprocessor. This fact is rather insensitive to variations in the architecture of these systems. Many workers cannot admit this situation and the results of performance evaluation of parallel computer systems are often misrepresented [6].
The popular appeal to the analogy with the brain as a justification of the validity of massive parallelism is unwarrantable. Since the formal neuron is an algorithmically complete device, so a system of neurons can implement an arbitrary algorithm. However, the question is whether such a system can implement all the required algorithms in real time, i.e. in a fraction of a second. There are no grounds to expect that any yet unknown arrangements of the slow neurons could bring the productivity of a computing system to the point where it can even be compared to that of the brain. Unfortunately, the effectiveness of various computing systems is difficult to quantify since there is no good lower bound theory for parallel algorithms. But it seems obvious that the productivity of the brain exceeds that of any system of neurons irrespectively of its size, structure, or way of operation by manymany orders of magnitude.
In attaining high productivity of information processing there is no substitute for fast switching elements. As a matter of fact, the history of computer technology clearly indicates that most of the progress in the productivity of information processing is due to increase in elements speed rather than to anything else. Sacrificing some productivity, other qualities of information processing systems can be developed through sophistication in their structural organization.
Three basic issues in the conceptual design of an information processing system should be addressed: architecture, software, and hardware. There is no possible way how the brain can achieve high computational power without an extremely fast "hardware". Therefore, employing a new, may be yet unknown, rapid phenomemon is necessary for the very subsistence of the information processing in the brain.
The organization of the brain can be adequately described by attracting the holographic mechanism (see [7]). Thus, the holographic mechanism naturally explains such a fundamental property of the brain as associative retrieval in the presence of noisy distortions. A thorough exposition of this subject and a considerate argumentation. In support of different aspects of a holographic organization of the brain are given, in particular, in [8]. The necessary condition for the development of the physical world is that it must support the existence of intelligent life, the so-called anthropic principle (see, e.g., [10]). In view of further indications on a close interrelation of the brain with the constitution of the physical world it is interesting to note that the holographic mechanism leads to the requirement that the space of perception has to be three-dimensional [9]. By virtue of the anthropic principle this implies the three-dimensionality of the physical space.
An effective associative holographic memory allows the organization of "software" to be simplified because the implementation of algorithms can be content-driven. Being composed of sequences of "atomic" operations of associative accesses to the holographic memory, an algorithm can recover a subsequent operation from the information contents of a previous one. As soon as these atomic operations are self-determined and distinct they can be interleaved in different algorithms. This ensures effective interrupt facilities for multiplicity of concurrent unrelated tasks.
DISCUSSION ON THE IMPLEMENTATION ISSUES
The problem with the holographic model of the brain is that its implementation requires a wave mechanism which is not at hand. Electromagnetic waves cannot constitute the functionning mechanism of the brain, other supposed types of wave-like activities, such as ion-displacements waves [11], are slow. A faster wave mechanism may be looked for among various phenomena in the cytoskeleton constructions like conformational movements of the protein molecules in the model [2].
However, against the possibility that any conceivable phenomenon at the level of material formations of the physical world can provide a wave mechanism for the brain two basic objections can be raised. First, due to the speed of light limitation a traversal through the brain will take a significant amount of time, so in a brain of a linear size of about 10 cm, no physical mechanism can execute more than around 3× 109 global operations per second. In contrast to local operations those are crucial for the functions of mind relying on consciousness. Another objection comes from the reliability aspect, it is unclear how material formations can provide durable virtually indestructible storage of a tremendous amount of information.
So, whatever the reasons for this conjecture may be, we assume that there should exist a fast wave-like mechanism in Nature beyond the conventional material processes of the physical world. Actually, the existence of a faster-than-light phenomenon in the form of a "spooky" action-at-the-distance might be reluctantly acknowledged. This problem has been brought up long ago in conjunction with the theory of gravitation. A remark in [12] regarding the propagation of gravitational influences with the speed of thought can be considered as a first hint on the relation of this phenomenon to the mechanism of the brain. The most apparent evidences of the action-at-the-distance represent long range correlations in quantum mechanics as have been observed in violations of Bell's inequality (see, for example, [13]). A fast propagational mechanism is required, in particular, by the inflational scenario of the Big Bang; according to this scenario at the beginning there was a brief period of extraordinary rapid expansion when a spacetime organizing factor has spread through the observable universe in about 10-30 sec [14]. In this work it is assumed that the wave-like holographical mechanism of the brain is provided by the fast propagation of the action-at-the-distance in the physical world.
