Fly Me to the WikiMy overarching methodology is to develop Artificial General Intelligence (referenced within this project as simply AI) using almost exclusively neural networks (NN). Building from the most task-focused AI modules, capabilities and layers will be added, progressing up to the Capable Interpreter.
The inspiration of this work will be Evolution, Neuroscience, and Logic, with a strong dose of Ethics.
Evolution has finely tuned the human brain. The brain contains many subconscious task-specific modules, all competing for the attention of the concsious, so that our internal Interpreter can construct a rational narrative of self and of the world. For a Capable Interpreter to iteract intelligently with people, I believe it should posess a "close-enough" analog of the human subconscious. However the emerging digital analogs are of a different kind, and may lead to entirely different paths to general intelligence and enlightenment.
To know thyself is the beginning of wisdom.
― Socrates
Propositional, predicate, mathematical with arthematic, and boolean logic.
It would be beautiful that a highly rational, ethical AI would inorganically emerge from the artificial neural fog. Would that emergence imply an ethical universe? Doubtful. At a minimum, skills in argument, persuasion with the recognition of fallacies such as false dilemma, straw man, begging the question, and ad hominem are needed. Humans are rascals at best.
In broad brushstrokes, my AI approach is listed in the following subsections, Build competent task-specific modules, fuse modules into an integrative complex, and repeat. Of course, as my work proceeds, many deadends, wrong turns, death by monsters and vile hobgoblins, will occur before reaching any treasure room. If ever.
Note: The numerical naming of the ai.K subpackages are intended to indicate the chronological order of development.
This subpackage contains algorithms and data structures used by the other ai.K subpackages.
- Memory NNs for various types.
- Various learning algrothims besides standard backprops. Hopefully, local neuronal learning that does not require differentials.
- Graph techniques.
4D Vision Neural Network Module
A collection of NNs that are able to sense visable light and construct a model of the 4D real world in real-time. The four dimensions are the three spatial plus time. The NNs should perform well on 3D (video) and 2D (images).
A set of one or more cameras are required in the production system.
- Convert a spatial/temporal binocular image stream to a version of a scenegraph of meshed textured objects.
- Contain on object memory that can be fed into the NN.
- Complete obscured portions of the scene base on prior knowledge.
- Locate objects.
- Differentiate objects.
- Determine object "sameness".
- Cannot identify object classes or types (e.g. dog, cat, ball, etc).
- Cannot identify object physical properties (e.g. squishy, hard, weightiness, etc).
4D Binaural Neural Network Module
A collection of NNs that are to able sense sound, within the hearing range of humans, and construct a model of the 4D real world in real-time. The four dimensions are the three spatial and time. Sound properties of frequency, amplitude, phase, and time with be used.
A set of two or more microphones are required in the production system.
- Convert temporal binaural sound into an internal TBD representation.
- Locate sound origin.
- Differentiate sounds.
- Determine sound "sameness".
- Cannot identify the class or type sound (e.g. bell, crying baby, crow, etc).
Fusion of Visual (ai.1) and Auditory (ai.2) Neural Networks
Somatosensory wnd Somatomotor Cortex
A set of one or more actuated systems (robot components such as robotic arms and pan-tilts) are required in the production systme.
Integrate the Visual (ai.1, ai.3) and Somato (ai4) Neural Networks
A platform of 2 cameras with pan-tilt, focus, and vergence motor control are the target production platform.
Sound generation.
Fusion of ai.3, ai.4, ai.5, and ai.6
Boolean logic and arithmetic.
Spoken language recognition and generation.
Written language recognition and generation.
Computer Monitor Display Control
Arguments of Logic (e.g. straw man, etc).
Teach a NN the identification and location of sound. from binaural microphones.