Summary of "An Evolutionary Model of Human Intelligence"

An Evolutionary Model of Human Intelligence

By Chris Reynolds

Draft Summary (full paper to follow)

While there is a vast amount of published scientific research about the structure of the human brain, how it evolved, and what it can do, there is a significant gap in our knowledge about it. Basically we evolved in a complex environment and there is no adequate mathematical model of the “complex biological computer” in our heads. What is missing is an explanation of how the activity of single neurons evolved to support the intellectual performance of the human species.  

This paper lays the foundations for a predictive evolutionary model which suggests why there was a significant spurt in human intelligence, compared with animals, and provides an explanation for human brain information processing strengths, and also its limitations, such as confirmation bias.

The proposed model is based on very general decision making nodes which are mapped onto an infinite recursive neural network. This network can be considered the biological equivalent of the infinite numeric array of numbers which is the logical basis for the well known deterministic stored program computer model. Within the model the decision making nodes exchange messages, and in theory all nodes and messages can be broken down into collections of simpler nodes and messages. Individual nodes may be anything from a single neuron, via complete brains, to a human committee, or even man-made tools such as computers. Because the network is infinite, it is capable of representing every neuron of every animal and human brain that has ever existed – and the messages exchanged between them. Evolution from the simplest animal brains into the far more powerful human brain represent a pathway through the logically simple (but very infinitely extensive) network of nodes over a period of about a billion years.

The critical feature of the model, which makes it a complex model, is that virtually all the information needed to construct the deterministic network has been irretrievably lost or is otherwise inaccessible. For this reason the model allows nodes to morph between being deterministic or complex depending on context, and what is known of their history.

Not all aspects of brain evolution are covered. The model describes the way information stored in the network of logically identical nodes is used to make decisions. It is not directly concerned with the physical form of the nodes.  However it is often useful to consider, for example, how the brain is physically constructed and that neurons occur in brains which have a limited life time. Because significant research has been published on how neural networks can learn to recognise patterns the paper does not consider alternative algorithms for such trial and error learning, but simply assumes that in evolutionary terms it is an expensive process. Instead the paper concentrates on the ways language is used by humans to construct a more effective, and significantly more efficient,  decision making network based on the same basic structure used in animal brains. 

The paper is divided into the following sections:

(1)   The Evolution of Species and Trail and Error Learning

The infinitely recursive network model is described in detail,  and used to map out the evolution of the species as a continually growing network which passes genetic information from generation to generation. The paper then looks at animal brains as short-lived self-contained learning packages where the information they learn, by trial and error, is lost when the host animal dies. It considers the evolutionary implications of such loss on the functionality of the brain and it is clear that what was needed was a low cost information processing system designed to be discarded, and the "blind watchmaker" of evolution did not plan an expensive, reliable and mathematically sophisticated super-intelligent brain. Finally the impact of the limited exchange of information between generations in social animals is considered, up to, but not including, the impact of tool-making and language.

(2)   Information Flow within Brains

A brief description of how information can be stored in a network, together with simple pattern matching, trial and error learning and the naming of nodes.

(3)   Modelling Human Short Term Memory using CODIL

The key step is to define the mechanism by which the brain makes significant decisions, and an assessment of an unconventional computer language called CODIL demonstrates how pattern matching within the network can morph into set processing and from that into a powerful procedural processor.

CODIL arose from a 1967 study of how it might be possible (once computer terminals became available) to provide a human-friendly interface between the salesmen working in a complex and ever changing market place and the invoicing program of a major oil marketing company. This study led to research into how it might be possible to build a “transparent” computer, which could automatically explain what it was doing in user-meaningful terms – in contrast with the conventional black box computer systems we are all familiar with. Significant research using simulator packages was carried out into CODIL, but the project was closed down, restarted and finally closed again in 1988. These closures were partly because it was deemed inappropriate in a department which was concentrating on current commercial computer systems, and partly because insufficient research had been done on the project's theoretical foundations.

