Many critical user interaction design decisions are made in the heat of detailed development. These include simple parameter choices or more complex weightings in intelligent algorithms. Many would be appropriate for expert design review, user-preference choices or optimisation by machine learning, but they are buried deep in the code. Although the developer may realise this potential, the location of the decision is far removed in the code from where user feedback occurs, data can be collected and machine learning could be applied. This position paper describes several case studies and use them to frame an architectural challenge for tools and infrastructure to uncover these hidden variables to make them available for machine learning and user inspection.
|