Program Representation for AGI
Program Representation for General Intelligence
Authors:
Year: 2009
Venue: Proceedings of AGI-09, Atlantis Press
Links: PDF
Summary
Argues that programs — well-specified, compact, combinatorial, and hierarchical — are the right representation for the complex relational, typed, and functional structure of real-world data, and that AGI systems need direct support for programmatic representation and learning rather than flat feature/grammar representations. This is the published statement of the program-representation foundations underlying MOSES (combinatory logic and evolutionary program learning); the AGI-09 conference companion is "A Representation of Programs for Learning and Reasoning".
Relevance to Hyperon
The canonical published statement of the program-representation foundations underlying MOSES — directly informing MeTTa's program representation and the MOSES evolutionary program-learning algorithm being reimplemented for Hyperon.