Cycorp CEO Doug Lenat coordinated the AAAI Spring Spring Symposium at Stanford University held on March 23-25, 2020 on Combining Machine Learning and Knowledge Engineering in Practice.
More specifically, there are three different types of representation (and therefore reasoning) worth differentiating here, each of which has its own advantages and limitations:
- statistical training of neural nets on big data
- knowledge-graph triple- and quad-stores that make taxonomic relationships explicit, and
- higher-order logics that have the same expressive power as natural languages such as English.
In Cycorp terms, these correspond to:
- machine learning Heuristic Level modules,
- CycL assertions that happen to be first-order (or propositional), and
- modal and other meta-level assertions and rules in CycL.
The talks at the symposium focused on (i) how these different sources of power might synergistically work together and (ii) empirical case studies of AI applications where two or all three of these have been brought to bear and harnessed.