Advanced Logicist Machine Learning (ALML): Phase 1
Abstract
Tools useful to humans needn t be capable of human-level learning. AlphaGo might for example be a wonderful training tool for humans competing against other humans in Go - despite the fact that AlphaGo knows next to nothing. We seek to engineer not mere tools, but machines that can learn as humans do, and thereby acquire suffcient knowledge to be not tools, but teammates. Seeking this, we advise the pursuit of a new form of learning rather more powerful than today s "ML." This form we dub `advanced logicist machine learning (ALML; rhymes with "pal bull"). Two sub-types of ALML will be formalized and implemented in the course of Phase 1, one based in a revolutionary kind of plan generation and plan recognition, and the second in a new kind of reasoning that is lead by analogy, but blends it with deduction and other inductive modes of reasoning. A number of original-in-their-own-right "facilitating" structures and algorithms will be needed to enable ALML, and they will be supplied. Phase 1 s testbed environment will be that of a classroom and supplementary "recreational" space, suitable for collaborative learning for artificial and human agents.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2017
- Source ID
- N000141712115
Entities
People
- Selmer Bringsjord
Organizations
- Office of Naval Research
- Rensselaer Polytechnic Institute
- United States Navy