Thesis Rhymes With?
by Carson Reynolds
Thesis: adaptation can be applied generally to improve interaction.
The thesis could be decomposed into:
- a theory of simple adaptation
- a discussion of how and why adaptation in interaction has failed and succeeded in the past coupled with a discussion of the attributes of successful adaptive interaction
- an artifact motivated by my theory of adaptation would be necessary to test the thesis.
Below is a brief first stab at each of these sub-topics:
Many have provided theories of how complex adaptive systems operate. Few have tried to reduce adaptation to its essential components, and the behavior of these components. To understand adaptation, one must understand how an adaptor behaves. An adaptor maps one system onto another, or itself. For instance, an adaptor plug takes one system’s physical structure and converts it into another system’s standard. Alternatively, one can imagine one population’s genes being mapped into a new genetic space culling, recombination, and mutation. Here procreation acts as the “adaptor” between two generations. Adaptation is the process of continually (recursively?) creating adaptors. It is the continual process of making the output of one system suitable
Adaptive interface systems have often failed in the past. Many of these failures can be attributed to violations of known usability and design heuristics (consistency, predictability, user-control), for instance. They have also failed because many of the systems acted to alter the interface based on very little information. It would be interesting to re-run some of the most significant experiments on adaptive interface systems after correcting some of these shortcomings. For instance, re-running Mitchell and Shneiderman’s adaptive menu experiment with a menu that offers an alternative (reordered) menu at the bottom, while leaving the original intact. Users could select the alternative reordered menu at their convenience.
To test this theory of adaptive interaction, an artifact should be created that doesn’t violate known usability and design heuristics. It should furthermore collect enough information such that adaptors can be created that map interaction into new, user selected, points in “interaction space.” One idea would be to create a set of user-adjustable control panels that automatically suggest new settings, and can be further decomposed down into a fully user-programmable language for interaction configuration. The suggested new settings would be informed by a large amount of information collected about interaction: user affect, error rates, performance statistics, stability statistics, user critical incident report, expert evaluation, and validation. An extension of this might be an entire operating system on top of a reconfigurable computer that adjusts its behavior in the same manner.