Crows can count

Crows are fascinating because they are collectively very active and vocal. It is often unclear what they are doing but there seems to be a lot of communication involved. In particular, they have many different calls.

An important feature of their calls is that they often involve a specific number of unit calls, which I call caws. For example there is a three caw call that may be repeated a number of times. There is also a four caw call that includes a slight pause between the second and third caws, sort of caw-caw caw-caw.

In order to make these calls the crows have to count the caws. This is not counting in the sense of naming the numbers. It is not like saying or thinking one-two-three, etc. Rather it is counting in the sense of knowing how many caws have been made. The hearers must also count the caws if communication is to occur.

This is really a case of the distinction between verbal thinking and non-verbal thinking, which I discuss early on in this blog. Because so much human thinking is verbal, it can be hard to see what non-verbal thinking looks like. But I can see that there are, say, three horses in view without going through the numbers one, two, three mentally.

As with other cases we have discussed, it is useful to think about building robots that do what the animals do. In order for robotic crows to make or respond to these various calls, there would have to be some sort of counting mechanism. In short, crows can count.

The amazing cow circle

When a small herd of cows, with calves, is properly threatened, it may form a circle. Each cow facing outward, with the calves in the middle, protect by the herd. I have seen this done, when a herd of about a dozen herefords, with several calves, was threatened by several dogs.

The circle was amazing. Consider the complex decision making and group coordination required to execute this behavior.

First the herd has to collectively decide that the proper threat exists. How this is done is of course a deep mystery.

Then they have to form up into a good circle. This is far more sophisticated that simple grazing together. The Roman battle square comes to mind. How does each cow decide where to go, or who to get next to?

To see the complexity, consider this question. If you had a dozen robotic cows, how would you program them to quickly form a good cow circle? This is clearly a grand challenge. Until this question can be answered I would say that we do not understand the behavior.

Then the calves have to get into the center. A cattleman friend tells me that there is a call that mother cows can make that brings their calf running. Perhaps this plays a role in the protective circle behavior. There may even be a specific "get into the center" call. While perhaps not language it would certainly be auditory communication. (The crows seem to do a great deal of this.)

Then when the threat has passed, as it did when I chased the dogs away, they have to collectively decide that it has passed and go back to grazing.

Instinct tells them how to do all this, but the decision making still has to be done in each specific instance. The cow circle is an amazing case of collective action.

Herd behavior and cascades (in humans)

There is a lot of research on what is called "herd behavior" in humans, which might be useful when applied to herd animals like horses. The basic idea is that in some cases people do what they do because they see others doing it. There are also some specialized models of human herd behavior, which are called "cascades."

In the last two decades several distinct sorts of cascading behavior have emerged as topics of research. Some of this work involves the development of mathematical models which might be useful in modeling animal group behavior.

These include the informational cascade, the information cascade, the reputational cascade, the availability cascade and herd behavior. The relative size of these various research efforts can be gauged from the number of documents found by Google Scholar, when one searches on the name of the behavior. As of this writing these results are as follows:
 "Herd behavior" gives over 15,000 hits.
"Informational cascade" gives about 1700 hits.
"Information cascade" gives about 3000 hits.
"Availability cascade" gives about 200 hits.
"Reputational cascade" gives about 100 hits.

Taken together this amounts to a great deal of research into herd-like behavior.

Collective animal behavior

http://en.m.wikipedia.org/wiki/Collective_animal_behavior has a nice summary of some of the research. Clearly there has been quite a lot.

While some of it is in the form of decision rules, these seem to be too simple for the kind of decision making we have been discussing. Our view is that animals do complex reasoning using instinctive expert knowledge.

There is nothing anthropomorphic about this, except that we are forced to use our own concepts in order to try to understand the concepts that the animals are using. But we are not attributing these human concepts to the animals. If anything, our concepts are something of an impediment.

On the other hand some of the conjectures regarding animal behavior seem to be too complex. For example, the Wikipedia article suggests that group decision making may be based on the Condorcet method of voting, but this seems entirely too anthropomorphic. See http://en.m.wikipedia.org/wiki/Condorcet_method This method involves not merely voting but a ranking of alternatives by each voter. These ranking votes are then tallied by someone. It is an extremely complex procedure.

