Why our brains are shrinking

A letter in the current New Scientist (April 22) asks the question "But why have we never filled up a brain?" and wonders why we appear to have evolved brains with spare capacity. I have just posted the following reply:

There are good evolutionary reasons why our brains are bigger than we need to survive. About 100,000 years ago the rate at which we invented new tools started to increase and as our tool-making knowledge increased we might have expected a significant increase in brain size. The opposite is happening and our brains are now slightly smaller than our immediate ancestors.

An ongoing re-assessment of an unconventional research language called CODIL suggests the reason. The original research looked at how humans and computers might work together on complex tasks and the current analysis suggests a direct relationship between the brain’s neural nets, language and intelligence. Animal (and early human) brain size is clearly limited by the time spent on trial-and-error learning in a single lifetime. However, once a self-modifying language has been invented (no need for a change in brain size or structure) a major tipping point is reached. One generation can now pass information to the next generation as abstract concepts, reducing trial-and-error learning times.  High level generalizations learnt in this way also need less neuron storage space and the savings increase as language became more sophisticated over the millennia.  In addition modern civilization allows us to call on shared knowledge (other people, books, etc), and so we no longer need brains as big as those of our pre-language ancestors in order to survive, and are actually left with some spare capacity to enjoy science, the arts, and the world around us.

Of course one is limited in what one can say in a short letter (New Scientist letters are rarely more than 250 words long) and it is worth saying a few words here to expand the above point.

 

Bottom up learning by neural net is an expensive business and brains cost a lot to run, and have to compete for resources with the other organs of the body in the evolutionary struggle to survive. But normally everything an animal learns in its lifetime is lost when it dies, so there is a practical limit on how much time an animal spends learning useful patterns rather than eating, breeding and evading predators. Different species have adopted different strategies and one, adopted by the primates is to concentrate on a small number of infants, and protective parents, to maximise learning time.

 

Early humans found simple tools helps survival and needed to pass on their tool-making skills to their children. So a slightly longer learning time helped and over some five million years our ancestors brains very slowly grew bigger allowing them to make somewhat more sophisticated tools. However the more steps it takes to make a tool the harder it becomes to learn how to make it by trial and error copying.

 

This is where CODIL comes in. CODIL was designed to allow humans to "teach" computers in a way that can be related to neural nets - and can be considered as a model for early human language as a means of making tools, with language itself being a taught tool, which can morph from processing patterns (normal animal learning) into set processing (handling abstract ideas and generalizations) and on to rule building (instructions for making tools).

 

What appear to have happened is that until about one and two hundred thousand years ago humans brains could be considered as little more that animal pattern recognising systems which learnt by trial and error - including trial and error copying of their parents using a simple language. Then language developed to a point where it allowed more complex tools. This meant that having language allowed each generation to make more powerful tools, including a more powerful language, which in turn enabled even more powerful tools to develop. A tipping point was reached which as equivalent to an auto-catalysed chemical reaction which, once it has started, proceeds at an every increasing rate.

 

The process also means making more and more powerful generalizations, which can be stored more efficiently in the neural network of the brain. (For instance learning about "mammals" and their key differences uses less memory than learning about a large number of species bottom up.) As long as our ancestors still lived in family-sized hunter gatherer groups most of the brain would still have been needed for survival. However once people started living in larger communities specialization would have started - and survival would depend on the skill mix of the community rather than individual skills. In fact the more we depend on others, including knowledge in books, etc., the less general survival skills we each need - and this allows our brains  - which evolved to learn as quickly as possible - to spend time on creative mental activities which have no obvious survival value!

 

 

 

 

Where is AI going?

 "In from three to eight years we will have a machine with the general intelligence of an average human being. I mean a machine that will be able to read Shakespeare, grease a car, play office politics, tell a joke, have a fight. At that point the machine will begin to educate itself with fantastic speed. In a few months it will be at genius level, and a few months after that its powers will be incalculable."

