Mutation, Variation, and Novelty • artificialLifeR

library(artificialLifeR)

Purpose

This article explains mutation, variation, and novelty in artificial-life models. Variation is essential because selection requires differences among individuals (Darwin 1859; Nowak 2006).

The purpose of this chapter is to show how novelty can enter an artificial population and why variation matters for adaptation-like dynamics.

The guiding question is:

How does novelty enter an artificial population?

Why variation matters

A population with no variation has limited capacity for adaptive change. If all agents are identical, selection has little to distinguish. Every agent has the same traits, the same potential advantages, and the same vulnerabilities.

Variation changes this. Once agents differ, some traits may become more successful than others under particular environmental conditions.

For example:

faster agents may reach resources more easily;
more efficient agents may convert resources into energy more effectively;
agents with lower reproduction thresholds may reproduce earlier;
agents with higher energy may survive longer.

In artificial-life models, variation provides the raw material for selection-like processes.

Variation versus mutation

Variation and mutation are related, but they are not the same.

Variation refers to differences among agents in a population.

Mutation refers to one mechanism that can create or modify those differences.

A population can have variation because agents were initialized differently. It can also gain variation over time because offspring differ from parents.

Concept	Meaning
Variation	Differences among agents
Mutation	Random change to a trait
Novelty	New trait values or new combinations
Selection	Differential persistence or reproduction
Adaptation-like change	Shift in population traits over time

In artificialLifeR, variation can appear through initial trait differences and through mutation.

Mutation as controlled noise

In biological systems, mutation refers to changes in genetic material. In artificialLifeR, mutation is much simpler. It is represented as random numeric change applied to a selected trait.

This is not a full genetic mutation model. It does not represent DNA, RNA, molecular replication, recombination, repair, or developmental biology.

Instead, mutation is used as a teaching abstraction:

Mutation introduces small changes that can alter trait distributions.

This makes it possible to explore how novelty enters a population without modelling molecular genetics.

Relation to the package

The function simulate_mutation() applies mutation-like changes to a selected trait.

Argument	Conceptual meaning
`agents`	Starting population
`trait`	Trait to modify
`mutation_rate`	Probability that an agent’s trait changes
`mutation_sd`	Size of random mutation effect
`lower`	Lower bound for trait values
`upper`	Upper bound for trait values
`seed`	Reproducibility

The mutation rate controls how often mutation occurs.
The mutation size controls how large the changes are.
The bounds prevent trait values from moving outside the allowed range.

Example: create a population

Start by creating a simple population.

agents <- create_agents(
  n_agents = 50,
  seed = 5
)

head(agents)
#>   agent         x         y    energy      speed efficiency
#> 1     1 0.2002145 0.3845769 1.2194873 0.01009226  0.5946089
#> 2     2 0.6852186 0.5662727 1.0281589 0.07270623  0.5751771
#> 3     3 0.9168758 0.9219906 1.1533034 0.06351589  0.4482623
#> 4     4 0.2843995 0.9758776 0.9112248 0.05416967  0.5808336
#> 5     5 0.1046501 0.9330338 0.9831699 0.04884309  0.4385465
#> 6     6 0.7010575 0.3811627 0.8612570 0.06787623  0.6238259
#>   reproduction_threshold age alive
#> 1               1.488632   0  TRUE
#> 2               1.470490   0  TRUE
#> 3               1.598917   0  TRUE
#> 4               1.422487   0  TRUE
#> 5               1.527590   0  TRUE
#> 6               1.541078   0  TRUE

Measure the starting variation

Before mutation, summarize variation in the selected trait.

measure_life_like_complexity(
  agents,
  trait_col = "efficiency"
)
#>    n unique_values  entropy      mean        sd temporal_variability
#> 1 50            50 2.889669 0.5066717 0.0989657                   NA
#>   mean_abs_change
#> 1              NA

Apply mutation

Now apply mutation to the efficiency trait.

mutated <- simulate_mutation(
  agents,
  trait = "efficiency",
  mutation_rate = 0.50,
  mutation_sd = 0.08,
  lower = 0.01,
  upper = 1,
  seed = 5
)

head(mutated)
#>   agent         x         y    energy      speed efficiency
#> 1     1 0.2002145 0.3845769 1.2194873 0.01009226  0.5711303
#> 2     2 0.6852186 0.5662727 1.0281589 0.07270623  0.5751771
#> 3     3 0.9168758 0.9219906 1.1533034 0.06351589  0.4482623
#> 4     4 0.2843995 0.9758776 0.9112248 0.05416967  0.6943207
#> 5     5 0.1046501 0.9330338 0.9831699 0.04884309  0.5584484
#> 6     6 0.7010575 0.3811627 0.8612570 0.06787623  0.6238259
#>   reproduction_threshold age alive
#> 1               1.488632   0  TRUE
#> 2               1.470490   0  TRUE
#> 3               1.598917   0  TRUE
#> 4               1.422487   0  TRUE
#> 5               1.527590   0  TRUE
#> 6               1.541078   0  TRUE

