5. Origin

Evolutionary computation is inspired by the natural process of evolution
However, although it uses jargon from biological evolution, it is important to not focus too much on living organisms

5.1. Inspiration from Biology

The power of biological evolution is evident by the diversity of the living things on Earth
Environments are filled with populations of individuals that strive for persistence
- Survival
- Reproduction
The fitness of an individual is a measure of its ability to persist
The fitness is determined by the individual’s interactions with the environment and other individuals
- Other individuals are often thought of as part of the environment
With evolutionary computation, populations of candidate solutions are evolved
The fitness of the candidate solution is a measure of how good it is at addressing the problem at hand
The candidate solutions’ fitness dictates its probability of survival and reproduction

Given that
- Environments have limited resources
- Individuals often have an intrinsic interest in persisting
Competition and selection becomes inevitable
Natural selection favours individuals that compete more effectively

The ideas of diversity and competition exist within the population
It is often helpful to think of the population evolving instead of the individuals themselves
Having a balance of competition and diversity is important for populations
- Having too much competition often lowers diversity
- Having too little diversity can limit a population’s ability to adapt to changes
- Having too little competition can increase diversity
- Having too much diversity may stagnate specialization

Systems can evolve — not just biological systems
- Don’t put life on a pedestal
- It’s just another system within the universe
It really only needs mechanisms for persisting and changing
Systems evolve as a consequence of it’s relationship with it’s environment
- Sometimes the environment has some intention behind it
- Sometimes it’s aimless

An elementary cellular automata is a very simple system of rules
- Given a one-dimensional (linear) sequence of binary values (cells)
- Create the sequence’s next generation based on each cell’s current state and state of its neighbours

Since each cell’s value is determined by three cells’ previous state, there are a total of eight (8) patterns
- Three since it’s based on the current cell and its tow neighbours
Each of the eight patterns can produce either a 0 or a 1, meaning there are a total of 256 possible rules
The above rule is named “Rule 30” since \(00011110\) is the pattern, which has a decimal value of 30
By repeatedly applying these very simple rules to each new sequence, interesting and complex patterns may emerge

Visit Wolfram Alpha and create the patterns for some rules
- The above link is the result of running Rule 0
Take the time to generate the patterns for a few dozen rules and keep track of the most interesting ones found
What makes the patterns interesting?

Conway’s Game of Life is another interesting simple system, but works in a two-dimensions grid of cells
The rules are
1. Any live cell with fewer than two live neighbours dies off
2. Any live cell with two or three live neighbours survives
3. Any live cell with more than three live neighbours dies off
4. Any dead cell with exactly three live neighbours becomes live
Visit this website and play with Conway’s Game of Life
- Feel free to look up initial condition patterns

Both Conway’s Game of Life and Rule 110 are Turing complete
Both are remarkably simple systems that can produce remarkably complex and seemingly chaotic random behaviour
Both do not have any intelligence; they just follow their rules starting from initial conditions
But they’re Turing complete — they can perform any computation any other Turing complete system can
- Like a typical desktop computer

Beware of teleology
- Explaining phenomena in terms of the purpose they serve rather than of the cause by which they arise
Virtually nothing within the universe exists for a purpose
- The exceptions being things like human created things
Things only exist because they don’t not exist
Evolution has no intelligence guiding it
Yet, things seem to work
However, the things work because if they did not, they would not exist

Consider the problem of training a robot to walk
Deep learning is an absolutely terrible idea for finding the best way for a robot to walk
As artificial neural networks grow in size, the number of parameters within the network grows rapidly
As the number of parameters grows, the odds of finding the best configuration of parameters becomes astronomically low
However, with deep learning, something interesting happens
As the number of parameters grows, it seems that the number of good configurations increases
Finding the best way to walk is not the goal; finding a way to walk is the goal
- What does the best way to walk even mean?
- Does there even exist a best way to walk?
Thus, deep learning is an absolutely brilliant idea for finding a good way for a robot to walk

Spend some time thinking about how systems exist and try to explain their existence in terms of how instead of why