A recent re-assessment of the project suggests that CODIL is best described as a language for exchanging information between two neural nets system – one the human user’s brain and the other a computer simulation. The language using structured lists of concepts meaningful to the human, and the interface is via a context window designed to reflect the limited capacity of the human short term memory. The decision making unit processes the information in a way that appears natural to the human user. The approach therefore models some important aspects of human short term memory and also fits in well with infinite recursive network model. 

The paper outlines how the original system worked, and how it can be re-interpreted as a neural net. It describes the various ways in which it was used to process information (including sophisticated problem solving, information retrieval of messy data, processing fuzzy and uncertain information. The  CODIL interpreter could support simple trial and error learning, and while not explored in detail at the time the approach should support at least some aspects of natural language.

(4)   Human Evolution over the last 5 million years

Over most of the last five million years there has been a steady increase in both brain size and tool making, but sometime after the appearance of Homo sapiens, the rate of new tools started to increase at a rapid rate while the brain has, if anything, got slightly smaller. The CODIL model suggests how language started with the ability to give named to concepts and provides an explanation for a number of tipping points in the evolution of language, tool-making  and the associated culture. These tipping points allow both faster learning, and more efficient use of memory, making it possible to handle high level concepts. These factors predict an exponential growth in the amount of culture passed between generations. Sharing information, with humans specializing in different areas, reduces the need for individuals to have bigger brains, while the collective intelligence of the species continues to increase. At the same time the inbuilt "animal brain" limitations relating to the reliability of long term memory, the size of short term memory and the problems of handling negative ideas (leading to problems such as confirmation bias) become apparent as more and more complex ideas are being handled.

(5)   Overall Assessment of the Model

The model suggests an evolutionary pathway to explain the evolution of human intelligence which suggests that at a biological level the human brain works in the same way as in other animals, although it may be bigger and more efficient. The intelligence of animals is limited by the costs of trial and error learning, but language allows significant cultural information to be handle efficiently. Human mental superiority is almost entirely due to the use of language, rather than any changes in the basic brain mechanism, and this has highlighted some of the inherent limitations in a network which evolved to maximise survival of individual animals, where all learning was lost when the animal died.

Because of the complex history of the project there is much more that needs to be done to complete the research, and there are many gaps where I feel fresh minds could take the model further. CODIL started in 1967 as a purely technical computer project and work was aborted because it did not conform to the way the computer industry was going, In retrospect it is clear that if it been in a less technical environment the project would not have been abandoned. While retirement means I have at long last been able to do the relevant “blue sky” research the priority in writing this paper is to document the work that has been done, despite the omissions, while I am still in a position to  help others to restart the project if that is deemed appropriate. 

The Future of the Project

The reassessment of the CODIL project started when my son pointed out that when I pass on the contents of the garage would probably go into a skip - and I decided to reassess the project to see if there was anything worth saving.

It would seem that the project failed because in the 1960s and 70s everyone was so busy climbing onto the stored program computer bandwagon that promoting a possible alternative information processing model was doomed to failure. However some 50 years later it is now obvious that the human brain works in a very different way to a conventional computer and the project could be worth starting up again

 - so what do you think????

I will be completing the draft of the above paper  - and continue to post more detailed topics on the blog. If you have any comments or questions I am always receptive - see Science is about asking the right questions? I am also looking to see if any of the project records should be archived and have prepared summaries The SMBP Story (1965-1967) and The EELM/ICL Story (1967-1990) for the LEO Computer Society Archives at the Cambridge Museum for Computing History, and because the CODIL project was originally supported by the LEO pioneers David Caminer and John Pinkerton, it may be appropriate for some of the later records to go in the same direction.

If anyone is interested in following up the ideas I will be delighted to hear from you, either because you are interested in research into the evolution of human intelligence, or because you are interested in the history of computers and the reasons why this project was originally abandoned.

A Guide to published information on the original CODIL project

A blog which discusses the recent developments in the research