Lessons from other species

Wed, 26 Sep 2007 11:40:45 -0700

It is often helpful to study species other than ourselves. They must survive in the same world as we do, and do so with even less brain power than humans. For example, consider a young animal following a mother minutes after hatching or being born. It appears instinctive, but is it? How do the animals know it is their parent? Some insight into these questions came from Konrad Lorenz working with geese. He showed that newly hatched goslings will follow the first moving object they encounter after hatching. Normally that would be the mother goose, and this "imprinting" allows the young birds to recognize who their parents are, and thus who to follow around to learn more about the world. This exposure must occur within a narrow "critical period" or "sensitive period". If a human rather than an adult goose is present during the critical period, the goslings will imprint on the human, and follow them around.

Usually imprinting is studied as a simple form of learning, or as an example of the importance of early experience; however imprinting is also interesting for a vision scientist. It illustrates just how pragmatic vision can be. What is happening when a gosling imprints on an adult goose, human, or even a mechanical contraption? It seems that the young geese use a few simple clues to solve a rather difficult but important problem. Making the recognition process completely instinctive would be expensive, requiring many genes to create a specialized parent recognition circuit in their brains. Instead the birds seem preprogrammed to use some simple sensory cues. The parallel motion of different patches of light can be used to link cues together, to separate them from the background patterns, and to discover the shape of their parent. Goslings are solving the immensely difficult perceptual problem of identifying it's own species and individual parents by first reducing it to a simpler one of motion perception. This works well because normally the parent is the only thing moving around a nest. Such learning and perceptual shortcuts and learning can be useful, especially if they rely on a reasonably unchanging attribute of the environment.

Another important lesson is that nature-nurture distinctions can be difficult to make. Typically the young birds will appear to follow their parents “instinctively”, but this is not true as they must first imprint or learn about their parents. On the other hand there seems to be a built-in tendency to use certain sensory signals to learn something quite specific, and to do so as rapidly as possible.

Perception is hard

Wed, 26 Sep 2007 11:32:06 -0700

Most people give little thought to perception. It seems automatic and effortless. Even when our perception fails, we just laugh at the “illusion” and assume that it could not have been seen any other way. It seems easy to us, but is it really? When computer scientists look at the requirements for building a machine to recognize things, they quickly discover that perception is quite difficult. A visual system has to interpret light that varies over a huge intensity range, consists of different wavelengths, and comes from many directions. Furthermore the pattern is constantly changing. To make things even worse, the external world is immensely complicated, and the range of possibilities from which to choose is large. Every single patch of color we see could arise in multiple ways. Is it a shadow, or is it a surface marking? Is it small and close, or large and far way? Is it coming from one object or two? Scientists have not even come close to building a computer that can match what any human can do. Or even what most “simpler” animals can do.

Our perceptual system has limitations and it can make mistakes. Everyone is familiar with optical illusions, but these are trivial compared to how visual art and modern technologies exploit similar weaknesses "to fool the eye". Science still has a long way to go before we understand how we comprehend the ever-changing patterns of light that reach our eyes. However considerable progress is being made by combining computational and biological approaches, and the next few pages will give a general idea of the sort of mechanisms that we think have evolved to allow us to see. Later entries will be devoted to a more in depth exploration of specific issues.

RANTS

Lessons from other species

Perception is hard