The probability that a word will appear in the New York Times headline tomorrow reduces as more days pass since it last appeared.

In exactly the same way the chance that someone will email you tomorrow reduces in line with the number of days since they last mail you. Thus the odds that an event occurs is a function of recency.

Stuart Card and Peter Pirolli from PARC looked at research into memory. Most people you assume that memory is an imperfect device (because we forget). They were interested in discovering what happens if you assume that it has evolved perfectly to its environment and thus reverse engineer the brain to discover why it is so perfectly adapted.

The results of memory tests show that recollection decreases according to the number of days since that data was last required. In exactly the same way (as a power function) as the email and headline examples above. When put this way it seems no great shock, but looking at human capabilities by asking why did they evolve is very powerful.

They asked three key questions:

• What environmental problem is solved?
• Why is a given system a good solution to that problem?
• How is that solution realised by mechanism?

(Out-thought: these three questions seem an excellent way of tackling any system design problem)

In this context:

What environmental problem is solved?

• Need to remember information that is likely to reoccur

Why is a given system a good solution to that problem?

• Retrieval functions optimised for the predicted need

How is that solution realised by mechanism?

• Network of interconnected concepts

The basic idea is that memory contains concepts. These concepts are associated. Concepts that are more 'activated' are easier to recall. The activation of a given concept depends on two things:

• the base activation of that concept (i.e. probability of occurrence according to model above - put simply stronger if it happened recently)
• the sum of the activation of associated concepts multiplied by the strength of association (i.e. how frequently the two concepts occur together)

In this illustration my concept of rabbit has a base level activation of 3 (a rabbit hasn't occurred in my environment for a while now). This is associated with Easter (which last occurred even further back), but the strength of association is quite high as the concepts of rabbits and Easter often occur together.

This model was termed a spreading activation model, and turned out to be fantastically useful tool when determining how to sort search results. This is also a very close approximation of how neurons interact via synapses and even how the Google search algorithms work.

The concept of a spreading activation model can likely be applied to many different other situations involving information structuring and memory.



Got a suggestion?, email Dave

Can't find what you want? - use the search!

WWW www.arrod.co.uk