The suggested paradigm relies on our cellular automaton model of the physical world [15, 16]. This cellular automaton model represents a grid of nodes containing circular counters which maintain mutual synchronization by a distributed operational rule of weighted phase averaging. According to our interpretation, the physical world is a reflection of various aspects of this activity. The considered model has two basic classes of solutions: ordinary diffusional solutions and traveling wave solutions propagating in a helicoidal form. The latter provide a spectrum of formations which exhibit characteristic features of the elementary constituents of the material world. Thus, for example, there are two dual types of solutions with the opposite sense of rotation corresponding to matter and antimatter, the property of inertia is maintained by the driving mechanism of the model, there is an upper limit on the propagational speed of the traveling wave solutions determined by the congruence of translational and rotational motions, the traveling waves solutions undergo intermitting states which may be interpreted in terms of the wave-particle duality [17].
However, a new interpretation of known physical phenomena in the framework of the cellular automaton approach is not of primary importance for the explanation of the brain. What is lacking for the advocated principle of holographic organization of the brain is a new wave-like phenomenon. The essential feature of the presented cellular automaton approach is that besides the material world of traveling wave solutions it provides a fast operational background of diffusional solutions. The diffusional processes have a paradoxical property of "instantaneous" propagation incorporated in the parabolic equations which are used to describe these processes.This fact implies that this description is not complete and the diffusional processes may involve some fast wave-like propagational factor [18]. The propagational mechanism of the diffusional processes is different from that of the helicoidal traveling waves, therefore the operational background of the physical world is not affected by the speed of light limitation. It has been shown in [15] that this model can be initialized forcing a pinpoint blockade of the mutual synchronization process which causes generation of helicoidal formations interpreted as a Big Bang creation of matter. This process is preceded by a central symmetric diffusional process which can be identified with the inflationary stage of the Big Bang. Spreading of this diffusional process through the observable universe will occur in a time period of about 10-30 sec. The diffusional processes other than the central symmetric propagation are not necessarily associated with the creation of material formations and may be involved in less perceptible action-at-the-distance effects. In this work, we assume that the rapid wave-like spreading of the diffusional solutions constitute the missing mechanism for the holographic model of the brain. The possibility of a linkage between holography and diffusional processes has been considered in [11, 19].
The crucial question in this development is where can the holographic information be stored. Remember, that the material formations of the physical world are traveling wave solutions relocating through the cellular automaton grid. It would be effective if the holographic storage of the information is organized in the nodes of the cellular automaton grid. Such an organization being similar to that in solids, as analyzed in [20], can enjoy to the full extent all the benefits of the content-addressable holographic mechanism. The unusual aspect in the suggested paradigm is that the main information processing should take place outside of the brain. The role of the material formations of the brain is to provide an access to these facilities. With the arrangement of accessing to the information outside of the brain the inscrutable properties of human memory - unlimited storage capacity and reliability - becomes immediately clarified. The fault-tolerance can be easily provided because what is required from the brain in this case is not to process the information in a fault-tolerant mode (a very complicated task) but rather to ensure a much simpler task – a fault-tolerant access to outside facilities.
According to the suggested paradigm the cellular automaton grid underlying the physical universe should serve as a common storage for the whole multiplicity of the brains in the world. Figuratively speaking, the brain, is a "terminal" rather than a "computer" itself. In favor of this possibility speaks the fact that the patterns of electrical activities and energy dissipation in the brain are not substantially affected by the type of information processing. Although, as soon as we appeal to unknown physical phenomena, it might seem simpler to try to get an explanation of the organization of the brain remaining within its physical boundaries. However, with the given cellular automaton interpretation of the material world this possibility is less likely. Also, the volume of the brain, about 103 cm3, may not accommodate a sufficient storage capacity. Thus, according to the results [20], a holographic mechanism with a wave length l = 1 m m can store about 1012 bits·cm-3 which is much lower than the above estimated value of 3·1017 bits·cm-3. Increasing the density of information storage by having a holographic mechanism with a substantially lesser wavelength would cause problems in establishing connections with the sensory structures at the level of material formations.
The information processing activities outside of the brain can play a role of an organizing source for the organism. The innate state of the brain does not appear to be a tabula rasa, some of the facilities of the brain cannot be acquired from its development. A particular example of this situation represents the effect of phantom limbs which can be perceived by a brain without ever having a sensory input [21].
