Knowledge-based artificial intelligence is a knowledge-based system that provides these benefits, using KBpedia as the reference example:
It provides a rich, easily selectable pool of labels for doing supervised machine learning. In KBpedia, about 47K of the 55K reference concepts (RCs), or about 85%, are fine-grained types, that are logically organized and can be sliced and aggregated to create positive and negative training sets and reference ("gold") standards. Definitions, synsets, and robust text are available for (nearly) all RCs. Types exist for entities, events, concepts, attributes, and relations;
Because of its rich structure, KBpedia may be sliced and diced to create target pools from 20 million entities (standard version) and their features for unsupervised learning; deep learning leverages both supervised and unsupervised;
The knowledge graph is organized to serve KM and AI purposes; it is grounded as a digital reference -- for primitives, particulars and generals -- that is coherent and computable. Aggregation, restriction, inference, and other logical operations may be applied against the graph using semantic Web standards.
The purposeful design for computability and to support AI and machine learning are what distinguish KBAI from standard knowledge bases. Fortunately, existing knowledge bases may be restructured to serve KBAI purposes, as is the case with the six (6) KBs in the core KBpedia.
KBAI is geared to applications in knowledge representation, natural language understanding, entity and relation detection, 'natural' classing (via attribute evaluation), mapping to external instances and schema, search, feature generation and extraction, and ML prep and testing. Vision, speech, image or other recognition systems would likely not train against KBAI, though portions of KBAI may apply to common sense, decisionmaking or planning systems.
KBpedia exploits large-scale knowledge bases and semantic technologies for machine learning, data interoperability and mapping, and fact extraction and tagging.