Abstract
We propose a framework that replaces scalar confidence estimates in AI outputs with structured, multi-dimensional representations termed knowledge tokens. Each token encodes orthogonal epistemic attributes including provenance weight, empirical grounding, consensus density, and temporal freshness. Unlike conventional probabilistic confidence scores, knowledge tokens provide composable, inspectable, and tradeable representations of informational quality. This enables downstream systems to reason not only about likelihood but about the structure and reliability of knowledge itself. We formalize the representation, integration into model architectures, training objectives, and evaluation methodologies, and outline how such tokens enable new forms of reasoning, aggregation, and economic exchange over knowledge.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Manuel-Devadoss, Johny and Prasad, Deepthi, "Knowledge Tokens for Multi-Dimensional Epistemic Representation in AI Systems", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10036