Evolution Flashcards
Name the features necessary in an evolution model
- Replicators: Agents that reproduce
- Variation: Variability in population
- Differential survival or reproduction: variances affect ability to survive and reproduce
How do we simulate replicators, variation and differential survival?
- Replicators: Give each agent genetic information called genotype, which can be a sequence of N digits
- Variation: Create a population with different genotypes
- Differential survival: define a function that maps from each genotype to a fitness (ability to survive/reproduce)
How do we map survival?
To a fitness landscape, where each genotype is a location in an n-dimensional space, known as a fitness landscape.
It can be random, and evolution still occurs.
What is the outcome of the evolution model without differential survival?
A random walk of various genotypes dying and reproducing
What happens if we add differential survival with no mutation?
Agents with higher fitness are more likely to reproduce.
Agents with lower fitness are more likely to die.
Over time we expect more agents with higher fitness and less agents with lower fitness.
The ultimate outcome is all agents with the same fitness/same location.
Now what happens if we add mutation?
We see diversity. Agents can move location and high fitness agents reproduce, while low fitness locations die out. Over time, the population migrates to the high fitness location.
Higher overall fitness compared to no mutation.
How can we identify species?
Species can be identified as clusters of agents with small genetic differences
What is the rational decision in the prisoner’s dilemma?
Both to defect
What is the problem of altruism?
If natural selection is highly competitive landscape, why haven’t the genes for altruism died out?
What is the human outcome of prisoner dilemma tournaments?
Tit for tat, they do the same outcome as the opponent last round.
What are properties of successful strategies in PD tournaments?
Nice- generally cooperate as often as they defect
Retaliating - strategies that retaliate if their opponent defects do well
Forgiving - strategies that forgive do well
Non-envious: most of the successful strategies do “well enough”
TFT has the above properties
What is the outcome of simulating PD evolution?
Populations of defectors are vulnerable to invasion by nicer strategies
Populations of strategies that are too nice are vulnerable to defectors
The average level of niceness is high, but oscillates
There is no specially optimal strategy, but TFT is somewhat optimal
Only some retaliation is required
