from data to rules, from intention to form

A dialogue between computational learning and design, to foster academy-industry transfer and innovation.

Where many interdisciplinary initiatives merely juxtapose competences, the Center sets out to investigate the deeper level at which the two practices are already speaking to one another: the level of inference, of constraint, and of representation.

The first aspect is the form of inference. The classical distinction, deduction (from the general to the particular, with necessity), induction (from the particular to a regularity), abduction (from an effect to the hypothesis that best explains it), offers a precise language for describing what happens in both practices. 

Computational learning is predominantly inductive: it generalizes from examples. But in the moment a model proposes the most plausible explanation of an ambiguous input, it slides toward abduction. Design has historically been the art of abduction: faced with a need, the designer does not deduce the solution, but conjectures it as the best available hypothesis, to be tested afterward. 

The second aspect concerns how logic is embedded in the system. At the two extremes: rule-based systems embody an explicit logic, declared in readable, binding rules; generative systems embody an implicit logic, distributed across a model’s weights and not directly inspectable. Design operates precisely in the space between these two poles: its function is to decide what to make explicit — which constraints to bring to the surface, which to leave implicit within the system, how to give perceptible form to a logic that would otherwise remain opaque. Along this axis, to design is to govern transparency.

The third apsect is that both computational learning and design operate by building and inhabiting a space of possibilities. In machine learning this is the feature space, the choice of dimensions along which data is represented determines what the model can or cannot learn. In design this is the design space, the set of conceivable solutions, likewise defined by the way the problem was framed. In both cases the decisive move happens before the solution: in structuring the space within which the solution will be sought.