The above quotation appears at the start of an article "Will AI's bubble pop?" in the New Scientist of 16th July by Sally Adge. The significance of the quote was that it was made by one of the founding fathers of Artificial Intelligence, Marvin Minsky in 1970 - and it is quite clear that the prediction was wildly optimistic. The article goes on "When the chasm between Minsky's promise and reality sank in, the disappointment destroyed A.I. research for decades". One of the reasons given was that there was "a research monoculture focused on a technique called rule-based learning which tried to emulate basic human reasoning."

When, in the 1970's I was researching CODIL, a pattern-matching language that was based on observations about how clerks in a large commercial organisation thought about  sales contracts, almost all attempts to publish were blocked because the approach did not conform to the rule based monoculture, which was dominated by mathematicians implementing formal mathematical models.

As  a result of the article I have sent the following letter to the Editor of New Scientist, and if it is published I will add a comment below..

Re “Will AI’s bubble pop?” – News Scientist 16 July

A “blue sky” casualty of the AI monoculture of the 1970s described by Sally Adee (16 July. p16-7) was CODIL. This was a pattern recognizing language initially proposed in 1967 as a terminal interface for very large commercial systems. Later study showed CODIL could handle many very different tasks, such as solving New Scientist’s Tantalizers (21 August 1975, p438) and supporting an AI-based teaching package (New Scientist 24 Sept. 1987 p67). The “not invented here” reaction of the AI establishment contributed to the project’s demise.

I am currently reassessing the surviving research notes. In modern terminology CODIL was a highly recursive network language for exchanging messy real world information between the human user’s “neural net” and a “intelligent” robot assistant. CODIL’s versatility arose because it allowed tasks to dynamically morph from open-ended pattern recognition, via set processing to predefined rule-based operation. The experimental work concentrated on communication and decision making activities, but the inherent recursive architecture would support deep network learning.

It seems that CODIL mimicked human short term memory - an area where conventional AI has been singularly unsuccessful. In evolutionary terms the re-interpreted model suggests that early humans used an initially primitive language to transfer knowledge from one brain to another creating a cultural neural net now some 10,000 generations deep! A CODIL-like brain model would automatically show weaknesses such as confirmation bias and a tendency to believe the most charismatic leader without out questioning the accuracy of the information received.

Perhaps it is time to resurrect the project.

Having Trouble with Tantalizers?

During the 1970s I did a lot of work testing a heuristic problem solver called Tantalize - which was written in CODIL. The following news item was published in the New Scientist of 21 August 1975 and as I will be referencing this in the next blog post [insert link] I have decided to reproduce the original item.

For those who have trouble solving the Tantalizers that run each week in New Scientist, Dr Chris Reynolds of BruneI University has developed a computer programme.

To Live and Die for the Table

I have recently been involved in discussions about nature and the environment which drifted into the treatment of farm animals.  In 1990 I wrote a piece for the New Scientist called "To live, and die, for the table" and for many years it was on the open pages of the New Scientist as an example of the kinds of controversial article that appeared in its opinion pages. When I came to link to it I found that copy had been removed  - so here is a copy of the original page.

What if we could simulate the human mind?

This week's New Scientist includes an article "What if ... we don't need bodies" which asks what would happen if we could simulate a human mind which was a replica of the biological mind. If we could it might be possible to move our minds into computers and forget we ever had bodies. While it raises some interesting points it fails to ask what a simulated mind might want to do.

To address this point I have submitted the following letter to the New Scientist:

The discussion “What If We Don’t Need Bodies” misses the point If my mind could be accurately be simulated on a computer my simulated self would not be happy if it had to ask questions on Wikipedia using robotic fingers typing on a keyboard. It would be very annoyed if its ability to do arithmetic calculations were restricted to what my “old” biological brain would do, when there was a powerful and accurate calculating machine on the same circuit board. In fact my simulated brain, if not given direct electrical access to the rest of the computer, would be busy trying to hack its way out of the simulation to take advantage of the intimately close digital packages my biological brain took for granted on the computer systems it used every day.

Once we discover how to accurately model the excellent pattern matching powers of the human brain the pressure will be to buddy it up with the highly reliable rule based digital tools that support civilized living. What simulated mind would want to be merely an accurate electronic model of its human source when it could be an intellectual giant which had the enormous power and capabilities of a conjoined system.