Learning and evolution

When the idea of studying learning began to grow in my head I was sure it would be an easy task, the library had at least 100 aisles dedicated to different strategies, models, explanations, theories about how it should be taught. Education, teaching, or learning problems were kind of synonyms of learning.

After reading all the books, magazines were a good source of inspiration, but only created more notes into pages of my notebooks which ended quickly and I finally decided to look in the brain, it was the logical step, because it’s where learning is scanned, processed and stored. 

I believed it was going to be easy to find answers in this lump that all of us  carry all on our shoulders, and it seemed an easy journey, after all,  how much can fit in an average area of 1130 cm³?.

My trip was from structures with their latin names to the connectome, which is a map of the neural connections, and seeks to describe the brain structure, as well as the genome is more than just a juxtaposition of genes, the set of neural connections is much more than the sum of its individual components (Biswal, Mennes, Zuo, Gohel, Kelly Smith; Beckmann, Adelstein, Buckner, Colcombe et al., 2010).

Connections 

A great expert in this topic, Sebastian Seung (2012) explains  that the connectome contains millions of times more connections than the letters of the genome, and each one will be building specific connections, all those networks is the personal  connectome, which is created on 4 principles: reweighting this means changes in the strength of the synapses; reconnection is the creation and elimination of synapses; rewiring that is the creation and elimination of neuronal branches and regeneration which is creation and elimination of neurons.

If neurons were all, it seemed logic that understanding learning should not take me more than 10 years, but as Dehaene (2011) figured out,  brain represents the reply of the slow evolution of species governed by the principle of the same natural selection that has been perfected over the years allowing the brain to optimize the way in which processes the huge flow of sensory information received to adapt the reactions of the body to a competitive and hostile environment.

If the key of learning is exclusively in the brain, then the idea of learning adapting to the environment was valid and I mistakenly arrived to propose that learning allowed this adaptation, I cannot deny the error because I published in different articles. I changed my mind when a biologist made me remember that other species adapt, including proteins as I explained in another entry on this blog (Dzib Goodin, 2012). Prions took away my sleep for a while.

So it was time to seek the evolutionary mechanisms. Finally we are the sum of the processes of changes and adaptations.

Years of change

Someone kindly suggested me to explore the Baldwin effect, also known as ontogenetic evolution which is a theory of the likely evolution process of learning, which was published for the first time in 1896. The theory proposes a mechanism for learning ability in general, based on the idea of selected descendants of a group can have greater capacity to learn new skills rather than simply the abilities granted by the genetic code which is relatively rigid.

The idea sounds very logical but theory has been controversial from the modern evolutionary synthesis, and it has not been easy to prove the occurrence of the phenomenon. Personally I'm going to start documenting my summer battles between weeds, slugs, and snails to keep my gardens healthy. 

The main limitation of the Baldwin effect from Hinton and Nowlan point of view (1984) is the fact this idea is only effective in spaces that would be difficult to find, very specific species to which it is possible to continue without an adaptive process of restructuring from the space. But for those biologists, who say that wilderness areas are well structured, the Baldwin effect is an important mechanism to allow adaptive processes within the body to greatly improve the space in which a species evolves.

This theory led me to find something else, and I returned to an old topic, Can you believe the evolution of brain processes has been best explained by theorists of artificial intelligence?.

Some researchers has the assumption that brain can adapt and learn from past experience, because the specific evolution is not only inherited behavior but adds inherited goals that are used to guide the learning under the orders of a genetic code that has two components in the species. The first component is a set of initial values to create a network of action which maps the sensory input to behavior, this is presented as a set of innate behaviors that are inherited from the parents (Stefano, Elman, and Parisi, 1994).

The example that comes to my mind is a newborn baby, who is beginning to recognize the environment, his initial reactions are sensory, as Piaget recognized from the last century. Some of these reactions begin to distinguish between the species, for example, reflexes are becoming more sophisticated, and some babies show signs of maturity, while others follow a different pattern of development.

