Fomitters

Fomitters are a speculative concept used in cognitive science and artificial intelligence to describe any agent, natural or artificial, that emits structured perceptual input—"forms"—which can guide downstream interpretation or learning. The term combines form with -itter, implying "emitter." In theoretical models, fomitters create controlled stimulus patterns that reveal how an agent's inductive biases influence categorization, generalization, and representation learning. There are two main interpretations: external fomitters, such as experimental stimuli that researchers design to probe perception, and endogenous fomitters, hypothetical neural or computational processes that generate internal forms for learning.

Mechanism: Fomitters generate form-rich inputs—structured patterns, shapes, or feature combinations—intended to be taken as priors by a learning system. The responses of the system to these inputs reveal its reliance on certain forms and rules.

Applications: In AI, fomitters help study the role of inductive biases and representation formation; in philosophy of mind, they provide a language for discussing inherent structured priors. Critics argue that the term is poorly defined and risks conflating mechanisms with effects.

History: The term first appeared in speculative discussions and has been used in thought experiments rather than empirical research. Some scholars treat fomitters as a metaphor for prior structures embedded in datasets, models, or curricula.

See also: Inductive bias, priors in machine learning, representation learning.