We want you to read chapter 1 of Mitchell (1998), which is an introduction to genetic algorithms. As discussed in the Mitchell reading, genetic algorithms (GAs) have various applications in artificial intelligence and computer science, but we will be using them as a way of simulating the biological evolution of signalling systems, in particular evolution of signalling under natural selection. On this page I provide a very simplified summary of what natural selection is – there are also literally dozens of textbooks or web tutorials you could consult. This should form a useful background to understanding the Mitchell chapter. FYI, I am working with a textbook called Evolution (they are all called Evolution), by Barton et al – see below for reference.
Don’t forget the short quiz (link at the bottom) – your answers to this are very helpful when it comes to preparing the lecture.
Adaptation
Everywhere you look in the natural world, you see animals which seem to be exquisitely well-designed for the world they live in: think of the superb camouflage systems of animals like stick-insects or chameleons, or the beautiful fit between flowers and their pollinators, of the ability of animals like camels to or penguins to survive in extreme heat or extreme cold, or the the complex social systems of ants and termites. These cases of apparent design are known as adaptations or adaptive traits (and the process by which adaptations are formed is also known as adaptation).
How can we explain adaptations, the appearance of design in life? Pre-Darwininan explanations were either theological, or appealed to the idea that organisms modified themselves to fit their environment and somehow passed those changes on to their offspring. These early theories have been replaced by one of the most important scientific ideas in human history: evolution by natural selection. All adaptations are the product of natural selection, and only natural selection leads to adaptation.
Evolution by natural selection
Natural selection is the process by which genotypes with higher fitness increase in frequency in a population, and it arises whenever you have heritable variation in fitness (Barton, 2007). I’ll explain what that means, and why it leads to adaptation.
The genotype of an individual is that individual’s genetic make-up – the genetic information that determines, through complex interactions with the environment, an individual’s phenotype, their body and behaviours and other observable characteristics. Genetics and the processes by which genes influence bodies and behaviours are incredibly complex, and we don’t need to know about them for the purposes of this course, or indeed for understanding how natural selection works (although if you’re interested, again there are numerous excellent textbooks and web tutorials) - Darwin didn’t know that genes existed when he formulated his theory of natural selection. The crucial thing is that genes are passed from parent(s) to offspring, via asexual or sexual reproduction. If an aspect of an organism’s phenotype (say something about its physical appearance, or the behaviours it performs) is determined or influenced by that organism’s genes, then that means that that trait is heritable: it can be passed from parent to child. More informally, for heritable traits the offspring will resemble the parent. For instance, you probably physically resemble your parents – that’s because many aspects of your physical appearance are strongly influenced by your genes, and therefore are heritable. Heritability is one of the requirements for natural selection: traits that can’t be passed from parent to offspring can’t evolve by natural selection.
The second precondition for natural selection is that the population exhibits variation. This is obviously true at the level of phenotypes – you’ve probably noticed that people vary in their physical appearance, tastes, abilities and personality, and you can similarly observe variation within any species of animals, insects, or plants (although you might need to be relatively expert to spot it). And at least some of this variation is underpinned by variation at the genetic level: different individuals have different genotypes, and variation in their genes contributes to variation in their bodies and behaviours. This is therefore heritable variation, and heritable variation is required for natural selection to operate: if everybody ‘looked the same’ (i.e. had the same phenotype), or if the phenotypic variation that existed was not heritable, natural selection would not occur.
The final requirement for natural selection is that variation matters. In particular, it must matter for fitness, which is a technical term referring to the number of offspring an individual has: in order for natural selection to occur, variants must differ in fitness. While you may not have noticed any particularly systematic correspondences between phenotypic variants and fitness in your personal experience, these have been painstakingly explored and mapped out by biologists using observational and experimental techniques (these examples from Barton et al., chapter 19). For instance, biologists might carefully note the size of birds and their probability of survival (it turns out that, at least for some bird species, being too big or too little makes you more likely to die young, and survival is of course a prerequisite for having offspring; somewhat surprisingly, the same link appears to apply in human infants). Or they might observe a link between the length of a bird’s leg and the number of offspring it has (it turns out that, for female song sparrows, a longer tarsus means more offspring – I don’t know why). It’s not hard to imagine that other heritable traits (i.e physical appearance, physical characteristics, abilities or behaviours) also impact on fitness – therefore, if there is variation in those traits, we will have heritable variation in fitness: individuals inherit their traits from their parents, different individuals inherit different traits, and some of those traits are ‘better’ (in the sense of leading to more offspring) than others.