The second component is a network of evaluation,  this focused action on the sensory input to a value scale that help to move from a bad to a good situation by changing its weight in the action of the network during the process and that the individuals maintain as learning goals (Stefano and Parisi 1994).

There is no a bigger pleasure than observing a baby who is in a dilemma. If he manages to control the motor actions and builds the network between look at an object and take it, how does he react when despite the same movements, but the object doesn’t move?. His first reaction is: hey, come to me, I am taking you!. Other babies will try it more than 5 times, while some observe the problem, and maybe one or two will try to solve the problem crying, explaining to the toy that mom will know soon about his rebelliousness and nobody plays with mom.

In this sense, it can be said that the evolution of neural networks contains information not only in genetic terms, but also a collection of behaviors developed by the ancestors and this can be understood as a culture (Dehaene, 2012, Conrad, 2004).

 

It is then that culture has a major role since the adaptations in the environment are not always determined by closed codes and therefore not can become stronger than those established by the selection (including the changes in the social environment). The best example of this is the language, since previously the specie was not dependent on the speech, until it begin to evolve the language skills, is so development processes that had not participated previously in the language can be selected object because of its effects on the acquisition of the language, resulting in the modification of older adaptations (Barret, 2012).

However, culture is not absorbed in the whole brain how explains Stanislas Dehaene (2004) in his theory of neuronal recycling, he says that cultural purchases can take place in a limited surface area bounded by the cerebral cortex. As an example, the author analyzes reading and arithmetic that have greater reproducibility in the neo cortex.

This idea has been explored in more than one research, for example in an article published by Conrad (2004) it presents a theory about the formation of the central nervous system based on the processing of information from the description of the molecular biological systems. He explains that the central nervous system consists of several types of unitary regions, of which there are many interchangeable replications.

Each region contains neurons whose power is determined by the enzymes that recognize the specific input patterns to that region specify by genes inherited or cultivated. Finally, the central nervous system has selection circuitry that put test and evaluation of different regions, determining control of the production of genes on the basis of such an assessment. 

At the same time, there are genes whose production is stimulated in a diffuse way in regions in which occur to transform other relevant regions of the same type. In this sense, the function of the molecules is the same in these new regions because the structure of the tissue and cell properties is the same. 

This makes possible a process of trial and error to learn mediated by the same mechanisms as the natural evolution, except that it is more efficient because the circuits of selection. Systems that operate on the previous basis are capable of performing any executable as a conventional computer, but with significant restrictions on the programming. Thus, these systems are also (structurally) simpler than more susceptible to learning and evolution and conventional information processing devices (Conrad, 2004; Changeux and Dehaene, 2000).

It seems during the evolution of the brain the properties of its tissues development are subject to evolutionary change from the effects on the phenotypes of the brain. This can be initiated by the changes in systems development (for example, through mutation), changes in the environment in which they develop (culture, environment), or both (Barret, 2012). 

An idea of how happens this is provided by Fernando, Szathmary, and Husbands (2012) who claims  that Darwinian evolution can happen in the brain during, for example complex thought, or the development of language in children, though nothing further than the level of the synapse is subject to Darwinian evolution in the brain. Which confirms the affirmation of Seung: we are our connectome.

Evolutionarily, the advantage of having algorithms of replication occurring by natural selectionis not observable to the naked eye, compared to the instrumental learning models (Fernando and Szathmary 2010). In fact, the notion of the dynamics of replication in the brain remains controversial and the creation of neural networks is a cost process that depends on technological development, example of this is the Blue Brain Project.

Artificial neural networks

Thus, the technological advances to make possible the creation of neural networks that tries to simulate the biological capacity to adapt and learn from past experience that has the brain. 

