Book: Natural and Artificial Intelligence by A. de Callata˙

- I have developed a brain model described in my book: Natural and Artificial Intelligence: Processor Systems Compared to the Human Brain, (1986) North Holland (Elsevier), Amsterdam. (500 pages, in quarto)
- (A Japanese translation is published by Maruzen Co., Tokyo).
- The book reviews the main features of computers (hardware and AI software) and robots.
- It develops and sizes special types of neural networks.
- It reviews the knowledge on biological brains relevant to the functional models.
- It describes a large-scale functional brain model.

mybook2.jpg (40790 bytes)
- Thought requires a stable memory to make repetitive deductions.
- The memory stored in the weight of the neural networks cannot achieve this persistence, as the synaptic weights get spontaneously modified or tend towards standard sizes (the lifetime of a modification could be as short a day).
- The model uses the algorithms successful for explaining the cognitive functions of the brain, such as the AI deductive algorithms based on production rule systems.
- The all-or-none mechanisms used in digital computers are the only devices on which convincing demonstrations of intelligent behavior have been shown.
- The purpose of the model presented in the book is to force digital operations within an organization not currently falsified by the current knowledge on the biological brain.
- The brain model suggested is opposed to several current ideas, such as the implicit view that a plastic, adaptive nervous system cannot make all-or-none decisions in a neuron or in a synapse.
- The problem addressed is not to find deep unifying theories but to understand how apparently incompatible simple systems can be integrated.
- In the engineering approach, the task is to integrate hybrid devices into a working system.
- This design method is called large-scale architecture and is not addressed by the reductionism methods.
- The cover of the book shows several functions that are all required for a comprehensive brain system (an updated colored version is shown below).
- Their integrated operations are compatible with the neuroanatomical brain connectivity.

brainbook.gif (6719 bytes) - After the first edition, I found necessary to first destroy the misconceptions preventing the model acceptance.
- A second edition includes a prologue (130 pages) about misconceptions (a list is given below)
- A main misconception is a narrow view on the adaptive, plastic nervous system preventing to see the biological winner-takes-all functions.
- Another misconception assumes that any logic must be strictly consistent. This prejudice prevents the use of any deductive model, making impossible a comprehensive view of the brain operations.
-The enlarged annexes (51 pages) are an update on neuroscience and on the corresponding features of the suggested brain model.
- The expanded edition was published in 1992.
- de Callata˙ A. (1992) Natural and Artificial Intelligence: Misconceptions about Brains and Neural Networks, New, expanded Edition, North Holland (Elsevier), Amsterdam (690 pages: hardcover or paperback).
  Link to my book at Elsevier site

Misconceptions about brains and neural networks
- (Pages P4 and P5 of the book).
- In the prologue, I try to shake the foundations of many current ideas.
- Should thought have a unifying principle?
- Should neuronal networks be homogenous?
- Are all-or-none switches biologically impossible?
- Must memory disappear in biological organisms continuously modified?
- Is the plasticity of neuronal maps incompatible with computer models?
- Are the neuronal computations mostly based on inhibition and excitation?

- Is the topographical organization of brains necessary everywhere?
- Do we forget our previous habits and beliefs after having changed them?
- Are our behaviors frequently perturbed by noise?
- Is the large variance of our movements incompatible with digital systems?
- Does machine behavior have less variance than that of animals?
- Does adaptive learning exclude irreversible memorization?

- Is symbolic computation incompatible with approximate reasoning?
- Is combinatorial explosion of cases unavoidable in symbolic systems?
- Is information distribution more reliable than redundancy?
- Do Lashley's experiments prove that brain memory must be distributed?
- Is addition of learned events impossible in conventional neural networks?
- Are grand-mother neurons impossible in the real nervous system?
- Are Gestalt phenomena incompatible with symbolic processing?
- Are attractor neural networks more reliable than grand-mother networks?

- Is instant learning rare?
- Are generalized rules necessary for reasoning?
- Is AI a rigid logic method?
- Is rote learning incompatible with understanding?
- Does intelligent classification need a teacher?
- Can classification be a continuous operation?
- Can a rhythmic system be non-oscillatory?
- Is the variance of reaction time incompatible with a rhythmic control?

- Does natural intelligence infer instead of deciding?
- Is making decisions a complex algorithm?
- Do decisions depend on the way the quantum wave collapses?
- Must we find mathematical formulae explaining thought?
- Is document retrieval a complex algorithm?
- Is reasoning by analogy a complex algorithm?
- Is intention a mental concept not implementable in hardware?
- Is consciousness not understandable as a mechanism?

- Do the limitations of expert systems prove that AI is an inadequate model?
- Do Gödel's and Turing's limitations of mathematics and computers prevent natural intelligence by machines?
- Are the primary elements of brain knowledge permanent things?
- Must we first find how similarities are computed?
- Is long-term memorization preceded by short-term memorization?

- Can parsing be a continuous operation?
- Is human behavior not stereotyped?
- Is spontaneous human behavior frequently efficient?
- Is human behavior optimized?
- Is it impossible that simple mechanisms combined produce intelligence?

- I explain in the book why the response to all these questions might be "no".

Back to contents    Best printing results when the browser fonts are scaled for size 10 in print.
Copyrighted: Armand de Callata˙, 1999, XLKL  Base location