Compare before and after mutation

rbind(
  before = measure_life_like_complexity(
    agents,
    trait_col = "efficiency"
  ),
  after = measure_life_like_complexity(
    mutated,
    trait_col = "efficiency"
  )
)
#>         n unique_values  entropy      mean        sd temporal_variability
#> before 50            50 2.889669 0.5066717 0.0989657                   NA
#> after  50            50 3.015854 0.5154477 0.1213256                   NA
#>        mean_abs_change
#> before              NA
#> after               NA

Interpretation

Mutation changes the distribution of efficiency values. It can increase, decrease, or reshape variation depending on mutation rate, mutation size, and trait bounds.

A careful interpretation is:

The simulation shows how random trait changes can alter variation in an artificial population.

An overstatement would be:

The simulation fully models biological mutation.

The first statement is appropriate. The second is not.

Visualizing trait change

If the plotting function supports point or histogram-style outputs, you can inspect trait values directly. A simple base R plot is also useful for this chapter.

plot(
  agents$efficiency,
  mutated$efficiency,
  xlab = "Efficiency before mutation",
  ylab = "Efficiency after mutation",
  main = "Trait values before and after mutation"
)

abline(0, 1, lty = 2)

Interpretation of the plot

Each point represents one agent. Points close to the dashed line changed little. Points farther from the line changed more.

This plot helps show that mutation is not a population-level command to become better. It is a source of variation. Some changes may be useful, some may be harmful, and many may be neutral depending on the environment.

Mutation rate

The mutation_rate controls how many agents are affected by mutation.

A low mutation rate means few agents change.
A high mutation rate means many agents change.

low_mutation <- simulate_mutation(
  agents,
  trait = "efficiency",
  mutation_rate = 0.05,
  mutation_sd = 0.08,
  lower = 0.01,
  upper = 1,
  seed = 5
)

high_mutation <- simulate_mutation(
  agents,
  trait = "efficiency",
  mutation_rate = 0.80,
  mutation_sd = 0.08,
  lower = 0.01,
  upper = 1,
  seed = 5
)

rbind(
  original = measure_life_like_complexity(
    agents,
    trait_col = "efficiency"
  ),
  low_mutation = measure_life_like_complexity(
    low_mutation,
    trait_col = "efficiency"
  ),
  high_mutation = measure_life_like_complexity(
    high_mutation,
    trait_col = "efficiency"
  )
)
#>                n unique_values  entropy      mean         sd
#> original      50            50 2.889669 0.5066717 0.09896570
#> low_mutation  50            50 2.856233 0.5084719 0.09675427
#> high_mutation 50            50 3.082754 0.5154907 0.12512048
#>               temporal_variability mean_abs_change
#> original                        NA              NA
#> low_mutation                    NA              NA
#> high_mutation                   NA              NA

Interpretation of mutation rate

Changing the mutation rate changes how often novelty enters the population.

A low mutation rate may preserve continuity but introduce little novelty. A high mutation rate may introduce more novelty but may also disrupt inherited structure.

This is important because artificial-life systems often depend on a balance between continuity and change.

Mutation size

The mutation_sd controls the size of the mutation effect.

Small mutation sizes create small changes.
Large mutation sizes create larger jumps in trait values.

small_mutation <- simulate_mutation(
  agents,
  trait = "efficiency",
  mutation_rate = 0.50,
  mutation_sd = 0.02,
  lower = 0.01,
  upper = 1,
  seed = 5
)

large_mutation <- simulate_mutation(
  agents,
  trait = "efficiency",
  mutation_rate = 0.50,
  mutation_sd = 0.20,
  lower = 0.01,
  upper = 1,
  seed = 5
)

rbind(
  original = measure_life_like_complexity(
    agents,
    trait_col = "efficiency"
  ),
  small_mutation = measure_life_like_complexity(
    small_mutation,
    trait_col = "efficiency"
  ),
  large_mutation = measure_life_like_complexity(
    large_mutation,
    trait_col = "efficiency"
  )
)
#>                 n unique_values  entropy      mean        sd
#> original       50            50 2.889669 0.5066717 0.0989657
#> small_mutation 50            50 2.880960 0.5088657 0.1008750
#> large_mutation 50            50 2.907012 0.5286118 0.1977515
#>                temporal_variability mean_abs_change
#> original                         NA              NA
#> small_mutation                   NA              NA
#> large_mutation                   NA              NA

Interpretation of mutation size

Mutation size affects how far traits can move. Small changes may preserve similarity between parent and offspring. Large changes can create more dramatic novelty but may also move traits away from previously successful values.