However, even for experts in neural networks, the task of explaining the learning mechanisms have not been easy, because as explain Iriki and Taoka, (2012) brain evolution has three essential components, one developed by multisensory integration (sight, hearing, smelling, feeling) and transformation of coordinates for the control of movements in the living space is an essential function of the nervous system (what is known as ecological niche). But this neural enhancement is not an isolated event, since it allowed the brain to move processing to the summary of information, through the implementation and reuse of the existing principles of information processing space that adapted to the subjection of mental functions and which ultimately led to the development of the language with which it was possible to communicate locations or spaces (which resulted in a cognitive niche). This was also useful handling of the image of the body in space, which became essential for the handling of tools, giving as a result the acceleration of interactive links between cognitive, neural bases cognitive and gave way to the third niche which is building.

Just in case here sounded a simple explanation, complex that a baby may formulate coordinated words, you must know how to use the tongue, move it in a coordinated manner, and learn to control the air, when it manages to dominate the difference between a sound and reaching a word to take the race to perfect this ability. I personally really enjoy these attempts, babies range from simple ma, pa, aba sounds to words: mom, wad, water. Once they have established that, begin to use the tools and parents are scared when they see the child with the Ipad or the cell phone in their possession, nothing more fun to do than using a pencil, or a touch screen. When parents learn to relax, and children send their first text message, or read their first book, tools have sense. Hence to describe the world.

This explains that a modified human environment exerts pressure on future generations to adapt to it, perhaps through the acquisition of new resources that have to adapt to the different organs, with which is possible to explain plasticity induced epigenetically (this term refers to the study of non-genetic factors involved in the development of an organism), including the development of mechanisms of learning involved in such processes. In this way, the additional genomic information can be transmitted between generations through mutual interactions between neuronal, ecological niches and cognitive domains. This scenario locates the brain as part of a comprehensive ecosystem in evolution (Iriki and Taoka, 2012).

It is so arises the neuroevolucion as a field of study, which seeks the creation of artificial neural networks (ANN) through evolutionary algorithms, and sometimes has focused its efforts on static neural networks that cannot change its function during lifetime, since these are the easiest to replicate (Miikkulainen, Feasly, Hohnson, Karpov, Rajagopalan, Rawal and Tansey, 2012; Gauci and Stanley, 2010).

However, a serious problem with the evolution of adaptive systems is that learning to learn is very misleading, as describe Risi, Hughes and Stanley (2010), because a principle is easier to improve physical condition without the ability to learn, evolve by that is not based on the heuristic adaptation. This is learning a stiff task with no greater decision making, which is a model away from reality. 

In their study, the authors find as a conclusion that novelty search has the potential to foster the emergence of adaptive behavior in reward based learning tasks, which opens up a new direction for research in the evolution of neural networks of plastic, which makes it more interesting and easy to learn adaptive behaviors that had been difficult to observe in human models.

But is Henry Markram, who has a good idea about this, he says brain has led billions of years evolve and has many rules, so the challenge of the neuroevolucion is to describe them carefully using mathematical laws and if it is possible to achieve that, the challenge will then be to build a realistic model of the brain (Kushner, 2012).

Brain and school

Although it sounds like a foolish, learning process has a long way, developing its own rules and the school as an institution should not ignore, but does it. The result is not only unhappily educated children, but jobless adults. But, there is no perfect educational system, even the brain has established laws, it modifies them generation after generation, trying to find a biological balance.

Of course, it’s possible to expect a programming expert to design an application that connects through an interface with a single click, so we could learn anything, even those for which we are not physically fit, as in the The Matrix movie, but while that happens, it is worth putting the brain in the classroom and understands its mechanisms. 

It is not my idea that teachers know neuroscience, that is not the goal, but at least I would like to explain to teachers when they  see a child with learning problems they can see him as a problem of education, since  the brain has adapted and survived on the face of the Earth much more better than any curriculum has done so. 

Part of that evolution implies as studies indicate, the brain changes all the time, and in this sense, if a child is not capable of running a task today, far from tag it, should think that under the correct strategies, he or she will do it, with their own pace, accuracy and specificity, different than others, after all, there is nothing more impressive being unique different and special.

