Sunday, 27 August 2017

Evolutionary Steps towards Human Intelligence

Humans are vain creatures and we like to think we are very clever, and concentrate on what we can do. However to understand how our intelligence evolved we actual need to look at our in-built mental weaknesses.

The reason for this is that the Blind Watchmaker of evolution does not plan ahead and often the results seem to be far from optimum. For instance the nerves in the human eye are at the front of the retina and mean that there is a blind spot in our vision. Our vagus nerve takes a roundabout route rather that taking the shortest path – and this becomes ridiculously long in an animal such as the giraffe.  
Similar defects apply to our minds. The short term memory is surprisingly small while our long term memory is unreliable. We think more slowly when processing negative ideas, and suffer from confirmation bias. Unless taught our logical skills have limitations, common sense is not always the most appropriate answer, and we are bad at handling numbers and even worse with more abstract mathematical concepts. We are also too keen to follow charismatic leaders without stopping to check whether their rhetoric makes sense.

The important thing to realize is that these limitations are caused because we are using our “animal brains” in novel ways and defects which were minor at the animal level have started to become significant.

The following draft notes suggest the main steps involved in the evolution of human intelligence starting with the simplest possible animal brain, and how what happened millions of years ago has put restraints on what our brains can do.

So let us set the Blind Watchmaker of evolution to work.

·        The Blind Watchmaker starts work. This is conventional genetic evolution at work and all the brain’s nodes and the links between them are determined genetically. This is the mechanism behind inherited instincts. There is random variation, and the more successful networks (as judged by the survival of the host animal) are passed to the next generation. This is an extremely slow process.

o   One would not expect such a hit and miss random process to generate a sophisticated mathematical model of how information is processed, What one should expect is a quick and inexpensive one which improves the chances of survival. This suggests that the lowest levels of the brain will be mathematically unsophisticated – as of course the Blind Watchmaker cannot plan ahead. Some of the basic weaknesses in untutored human logic, for instance relating to the processing of negative ideas, may well be a relic of this phase in the evolutionary pathway.

o   Genetic change is always comparatively slow, and sudden big changes are rare. The DNA code has been stable for over a billion years, and the basic structure plan for mammals is somewhere between 70 and 100 million years old. Because any major changes in the brain could have wide impacts on the rest of the animal one can rule out any major sudden change, and there is no evidence for such a sudden drastic change in our early hominin ancestors.

·        The Blind Watchmaker’s apprentices. Individual animals now have brains which, in addition to the instinctive nodes, include nodes and links which can be modified by trial and error learning. This learning is subjected to survival of the fittest tests – so brains with the best learning networks, and the best sensory organs, are favoured. Of course the individual apprentices all die and the information they have learnt is lost. So while the individual short-lived brains have to learn very quickly the process of evolving better genetic learning mechanisms is still comparatively slow.

o   If learnt information is prioritized it is possible to overcome the problem of handling negatives as long as the information being processed is unsophisticated. Scaled up to the human level this solution (perfectly acceptable at the animal level) could be why we suffer from confirmation bias.

o   In the survival of the fittest game if you don’t learn fast you are likely to be dead, and once danger has threatened it is better to play safe and totally avoid the situation in future. This means that there almost certainly must be two levels of learning – a quick “it could be dangerous” mechanism and a slower and  more easily modified “trial and error” process involving repetition. The quick “it could be dangerous” mechanism may later have been used to allow humans to “speed-learn.” When scaled up to the human level using the "it could be dangerous" route introduces both big advantages and serious traps in the way our mind works.

o   If you look at evolutionary economics of learning, an incremental model where links between nodes are established and then tweaked to get the best priorities rules out any A.I. big data type models or those involving sophisticated statistics. The result is of this simpler approach is that human long term memory is inherently unreliable.

·        The Watchmaker develops eyes. For animals which rear their offspring, limited information can be passed from generation to generation by the juveniles observing and trial and error copying the adults. This process becomes even more important in social animals, and means that at the instinct level the ability to copy other animals becomes important. 

o   This factor makes it easier for one species to diversify into two or more different species – and in human terms could help to explain why we seem to have exchanged genes with several different hominin species – in particular Neanderthals and Denisovans.

o   As brains gets bigger they become expensive to build and maintain and so evolutionary economics becomes important. No species will evolve a brain bigger than it can use in a lifetime, and one aspect of trial and error learning in a neural net is that learning times can become significant when the information being learnt becomes more complex. This fact would appear to put a cap on animal intelligence – but humans appear to be an exception.

·        The Watchmaker becomes a teacher. Tool making appears to have been critical in human evolutionary terms. Inventing a sophisticated tool means nothing unless the means of making it are passed on reliably from generation to generation. The key would appear to be the development of language as a means of transferring knowledge between generations.

o   Fossil evidence suggests that early human evolution continued with the brain slowly getting bigger – allowing more tool-making knowledge to be trial and error copied. There was a sudden change about 150,000 years ago when we started making many more varied tools and – counter-intuitively – our brain stopped getting bigger. This was probably the point when language became better are passing on tool-making skills. (The failure of the brain to continue to expand when handling "more" information can be explained if taught information uses less brain than information learnt by trial and error.)

o   It may seem counter-intuitive but in evolutionary terms the most effective way forward is to ditch the built-in safeguards of trial and error learning and instead blindly follow the rules the toolmaker teaches. What may have happened is that the fast danger-related learning mechanism has been modified to speed learning – bypassing the much slower but more accurate trial and error learning method. This is probably why humans tend to follow the most charismatic teacher without checking whether the information they are getting is correct. Apparently very intelligent people can be fooled by the fraudulent salesman or the convincing politician.

·        The Watchmaking Team become designers. Given an evolving language there is a significant change in the information flow without the need for any significant change in the underlying brain structure. Individuals take on specialist roles, ways of thinking such as mathematics and science are passed on, and information processing tools such as writing, photography and computers are developed. Our species is now using a brain that evolved for low-level pattern matching to process complex and often abstract rule-based information. The combination has some interesting effects. Our learnt information and shared knowledge includes rules about how to invent – so the human race are now prolific inventors although the underlying biological mechanism has evolved to be an excellent and unimaginative copycat. However the same underlying mechanisms mean that humans suffer from confirmation bias, can have problems with negative ideas, and in the absence of contradictory information tend to follow the most charismatic leader without considering the consequences.
o   Effectively we have started to evolve as a colony where different individuals have different roles – and no-one needs to know everything. In addition as learning involved more abstract ideas it needs even less memory – and this may explain why the human brain is slowly becoming smaller – because in no longer needs to be as big because we do far less trial and error learning.

o   Cultural evolution, and the expansion of knowledge that goes with it, is a very rapid process measured in years rather than thousands or even millions of years. There is no way that biological evolution at the genetic level can possibly keep up with it. So we are all still motivated by our underlying animal instincts and, unlike ants, have not learnt that the future of the colony requires everyone to work together with a common aim.
OK some tentative ideas – but what they suggest is that our brain is no more than a supercharged animal brain – and while this allows us to share information to great effect, many of the underlying limitations of the animal brain are becoming very apparent.
Pleaswe note that the above are discussion notes to help me draft a more detailed paper - any questions or comments wouls be appreciated.

No comments:

Post a Comment