If all of these three components are in place, then natural selection ensues. Organisms with high-fitness traits will produce more copies of themselves than organisms with traits with lower fitness (by definition), and over time the population will come to be dominated by traits which are associated with high fitness. Or, as the definition at the start of this section said, “Natural selection is the process by which genotypes with higher fitness increase in frequency in a population”, and this occurs whenever we have heritable variation in fitness.
If you don’t like my explanation of natural selection, find another one! I quite like this very simple explanation with nice pictures, or you could try this explanation from Ridley (1996), p71-72:
“Natural selection is easiest to understand, in the abstract, as a logical argument, leading from premises to conclusion…
- Reproduction. Entities must reproduce to form a new generation.
- Heredity. Offspring must tend to resemble their parents: roughly speaking, “like must produce like.”
- Variation in individual characters among members of a population. …
- Variation in the fitness of organisms according to the state they have for a heritable character. … individuals in the populatioon with some characters must be more likely to reproduce (i.e. have higher fitness) than others.
If these conditions are met, for any property or species, natural selection automatically results. If any conditions are not met, natural selection does not result. … When all four conditions apply, the entities with the property conferring higher fitness will leave more offspring, and the frequency of that type of entity will increase in the population” .
Why does natural selection lead to adaptations? My informal definition of an adaptation was that it is a feature of an organism that appears to be well-designed for that organism’s environment. A better, more technical definition would be that an adaptation is a trait which functions to increase fitness, and that evolved for that function. Hopefully the link should now make sense: adaptations are traits that enable organisms to better survive and reproduce (both of which influence that organism’s fitness), and natural selection leads to the accumulation of those sorts of traits in organisms. Since natural selection has been operating for as long as we have had heritable variation in fitness on the planet (i.e. probably as long as life has existed), we are surrounded by organisms which are the product of many many many millions of years of natural selection – they have been exquisitely finely-tuned by natural selection.
Mitchell Chapter 1
Melanie Mitchell is a computer scientist interested in, among other things, evolutionary computation and artificial life. As we just saw, natural selection produces beautifully well-designed organisms: evolutionary computation is based on the idea that we could use similar principles to solve complex programming or engineering challenges, by evolving solutions on computers. Artificial Life is the slightly less hard-nosed cousin of evolutionary computation, and seeks to understand life on earth by creating simulated ecological systems, simulated evolution, and playing with those models.
A key tool for evolutionary computation and artificial life is the genetic algorithm, and that’s what this reading introduces. A genetic algorithm is an algorithm (an explicit, step-by-step series of operations, e.g. that could be followed by a computer) which simulates evolution by natural selection. As discussed in the Mitchell chapter, genetic algorithms include models of all the things I discussed above: genotypes, phenotypes, heredity, variation, fitness.
Two quick warnings:
- Some of the terminology used in the GA literature is slightly different from the standard terminology in biology. In particular, the use of “fitness” is subtly different. In biology, this means the number of offspring you have. In Mitchell’s chapter, she uses “fitness” to refer to some score or measure which is calculated for each simulated organism, which then determines how likely they are to have offspring – so fitness in Mitchell’s sense determines fitness in the standard biological sense, but they are not quite the same thing.
- Mitchell talks quite a lot about bitstrings as a representation of genotypes, but you have to wait for a while before you see what a bitstring is. It’s a string of bits, i.e. a string of 1s and 0s, like this: 10100110010.
Post-reading Questions
Once you have read this page and done the Mitchell reading, take this short quiz, which tests you on a few basics (so I can see if you understood the readings!) and gives you a chance to flag up things you want me to go over again in class (or it does if you do it the day before class!). Then you can compare what you said to my comments on how I would have answered.
References
Barton, N. H., et al. (2007). Evolution. Cold Spring Harbor, NY: Cold Spring Harbor Laboratory Press.
Mitchell, M. (1998). An Introduction to Genetic Algorithms. Cambridge, MA: Harvard University Press.
Ridley, M. (1996). Evolution. London: Blackwell.