As a simple example of how natural selection can put functional information into a population of organisms, I like to point to dandelions growing in my yard. They take energy from the sun and grow. If there are two different genotypes, one of which processes that solar energy more efficiently, the frequency of that genotype will increase as it grows more than the other one, competes better, and produces more seeds.
When we consider a scale of fitness, or a scale of energy-processing efficiency, this process of natural selection shifts the distribution of genotypes so that more and more of the dandelions have higher efficiency of energy processing. It does not take much calculating to see that the population then has increased functional information. The information does not come mysteriously from nowhere: energy is flowing from the sun into the plants, and ultimately heat is lost to the environment. All of which is a net increase in entropy. Evolution is simply slowing down this overall increase, not reversing it.
This would seem to be an elementary example. But at the ID-advocacy website Uncommon Descent, their regular post author Dr. Eric Hedin gets examples like this wrong. In a recent post Natural sources of information? he quotes his own book "Canceled Science: What Some Atheists Don't Want You to See":
In systems which are far from thermodynamic equilibrium, differences or gradients in various thermodynamic variables may exist within the system and between the system and the environment. It has sometimes been mistakenly assumed that these gradients could generate the information found in living systems.[i] However, while thermodynamic gradients may produce complexity, they do not generate information.
This is particularly surprising since Dr. Hedin is trained as an experimental plasma physicist, actually at my own university. He obviously got a solid grounding in thermodynamics and information theory. But somehow he has failed to apply that correctly to evolving organisms. Let's look more closely at his discussion ...
In the excerpt he quoted, Eric Hedin gives examples of natural processes that tend to degrade information. For example:
Our sun is a low-entropy source of thermal energy that the Earth receives via electromagnetic radiation. This thermal energy is useful energy in the thermodynamic sense because it can be used to do work. The same is true of energy released by gravitational potential energy being converted into kinetic energy or heat. Waterfalls and solar collectors can produce energy for useful work, but they are sterile with respect to generating information. In fact, sources of natural energy (sunlight, fire, earthquakes, hurricanes, etc.) universally destroy complex specified information, and never create it. What will happen to a painting if left outside in the elements? What happens to a note tossed into a mulch pile? They degrade by the actions of nature, until all traces of information disappear. Or consider an unfortunate opossum killed on a country road. Will its internal, complex biochemistry increase or decrease with time due to the effects of natural forces? We all know the answer. If not eaten by scavengers, it eventually turns to a pile of dirt.
These examples are fine, but they do not consider what happens when there are biological organisms present, such as my dandelions. They can take sunlight, air and water, and make more dandelions, which would seem to contradict the generalization that Dr. Hedin just made.
He would probably answer that I have skipped over a part of his post where he addressed the case where there is a "design template":
Decreasing thermodynamic entropy can only be leveraged into information if a design template and the mechanism to employ it already exist. In this case, the desired information is not being created by the action of the low-entropy energy source; it is merely being transferred from the template to an output product.
For an individual clone of dandelions, he would be asserting that when the number of dandelions of that genotype increases, that this is no gain of information. But if the information is "transferred" to the offspring, this can happen multiple times, if the parent produces lots of seeds. Which is an odd way of bookkeeping the amount of information.
Things get even worse for his argument when we consider the information that gets put into the genome when natural selection favors one genotype over the other. That is the case I started with. Choosing one genotype out of a mixture, as the predominant genotype on my lawn, is a fundamental operation in information theory. And if that genotype is the better one on a scale of specification or function, that is an increase of functional information in the population of genotypes.
Is that information coming from a "template"? No. Each genotype might (or might not) be considered as one of Hedin's templates. But where is the super-template that says which clone of dandelions will ultimately predominate?
Someone might point to William Dembski and Robert Marks's concept of "active information", which seems to describe information existing in the fitness landscape, the table of fitnesses of all possible genotypes. But that concept, whether sensible or not, is bookkeeping information that is transferred by natural processes, such as mutation and natural selection, into the population of genotypes. It therefore already violates Hedin's generalization, so it cannot be used to rescue it. Hedin's sweeping statements are contradicted by the natural processes of reproduction, and the process of natural selection which flows from them.