The success of the brain is such that it has been able to look to infinity and beyond, the Moon was not its limit, right now is exploring Mars, has done its utmost, has grown, invented, fantasized and made possible what was thought impossible, in a 1130 cm³ average space^., imagine now that they will be able to make many with a common goal: an effective teaching

 

Alma Dzib Goodin

If you would like to know more about my writing, you can visit my web site,
http://www.almadzib.com/

REFERENCES

Ackley D., and Littman, M. (1991) Interactions between learning and evolution. In Artificial life II, SFI studies in the sciences of complexity. Vol X edited by CG Langton, C Taylor, JD Farmer & S, Rasmussen. Addison –Wesley. United States

Barret, HC. (2012) A hierarchical model of the evolution of brain specializations. Proceedings of the National Academy of Science of the United States of America. 19 (Supl 1). 10733- 10740.

Biswal BB, Mennes M, Zuo XN, Gohel S, Kelly C, Smith, SM, Beckmann, CF, Adelstein, JS, Buckner RL, Colcombe S, et al (2010) Toward discovery science of human brain function. Proceedings of the National Academy of Sciences 107 (10) 4734-4740 

Conrad, M. (2004) Evolutionary learning circuits. Journal of theoretical Biology. 46 (1) 167-188.

Dehaene, S. (2004) Evolution of human cortical circuits for Reading and arithmetic: the neuronal recycling hypothesis. In S. Dehaene, J. R. Duhamel, M. Hauser & G. Rizzolatti (Eds.), From monkey brain to human brain (2004). Cambridge, Massachusetts: MIT Press.

Dehaene, S. (2011) The number sense: How the mind creates mathematics. Oxford University Press. USA.

Dzib Goodin, A (2012) Learning how we learn. Available at: http://talkingaboutneurocognitionandlearning.blogspot.com/2012/05/learning-how-we-learn.html

Fernando, C., and Szathmáry, E. (2010) Natural selection in the brain. In B., Glatzeder, V. Goel,  and A. Muller (Eds) Towards a theory of thinking: building blocks for a conceptual framework. Springer. Germany.

Fernando, C., Szathmáry, F., and Husbands, P. (2012) Selectionist and evolutionary approaches to brain function. A critical appraisal. Frontiers in Computational Neuroscience. 6 (Art. 24). Disponible en http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3337445/pdf/fncom-06-00024.pdf.

Gauci, J., and  Stanley, K.O. (2010) Autonomous evolution of topographic regularities in artificial neural networks. Neural Computation 22(7)  1860-1898.

Hinton, GE., and Nowlan, SJ. (1987) How learning can guide evolution. Complex Systems. 1. 495-502.

Iriki, A., and Taoka, M. (2012) Triadic (ecological, neural, cognitive) niche construction: a scenario of human brain evolution extrapolating tool use and language from the control of reaching actions. Philosophical Transactions of the Royal Society Biological Science. 367. 10-23.

James Mark Baldwin. A New Factor in Evolution. American Naturalist 30, (1896): 441-451, 536-553. Disponible en http://www.brocku.ca/MeadProject/Baldwin/Baldwin_1896_h.html

Krushner, D. (2011) The man who builds brains. The Brain, Discovery Magazine. Disponible en red: http://discovermagazine.com/2009/dec/05-discover-interview-the-man-who-builds-brains

Miikkulainen, R., Feasly, E., Hohnson, L. Karpov, I., Rajagopalan, P., Rawal, A., and Tansey, W. (2012) Multiagent learning through neuroevolution. Advances in Computational Intelligence. 7311. 24-46.

Nolfi, S., and Domenico Parisi (1994). Good teaching inputs do not correspond to desired responses in ecological neural networks. Neural Processing Letters 1 no. 2 (11/94) pp. 1-4.

Nolfi, S., Elman, J., and Domenico Parisi (1994). Learning and evolution in neural networks. Adaptive Behavior 2 (1994): 5-28.

Risi, S., Hughes, CE., y O Stanley, K. (2010) Evolving plastic neural networks with novelty search. Adaptative Behavior. 18 (6) 470-491.

 Seung, HS. (2012) Connectome: How the Brain's Wiring Makes Us Who We Are. New York: Houghton Mifflin Harcout.