When constructing the fuzzy rules, the human was actively encouraged to take advantage of all human thought processes and percepts, as this forms the real basis of the project - if these human characteristics can be used to gain an advantage over standard ML techniques. However, the visualisation techniques available were limited in sophistication, due to the limited scale of the project. The techniques that were used were sorting of the training set on different fields (though usually with the class sorted first) and subspace plots, where two or three of the attributes were plotted on a plane or in a volume along with their classifications. For boolean valued attributes, such as those in the Voting Records dataset, subspace projections are essentially statistical, multi-dimensional histograms. It was originally hoped that several sophisticated approaches to the visualisation of the training set could be applied, such as scientific visualisation software such as AVS and IBM Data Explorer (IBM DX), and full statistical makeup properties of the dataset, however, these proved to be too large to be properly incorporated into the project. This subject is discussed more fully in Section 4.2.3.
It was also important that the human attempted to learn the patterns in the dataset, as opposed to using external knowledge base information. This was important because we were testing the human's ability to learn patterns, not their knowledge on the particular domain. For example, with the Congressional Voting Database, a US Political Scientist or Analyst would almost certainly be able to correctly predict a Congressman's party (Democrat or Republican) given the questions and the Congressman's votes on those questions. However, this is because they know what it means to be a Democrat or Republican, rather than simply observing how Democrats and Republicans vote. This is far from the objectives of this project, and luckily the human involved had no prior knowledge on what Democrats and Republicans believe in (and therefore would be likely to vote for). This subject is discussed further in Section 4.2.4.