ifc-0119

8.5 Active probe selection

Uniform exploration wastes experiments on redundant directions. An active system selects the next probe using its current uncertainty, the candidate it is trying to distinguish, and the cost of intervention. A generic policy has the form

\[ A_{t+1} \in \arg \max _{A\in \mathcal B_t} \Bigl[ \operatorname {Discrim}(A;H_t) +\lambda \operatorname {Novel}(A;\mathcal H_t) -\mu \operatorname {Cost}(A) \Bigr], \]

where \(H_t\) is the current hypothesis set and \(\mathcal H_t\) the probe history. This is a decision rule, not a universal scalar definition of creativity. Each term must be supplied by the application.

Proposal-aware selection asks a sharper question than generic information gain: which available experiment most directly separates the leading structural proposals? In a simulator this may be an intervention. In mathematics it may be a small countermodel. In a corpus it may be a targeted search for a claim that distinguishes two mechanisms. Such probes connect exploration to admission rather than rewarding novelty for its own sake.