Animal Learning: The Baldwin Effect

In a past post, we talked about how animals learn using learning algorithms like classical conditioning. But Donald Campbell pointed out that evolutionary theory has a hard time explaining the existence of complex animal instincts:

Complex adaptive instincts typically involve multiple movements and must inevitably involve a multiplicity of mutations at least as greater in number as the obvious movement segments. Furthermore, it is typical that the fragmentary movement segments, or the effects of a single component mutation, would represent no adaptive gain at all apart from the remainder of the total sequence. The joint likelihood of the simultaneous occurrence of the adaptive form of the many mutations involved is so infinitesimal that the blind-mutation-and-selective-retention model seems inadequate.

DONALD CAMPBELL in Evolutionary Epistemology, Rationality, and the Sociology of Knowledge, P. 61

But if animals evolved learning algorithms first then this problem is resolved:

The adaptive pattern being thus piloted by learning, any mutations that accelerated the learning, made it more certain to occur, or predisposed the animal to certain components responses, would be adaptive and selected no matter which component, or in what order affected. The habit thus provided a selective template around which the instinctive components could be assembled. (Stating it in other terms, learned habits make new ecological niche available, which niche then selects instinct components.) It is furthermore typical of such instincts that they involve learned components, as of nest and raw materials location, etc. (p. 61)

DONALD CAMPBELL, P. 61

The above theory that Campbell suggested was actually first proposed by James Baldwin and became known as “The Baldwin Effect.” Under this theory, species that are able to learn adaptive behaviors will then later experience an advantage to learning that behavior if they have part of the behavior built-in to their genome. The end result is that an initially learned behavior will soon become a genetic-based instinct. But the learned behavior comes first and first must create an environment where each individual mutation is an advantage.

As Nicholas Christakis puts it:

…over time certain behaviors that previously were learned may actually become genetic. This is know as the Baldwin effect, and it works as follows: In any given generation of a species, some animals might be born, by chance, with gene variants that make it easier for them to learn a particular behavior. For instance, some birds might learn a song more easily than others because they have brains more capable of producing the typical introductory note of the song. If this behavior is adaptive (for example, if the song helps the birds attract mates), the nit confers a fitness advantage on them, and the genes that make it easy to learn the behavior get reinforced and expanded in frequency as the population reproduces. The same thing continues to happen. In each generation, those animals fare best who are most easily able to learn the target behavior, because they innate manifest more and more of it. Having acquired the innate capacity to produce the initial note, some birds might now be born with other gene variants that make it easier for them to produce the motifs of the song. These too could become innate, across generations, in response to selection. Eventually, over time, the entire behavior – and not just its predicates—comes to be encoded in the genes. What was previously a nongenetic action has become a genetically encoded one.

Nicholas Christakis in Blueprint: The Evolutionary Origins of a Good Society, p. 325

The Baldwin Effect is today considered part of modern evolutionary theory.

On Chess Playing Dogs

Dennis Hackethal, on the Do Explain podcast, used an example of a hypothetical chess playing dog. His point was that it would be difficult to know where the knowledge came from, from learning or from genetic instincts, even if the behavior was quite sophisticated or appeared intelligent to us; so we can’t simply tell if the dog is intelligent by looking at how sophisticated its behavior is. And this is certainly true, the ‘smarts’ might all come from the genes rather than from the animal’s own intelligence.

But for selection pressure to even work in this circumstance (even if controlled by humans), some dogs must initially show some improved ability to play chess compared to others. Since today they have no instincts at all for chess, that implies the dogs must initially utilize their ability to learn to, say, make the best opening chess move. Since dogs do have learning algorithms already built into them, some dogs would be better at making that opening chess move than others. It is in this way that the ability to learn will proceed instinct. So all complex instinctual behaviors have to first have an ‘animal intelligence’ component. That is unless a reasonable account can be made of how every single part of that behavior was itself adaptive.

Animal Learning is crucial to animals being able to develop instinctual behavior, so we know animals are ‘intelligent’ in some meaningful sense and not all of animal behavior can be instinct.

3 Replies to “Animal Learning: The Baldwin Effect”

  1. I wonder if the same principle holds for the relation between the conscious and the unconscious mind. At first when learning a new task the conscious mind needs to control the body very precisely and with great effort. After a while the unconscious mind has picked up (evolved?) the common patterns and the conscious mind can start to give instructions to the body on a higher level. After a while the unconscious mind can become completely autonomous.

    How does the switch from detailed control to high level control happen? Could the conscious mind put out patterns for both the high level and the low level control at once and gradually scale back on the low level patterns while it evaluates how things are going with the unconscious mind’s learning process?

Leave a Reply

Your email address will not be published. Required fields are marked *