Building an AI
From Sciphi
Contents |
Interaction
- Speech synthesis: MBROLA Project - http://tcts.fpms.ac.be/synthesis/
- Speech recognition: CMU Sphinx - http://sourceforge.net/projects/cmusphinx
- Natural Language Processing (ontology database): opencyc, semantic media wiki - http://www.cyc.com/cyc/opencyc/overview http://semantic-mediawiki.org/wiki/Semantic_MediaWiki
- Vision Processing (hierarchical temporal memory): http://www.numenta.com/
- spacial self-modeling
- http://www.opencog.org/wiki/The_Open_Cognition_Project
Problem Solving
- Neural Nets
- Pattern Recognition (Hierarchical Temporal Memory)
- Evolutionary Software: 3d, patched into visual system so the AI can give its imagination a helping hand sometimes
- Research (internet), access to its own ontological database would be memory not research.
- Requires ability to estimate cpu time required and to switch between methods if its more efficient.
New idea
Forgetting useless information is going to be important
| Long term memory (hard drive) | Short Term Memory (RAM) | |
|---|---|---|
| High Value | Store until it hasn't been found to be useful for a very long time, then move to low value / long term | Store until it has been made use of or the associated problem has been abandoned. If accessed many times through out the day move then move to low value / long term |
| Low Value | If accessed many times through out the day move then move to high value / long term. Store until it hasn't been found to be useful for a medium long time, then move to high value / short term term to see if it should be considered. (this may be akin to dreaming) | Forget almost immediately |
solution: Sleep mode - a daily period of time when the AI all the data gathered from the day and compares it to previous totality, of information. Will need to start with some axioms (rough draft as opposed to a blank slate). the process of nightly re-evaluation of data could be similar to dreaming.
other
- freindly AI?
- requires some sort of reward and punishment system to motivate learning
- possibly basic act utilitarianism ethical system.
- any such system should be tied into the Evolutionary Software and other problem solving software so it cant dream up unethical actions accidentally.
systems integration: http://mindmakers.org/