In artificial-life terms:

Mutation explores nearby or distant possibilities in trait space.

The best mutation size depends on the model and question.

Trait bounds

The lower and upper arguments define the allowed trait range. These bounds are important because not every numeric value makes sense.

For example, efficiency may be limited to a positive range. Speed may have a minimum and maximum. Reproduction thresholds may need to remain biologically or conceptually plausible.

Trait bounds represent simplified constraints.

bounded_mutation <- simulate_mutation(
  agents,
  trait = "efficiency",
  mutation_rate = 0.80,
  mutation_sd = 0.30,
  lower = 0.01,
  upper = 1,
  seed = 6
)

range(bounded_mutation$efficiency)
#> [1] 0.01 1.00

Interpretation of bounds

The range shows that mutation did not push the selected trait outside the specified limits.

This is important because artificial-life models often need constraints. Without constraints, mutation can create values that are mathematically possible but conceptually meaningless.

Novelty is not automatically improvement

Mutation introduces novelty, but novelty is not automatically beneficial.

A mutation may:

improve performance in a particular environment;
reduce performance;
have no clear effect;
matter only under certain conditions;
become useful later if the environment changes.

Selection-like processes determine whether some variations become more common over time. Mutation introduces possibilities. Selection filters them.

This relationship is central:

Mutation generates variation. Selection changes the frequency of variation.

Balance between stability and novelty

Too little mutation may limit novelty. Too much mutation may disrupt inherited structure.

Artificial-life systems often become interesting when they balance continuity and change. This balance is related to broader complexity ideas: systems need enough stability to preserve organization and enough variability to explore new possibilities (Langton 1990; Mitchell 2009).

A useful way to think about it is:

Mutation level	Possible consequence
Very low	Population may stagnate
Moderate	Population can explore new possibilities
Very high	Population may lose stable inheritance

This is not a universal rule, but it is a useful educational guide.

Mutation and evolvability

Evolvability refers to the capacity of a system to generate heritable variation that can support adaptive change.

Mutation contributes to evolvability because it introduces new trait values. But mutation alone is not enough. Evolvability also depends on:

inheritance;
selection;
environmental constraints;
population structure;
survival and reproduction;
continuity across generations.

simulate_mutation() illustrates only one part of this broader process.

Responsible interpretation

Mutation in this package is abstract. It does not represent DNA, molecular biology, or real genetic mechanisms. It represents the general idea that traits can vary.

It is better to say:

The simulation illustrates mutation-like trait variation.

than:

The simulation models real genetic mutation.

It is better to say:

Mutation can introduce novelty into an artificial population.

than:

Mutation automatically improves the population.

Careful interpretation keeps the model clear and academically credible.

Relation to other artificial-life processes

Mutation becomes most meaningful when connected to other artificial-life processes.

Process	Role
Reproduction	Maintains continuity
Mutation	Introduces variation
Selection	Changes trait frequencies
Resource competition	Creates environmental pressure
Population dynamics	Shows system-level consequences

Together, these processes help explain how artificial populations can change over time.

Educational use

This chapter can support several classroom or self-study questions:

Why is variation necessary for selection?
How does mutation differ from variation?
What happens when mutation rate is low?
What happens when mutation rate is high?
What happens when mutation size changes?
Why is novelty not automatically improvement?
Why are trait bounds important?
How does mutation support evolvability?

These questions help learners understand mutation as one part of a broader artificial-life system.

Key takeaway

Mutation introduces novelty into artificial populations by modifying traits. It is most meaningful when combined with inheritance, selection, resources, and population dynamics.

artificialLifeR represents mutation as simplified numeric trait variation. This makes the concept visible and teachable while preserving the distinction between artificial-life toy models and real biological genetics.

References

Darwin, Charles. 1859. On the Origin of Species. John Murray.

Langton, Christopher G. 1990. “Computation at the Edge of Chaos: Phase Transitions and Emergent Computation.” Physica D: Nonlinear Phenomena 42 (1–3): 12–37.

Mitchell, Melanie. 2009. Complexity: A Guided Tour. Oxford University Press.

Nowak, Martin A. 2006. Evolutionary Dynamics: Exploring the Equations of Life. Harvard University Press.