The cardinal question determining the viability of the suggested paradigm is how this construction can simultaneously serve a tremendous amount of users such as the number of human brains. The organization of a multiuser environment with a holographic mechanism can be based on the principle of code division multiple access which has been recently introduced as a communication technique [22]. With this technique the multiple access is achieved by assigning to each user a unique code; users transmitting their messages modulated by this code can access the common holographic medium asynchronously and with minimal interference. Each brain must have a unique access code as well as each individual has a unique immunological response or a fingerprint pattern. This unique access code may constitute a part of the biological uniqueness of an individual.
The basic function of the neuronal apparatus of the brain is to modulate and demodulate messages to the outside holographic medium using its unique access code. Unavoidable imperfections in getting completely distinct access codes may lead to some cross-correlations among memory contents of different current or past users. Being rare and subjective, these cross-correlations transpire at the level of material world as inconclusive observations of extra-sensory perceptions and deja vu reminiscences.
The modulation and demodulation operations for access to the information storage can be performed by neural net kind of computations. Those involve sensory input/output procedures which can be associated with some pre-processing in a neuron supporting construction of the type described in [2]. The activities involving the neuron supporting construction may be essential for integration of various sensory signals, like overlapping visual images from different eyes or combining stimuli from various inputs. The connections among neurons themselves simply do not provide enough communication bandwidth. Employing slow electrochemical processes, the input/output procedures take a tangible fraction of a second. The high performance cognition of the brain comes from the information processing capabilities of the content-addressable holographic mechanism in between.
CONCLUDING REMARKS
Past chess champion of the world, Michael Tahl, once said that to win in chess is very simple - you just have to play stronger than your opponent. Paraphrasing this idea, we can say that natural intelligence is stronger than artificial intelligence simply because it employs more computer clock cycles. Any constructive explanation of the enormous information processing capabilities of the brain must acknowledge the existence of a very fast computational mechanism. Finding this unknown information processing mechanism in Nature requires a penetrative insight into the essence of physical phenomena beyond the descriptive approach of modern physics.
The suggested paradigm presents the brain as an apparatus which can provide sensory input-output facilities while its main cognitive information processing functions are performed outside in a holographic multiuser environment. The mechanism for this organization comes from the underlying cellular automaton model for the phenomena of fundamental physics. Besides generating the formations of the material world, the cellular automaton mechanism reveals in a less perceptible way as an action-at-the-distance through a very fast wave-like propagation of a diffusional activity. This activity is supposed to play key role in the organization of information processing in the brain.
The difficult issue in reconciling the proposed paradigm with our intuition is to accept the concept of the divisibility of the time. From the standpoint of human perception it is easier to accept the divisibility into smaller parts of material objects and space segments rather than the division of time intervals, because, to be meaningful, the latter must be filled with some kind of functionalism. The time in the material world is what is shown by a clock - a certain periodical physical process. The duration of physical processes is filled in with the ticks of the digital counters of the cellular automaton mechanism. The slow down of the time in moving systems, i.e. in relocating cellular automaton formations, is an objective phenomenon which is due to the enlargement of transition paths increasing the number of cellular automaton ticks required for the execution of the same physical process. The computational operations in the brain develop in discrete cellular automaton ticks rather than in physical clock cycles of the time perceived as a continuous entity at the level of material formations.
In this work, it is hypothesized that the very fast spreading diffusional cellular automaton activity represents the missing wave mechanism in the otherwise triumphant holographic model of the brain. This diffusional activity is not affected by the speed-of-light limitation of the material formations and thus appears in the physical world as action-at-the-distance. The time characteristics of the action-at-the distance process, "the speed of thought", can be estimated as follows. From the inflationary scenario of the Big Bang a central symmetric diffusion can spread through the observable universe in 10-30 sec. The human brain could access these facilities through the construction of the type [2] which can serve as an interface capable of operating with the estimated frequency of about 1023 sec-1. Therefore, roughly speaking human brain can be considered as a computer capable of performing around 1023 so global content-addressable operations per second. Such a tremendous computational power may be sufficient for the explanation of the fundamental functions of the brain, like emotions, consciousness, and the ability to think.
The proposed paradigm provides "engineering" grounds to explore the philosophical outlooks incorporating transcendental informational phenomena. The outside holographic information processing mechanism can be also employed by all highest animals. Presumably, the ability to access this mechanism determines the difference between the animated and inanimate objects in Nature.
REFERENCES