A Possible Evolutionary Neural Net Model of Turing's Child Brain

Picture source

A Possible Evolutionary Neural Net Model of Turing's Simple Child Brain

In 1950 Turing wrote a paper Compuing Machinery and Intelligence and suggested that one approach to building an intelligent system, that could play the imitation game, might be to start by simulating a child's brain.

“Presumably the child-brain is something like a note-book as one buys it from the stationers. Rather little mechanism, and lots of blank sheets. …  Our hope is that there is so little mechanism in the child-brain that something like it can be easily programmed.”

 While an enormous amount of research had been done relating to the brain and artificial intelligence since you will find nothing in the published literatureto suggest that Turing's hope was justified.

This paper looks at the archives of an early and very unconventional computer language, CODIL, which (like dozens of other experimental languages at the time) never became commercially viable. CODIL was designed to help a human user and a computer clerk work together symbiotically on complex problems where uncertainties made the conventional pre-define algorithm approach impractical. 

But the underlying mechanism supporting CODIL is very simple (like Turing's child brain), and could be a model of how information has passed from parents to their children for thousands of generation - making a very deep learning network. 

The assessment shows that CODIL could be considered to be a language for exchanging information between two neural networks - one the brain of the human user and the other in the computer.  This was achieved by handling information in a way that resembles human short term memory. In addition, because the human is providing the information in "network form" the computer does not need to use time-consuming trial and error learning, avoiding the learning problems of much A.I> research.

Because the work was abandoned in 1988 many of the crucial questions, such as "Could a CODIL-like model supporrt natural language?" were not examind before the project closed. As a result this paper looks ways in which the CODIL research relates to evolution of human intelligence.

Read the paper here

I would be very grateful for any comments you can make, with citations to any similar reseach I may have missed,  In particular I would be interested to know if you think the research should be restarted.

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:

Natural Language: The Evolution of Grammar

A reference in the blog Babel's Dawn alerted me to an interesting paper "An anthropic principle in lieu of a "Universal Grammar" by Hubert Haider, Department of Linguistics & Centre of Neuroscience at the the University of Salzberg.

Haidar starts by asking three questions:

Among the many unanswered questions in grammar theory, the following figure prominently.

1. First, what is it that enables children to successfully cope with the structural complexities of their mother  tongue while  professional grammarians  tend to  fail when  modelling them? 
2. Second, what determines the narrow system corridor for human grammars? 
3. Third, are the grammars of human languages the offspring of a single proto-grammar instantiating a "Universal Grammar" (monogenic) or are the shared traits of human grammars the result of convergent changes in the grammars of human languages of diverse ancestry (polygenic)? 

 The last of these questions is mainly concerned with debunking the ideas of Chomsky and I will pass over it in this blog post. More importantly he argues strongly that natural language and the grammars associated with them are due to normal evolutionary forces and are limited by the resources present in the human brain.  He does not detail the nature of the resources provided by  the brain - which is very relevant to my research into the evolution of human intelligence.  For this reason I quote extracts from his text below, and then give a brief example of how my model demonstrates the way in which the neural network in the brain supports the resources his approach needs to support language . 

A Simple Example of how CODIL works

J M asked "How would these symbiotic - easy to understand computers - look like? how would they work?"

The computer hardware would look like a conventional computer, and in fact any commercial system would probably combine both technologies. The difference is in the way you would interact with it.

The original idea (see The SMBP Story) related to a very large commercial sales contract program in 1967.  The workings can best be explained in terms of a very simple network model.

Imagine a board with a number of light bulbs on it - each light bulb represents a commercial concept such as CUSTOMER or PETROL - and some bulbs could also have numbers  associated with them. If a customer contract referred to a discount it might be written:

CUSTOMER = SMITH; PRODUCT = PETROL; 

QUANTITY >= 1000; DISCOUNT = 10%

This would indicate that the bulbs for CUSTOMER and SMITH were linked, as where the bulbs for PRODUCT and PETROL, etc. In addition there would be links between the linked pairs.

Similarly there would be a pricing table with statements such as:

PRODUCT = PETROL; QUANTITY >= 500; 

UNIT PRICE =  15

with the appropriate bulbs linked.

To Price a delivery the nodes equivalent to the following would be switched on:

CUSTOMER = SMITH; PRODUCT = PETROL; 

QUANTITY >= 1000

and the this would automatically turn on the lights for 

DISCOUNT = 10%  and UNIT PRice = 15

The important thing is that everything is done in terms of concepts the user can understand, and the system deduced what needs doing. The approach is application independent and can easily show the user what it is doing and why. Effectively it is a transparent system, where conventional computers are black boxes. 

In practice the experimental interpreters hold all the links in a table and for each input they automatically look up the relevant links. Instead of a very large pricing program which will price any sale, you have a table-lookup which - in effect - works out a purpose-build mini-program for each transaction. 

Papers on the Original CODIL System

A detailed paper is being drafted for this blog and will appear shortly

Why does CODIL differ from other computer languages

This query came up on a FutureLearn Course which read "Christopher, to be honest I don't think the world needs any more computer languages or most of the ones it already has for that matter"

The vast majority of computer languages, such as COBOL, Fortran, C, Java, etc. are designed to process information on a conventional stored program computer where the memory consists of numbered boxes which contain numbers . The numbers in the boxes may represent coded data (often several different formats), numeric addresses, or coded instructions. This approach was originally designed to handle a class of well-defined mathematical tasks which humans find difficult to do quickly and accuracy, so it is not surprising that modern computers are incomprehensible black boxes when viewed by the average human being. They were deliberately designed to do efficiently things which people are bad at doing.

CODIL uses a completely different memory structure which is based on a dynamic network which attempts to mimic the working of the brain's neural network. The aim is to produce a transparent information processor (rather than a black box) which is easy for the average human to understand and use for a range of potentially complex information processing tasks. It is particularly aimed at complex tasks where a dynamically flexible human interface is advantageous - and so fills the area where conventional computers are weakest.

In CODIL the array of numbered boxes which make up a conventional computer memory is replaced by a large number of nodes, where each node consists of an on/off switch and a label (which is for the benefit of the human user). The human user defines the names of the nodes and the wires linking the nodes.

How plans for a user-friendly computer were rubbished 50 years ago

In 1968 David Caminer and John Pinkerton (who were responsible for the world's first business computer, the LEO I, and who were directors of English Electric Computer) decided to fund research into a project to build inherently user-friendly computers and it was estimated that the market for such systems would be several hundred million pounds a year.

However, as a result of the government inspired merger to make the UK computer industry more competitive, ICL was created, and the project was closed down with no serious attempt to assess the successful research into CODIL which had already been carried out.

This was, of course, about 10 years before the first personal computers, and it is interesting to speculate what might have happened if the research had not been so rudely interrupted. Perhap the UK based idea would have been successful and the first personal computers would have been inherently friendly. This would have meant that there would have been no need for the hard to use MS-DOS operating system - and no one would have heard of Microsoft.

To get details of what happened read 

The EEC ICL Story

To see how the idea originated read

The Unlikely Origins of my current Evolutionary Research

This is relevant to this blog because the rubbished 1968 proposal is orign of the model of human working memory at the heart to the current research into human short term memory.