•Catalase Website Index •James •Techno •Evolution •Landscape •

Evolution - The Evolutionary Landscape

2.1 The Evolutionary Landscape Hills and Valleys - Limitations of a crude model of 'the evolutionary landscape' Levels of Feedback - Why it is important to consider different levels in feedback processes.

2.1 The Evolutionary Landscape

Hills and Valleys - Limitations of a crude model of 'the evolutionary landscape'
Levels of Feedback - Why it is important to consider different levels in feedback processes.

Hills and Valleys Evolution is often presented in terms of 'optimisation', using a model like the following: Imagine a hilly landscape - height above sea level represents fitness, positions correspond to different genomes. Small changes in genomes lead to small changes in fitness. Evolution is optimisation on this landscape - the finding of the highest point. The analogy is used to show that evolution may "get stuck". Having found the peak of one hill, any small changes in genome are deleterious. We expect further progress to be slow as it cannot happen through small changes. The mental picture can be useful, but it is as often misleading. There is for example no way for the model to represent repeated genes. Repeated genes are important in evolution. Even if you don't accept my arguments earlier that they are significant in accelerating evolution, there is no doubt that they occur. As an optimisation problem evolution also has some important differences to conventional optimisation. The number of parameters is huge. (The landscape model only shows two) The function to optimise changes over time. The landscape 'looks' very different for different parameterisations. The third of these points is probably the most important for the idea that evolution evolves efficient mechanisms for evolving. In conventional optimisation different parameterisations amount to rotating the landscape and don't really change its character. In evolutionary optimisation parameterisations amount to discovery by the genome of rules for successful designs - applying those rules radically changes the shape of the landscape.

Hills and Valleys

Evolution is often presented in terms of 'optimisation', using a model like the following:

Imagine a hilly landscape - height above sea level represents fitness, positions correspond to different genomes. Small changes in genomes lead to small changes in fitness. Evolution is optimisation on this landscape - the finding of the highest point.

The analogy is used to show that evolution may "get stuck". Having found the peak of one hill, any small changes in genome are deleterious. We expect further progress to be slow as it cannot happen through small changes.

The mental picture can be useful, but it is as often misleading. There is for example no way for the model to represent repeated genes. Repeated genes are important in evolution. Even if you don't accept my arguments earlier that they are significant in accelerating evolution, there is no doubt that they occur.

As an optimisation problem evolution also has some important differences to conventional optimisation.

The number of parameters is huge. (The landscape model only shows two)
The function to optimise changes over time.
The landscape 'looks' very different for different parameterisations.

The third of these points is probably the most important for the idea that evolution evolves efficient mechanisms for evolving. In conventional optimisation different parameterisations amount to rotating the landscape and don't really change its character. In evolutionary optimisation parameterisations amount to discovery by the genome of rules for successful designs - applying those rules radically changes the shape of the landscape.

•Index•2.1 Landscape•

Levels of Feedback The following example describes two levels of feedback in a piece of industrial design: In a technical article on positioning of the 'read head' of a Phillips video disc player there is a description of a two level feedback system. There are two servo systems that control position of the head: One moves the head left or right within a 'carriage', The other moves the 'carriage' as a whole, including the read head, left or right. The first servo system is much faster acting but can only move the head a few millimeters from a centre position within the carriage. The second servo system is less responsive, it uses the average deflection from the centre position of the fine servo system as an input signal. If the read head is on average slightly away from centre of the carriage the slower servo system will move the carriage to correct this. The combined system is responsive to tiny positioning errors and is able to track accurately over the full distance the carriage can move. The above example illustrates how different feedback systems at different levels can interact. Such interaction between levels is very important for biological systems and is central to mechanisms that accelerate evolution. Two examples of interactions of levels of feedback in evolution are given below: A widely accepted acceleration of evolution is called the Baldwin Effect . In essence adaptability of an organism improves the chances of adaptation of the genome. A crucial feature that makes the Baldwin Effect work is that the organism, once it finds 'a good trick' locks on to it and continues to use it. The Baldwin Effect relies on fine scale feedback that recognises that 'a good trick' has been found and that continues to use it. Natural selection is analogous to the coarser feedback. The example of regulation of multi-copy genes is a more striking example of a model of multiple levels in feedback. In that model the cell has evolved a fine scale feedback that evaluates individual genes and this works in concert with the coarse level of live-or-die feedback of natural selection.

Levels of Feedback

The following example describes two levels of feedback in a piece of industrial design:

In a technical article on positioning of the 'read head' of a Phillips video disc player there is a description of a two level feedback system. There are two servo systems that control position of the head:

One moves the head left or right within a 'carriage',
The other moves the 'carriage' as a whole, including the read head, left or right.

The first servo system is much faster acting but can only move the head a few millimeters from a centre position within the carriage.
The second servo system is less responsive, it uses the average deflection from the centre position of the fine servo system as an input signal. If the read head is on average slightly away from centre of the carriage the slower servo system will move the carriage to correct this.
The combined system is responsive to tiny positioning errors and is able to track accurately over the full distance the carriage can move.

The above example illustrates how different feedback systems at different levels can interact. Such interaction between levels is very important for biological systems and is central to mechanisms that accelerate evolution.

Two examples of interactions of levels of feedback in evolution are given below:

A widely accepted acceleration of evolution is called the Baldwin Effect . In essence adaptability of an organism improves the chances of adaptation of the genome.
A crucial feature that makes the Baldwin Effect work is that the organism, once it finds 'a good trick' locks on to it and continues to use it. The Baldwin Effect relies on fine scale feedback that recognises that 'a good trick' has been found and that continues to use it. Natural selection is analogous to the coarser feedback.
The example of regulation of multi-copy genes is a more striking example of a model of multiple levels in feedback. In that model the cell has evolved a fine scale feedback that evaluates individual genes and this works in concert with the coarse level of live-or-die feedback of natural selection.

•Catalase Website Index •James •Techno •Evolution •Landscape •

"Evolution - The Evolutionary Landscape" page last updated 5-July-2003