Abstract
A program evolution component is proposed for integrative artificial general intelligence. The system’s deployment is intended to be comparable, on Marr’s level of computational theory, to evolutionary mechanisms in human thought. The challenges of program evolution are described, along with the requirements for a program evolution system to be competent- solving hard problems quickly, accurately, and reliably. Meta-optimizing semantic evolutionary search (MOSES) is proposed to fulfill these requirements.
Co-authored with Moshe Looks (First author)
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
First because this will increase the number of samples needed to effectively model the structure of knob-space, and second because this modeling will typically be quadratic with the number of knobs, at least for the BOA or hBOA.
- 2.
A term borrowed from biology, referring to a somewhat isolated local population of a species.
- 3.
Another term borrowed from biology, referring to a group of somewhat separate populations (the demes) that nonetheless interact.
- 4.
This formulation is equivalent to using a general three-argument if-then-else statement with a predicate as the first argument, as there is only a single predicate (food-ahead) for the ant problem.
- 5.
That there is some fixed ordering on the knobs is important, so that two rotation knobs are not placed next to each other (as this would introduce redundancy). In this case, the precise ordering chosen (rotation, conditional, movement) does not appear to be critical.
- 6.
MOSES reduces the exemplar program to normal form before constructing the representation; in this particular case however, no transformations are needed. Similarly, in general neighborhood reduction would be used to eliminate any extraneous knobs (based on domain-specific heuristics). For the ant domain however no such reductions are necessary.
- 7.
The fact that reduction to normal tends to reduce the problem size is another synergy between it and the application of probabilistic model-building.
- 8.
There is in fact even more information available in the hBOA models concerning hierarchy and direction of dependence, but this is difficult to analyze.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
Copyright information
© 2014 Atlantis Press and the authors
About this chapter
Cite this chapter
Goertzel, B., Pennachin, C., Geisweiller, N. (2014). Probabilistic Evolutionary Procedure Learning. In: Engineering General Intelligence, Part 2. Atlantis Thinking Machines, vol 6. Atlantis Press, Paris. https://doi.org/10.2991/978-94-6239-030-0_15
Download citation
DOI: https://doi.org/10.2991/978-94-6239-030-0_15
Published:
Publisher Name: Atlantis Press, Paris
Print ISBN: 978-94-6239-029-4
Online ISBN: 978-94-6239-030-0
eBook Packages: Computer ScienceComputer Science (R0)