On Thursday, August 4, IIIM organized a workshop on self-programming in AGI systems at the Fourth Conference on Artificial General Intelligence (AGI) in Google’s headquarters, Mountain View, CA. The workshop featured presentations by seven authors on the subject of self-programming systems and a panel discussion among the participants.
Discussion and Presentations in a Nutshell
The presentations covered general issues such as what distinction, if any, should be made between self-programming and more conventional ideas of learning (Goertzel), and the appropriate role of self-programming in the development of AGI systems; e.g. can self-programming be used to help generate AGI systems, or does a human-level intelligence need to be in place for the system to meaningfully utilize these capabilities.
Other presenters focused on particular aspects or implementations of self-programming systems. Such as the role of creativity (Leijnen), imitation (Hall) and program-space search (Skaba). Several existing architectures were discussed as they partained to self-programming capabilities. One in particular was BECCA, a general system for feature creation and general reinforcement learning (Rohrer). Also NARS, an AI system based on non-axiomatic reasoning (Wang), and a proposed system based on the idea of “emergent inference” (Pissanetzky).
Videos, Papers and Resources
IIIM offers full papers and extra materials for downloading.