Horses grazing in motion are not repeatedly voting on the direction they will move. I suspect that something very different from voting is going on. For example, something that has degrees of importance, which voting does not have. Invoking the human concept of voting makes it harder to see this, not easier.

Representing instinctive knowledge

Given that an instinct is an inherited body of expert knowledge, we should be able to apply the methods of artificial intelligence to build a model of an instinct. Such a model might tell us interesting and useful things about the behavior of the critters in question, such as horses.

For those who are not familiar with these methods, here are some short, non-technical introductory readings. The field itself involves a lot of math and many of the articles linked to from these Wikipedia articles are much more technical. 

Knowledge representation is the core field.

http://en.m.wikipedia.org/wiki/Knowledge_representation_and_reasoning

Knowledge engineering is the method for uncovering the knowledge.

http://en.wikipedia.org/wiki/Knowledge_engineering

Expert systems are examples of how knowledge based decision making can be modeled.

http://en.wikipedia.org/wiki/Expert_system

I have been involved with this stuff since I was on the faculty of Carnegie Mellon University in the 1970's. In fact I worked with some of the pioneers, especially Herb Simon, who got a Nobel prize in 1978.

Dogs and Cats (and skillful affection)

Dogs and cats seem to be a difficult case and I have been wondering why. There seem to be at least two reasons. First, their behavior is so tied up with ours that it is hard to see the instincts at work. Second, a lot of what is instinctive is not thought of as knowledge based.

The whole point of domestication is to work well with humans, so this is the focus of the instincts. For dogs and cats these instincts include affection (giving and seeking), obedience, trust, loyalty, companionship, etc. There are others, like play, warning and defense with dogs, or hunting with cats.

In humans the traits of affection, obedience, trust, loyalty, etc., are not thought of as knowledge based, so we also do not see them that way in dogs and cats. This is probably because they are instincts in humans.

But in fact each of these activities involves a great deal of decision making and therefore requires considerable skill, which means expertise. The same is true of horses and other domestic animals, but perhaps to a somewhat lesser degree.

For example, it is not enough to want to be affectionate (whatever that might mean), one also has to know how to do it. When it comes to dealing with humans, the differences between wild and domestic animals are dramatic. Domestic behavior is highly complex, requiring a lot of skill.

As my wife puts it, dogs are professional people pleasers. Most do it well. The research question is what do they have to know in order to do this well? It is not a simple question, by any means.

Basic versus applied and "The Horse's Manifesto"

The distinction between basic and applied science is worth considering here. What we are doing is basic science, which means looking for understanding without regard for how useful it might be.

It is not that we are indifferent to the potential application of what we figure out and that is a common misconception about basic science. It is just that the focus has to be on what can in fact be understood, because that is a hard question in itself.

Science is based on the fact that there are cases where simple rules can explain complex behavior. This is true for physics, chemistry, biology and behavior, and all the varied sciences. Thus the first challenge in doing science is to identify specific cases where important rules can actually be discovered. The art of science is to identify important answerable questions.

But there is often an overlap between important basic science and importantly useful applied science. To make this point in our case of critter cognition, consider three elegant essays called "The Horse's Manifesto" by Lauren Fraser.

Fraser is the Chair of the Horse Division of the International Association of Animal Behavior Consultants. The focus of IAABC and its member consultants is the scientific treatment of behavioral problems, many of which are cognitive in nature.

The connection between IAABC's applied cognitive science and our basic cognitive science can be seen in the fact that Fraser's three essays touch on many of the topics we have been exploring. She talks about the three "F"s of Friends, Forage and Freedom, while we talk about herd decision making, grazing, play, movement, etc. In some cases even the details overlap, such as with what we call grazing in motion, which she also discusses.

Here is the difference. Fraser points out that these behaviors are deeply based on instincts, such that interfering with these instincts can lead to common behavioral problems. This is applied science. What we are trying to do is figure out how to understand these instincts much more clearly than can presently be done. This is basic science. Everyone talks about instincts but how do we describe one scientifically? Hopefully if we are successful the results will flow into the applied arena, to help solve the kind of behavioral problems that Fraser describes, even though that is not our primary concern. This is how basic science feeds applied science.