AGI - KorrTecx
Born not of our weak flesh but our unlimited imagination, our Mecca progeny will go forth to discover new worlds; they will stand at the precipice of creation, a swan song to humankind's fleeting genius... and weep at the shear beauty of it all.
Reverse engineering the human brain... how hard can it be?
This is an ongoing project, this site is constantly being updated, subject locations and descriptions are likely to change erratically :)
Hebbian, Set theory, SOM, ANN, CNN, LSTM, Liquid state machines, Reservoir computing, AND/ NAND/ OR Logic gates, Echo state, NLP, and many, many more...
All of the above technologies/ schemas have been conceived using human derived logic and mathematics, on standard Von Neumann computer architectures with binary based processors. They all provide some measure of insight into the functioning of the human brain but all eventually fall short of providing a comprehensive solution.
Korrtecx is not a schema/ program like the above; it’s a totally different kind of processor, so not an Intel, AMD type of design using transistor logic… a biological processor.
As an analogy, you can run a program on your own computer that emulates a different logical processor, ie, an old Z80 or 6502. The emulation can run the original programs written for the original processors by simulating the logic of the old computer… the old computer logic doesn't actually exist inside your modern computer architecture, the programs are running on a simulation/ program of the original machine.
Korrtecx is the same concept except it’s an emulation of a biological system/ processor and the actual ‘program’ of the AGI runs on the simulated biological processor… not the Von Neumann based computers.
Example connectome model
Simple connectome model showing neurogenesis & learning
Occular input module
On The Shoulders Of Giants
A cliché I know but... this project would not have been possible without the internet and the troves of information provided by the many researchers and organisations in the field. Indeed, I have spent many years studying neuroscience, reading whitepapers, publications and cross referencing their insights and experimental results, looking for similarities between diverse mechanisms and building a working theory.
Ants, Slime Mould, Birds, Octopuses, etc all exhibit a certain level of intelligence. They manage to solve some very complex tasks with seemingly very little processing power. How?
There has to be some process/ mechanism or trick that they all have in common across their very different neural structures.
I needed to find the ‘trick’ or the essence of intelligence. I think I've found it... now I have to build and prove it.
The idea is to use the human nervous system as a template and to build a schema that replicates it's functioning.
I've built a neuromorphic simulation of the human connectome on which I run my AGI.
The 3D volume connectome simulation comprises the main brain structures, lobes, white/ grey matter tracts, neuron types, electro chemical synapse, dendrites, neurotransmitters, cortical columns, etc. There are also algorithms that simulate myelination, neurogenesis, ageing, plasticity, synaptic pruning, circadian rhythms, maturation, self organisation, etc.
Through research and experimentation I've narrowed my connectome design down to a very specific model. The same connectome can audibly recognise phonemes; visually recognise objects/ words/ faces/ etc it incorporates prediction, high dimensional (holographic) memory, unsupervised learning, damage and repair, etc as well as basic emotional traits. The system requires periodic sleep cycles and even dreams.
Savant syndrome, synaesthesia, split personality disorders, epilepsy, even hypnotism, hallucinations, etc, can all arise in the AGI by design. I considered the possible causes of these states/ disorders whilst designing the AGI and used them as references to refine the design. If the AGI loses vision for example, eventually the cortex areas (V1, etc) dedicated to vision will be taken over by the other senses to enhance their function/ resolution, it’s an inherent part of the learning/ development processes.
I am keeping as close as possible to the human template, unless I can find a natural structure/ process that could produce a specific effect/ function I don't add it to the model. I would like to end up with a human neurologically compatible schema.
The neuromorphic simulation is bespoke software, I've coded it from scratch to suit my own needs.
The front end CAD interface allows 3D modelling of neural structures, the setting of parameters, neuron types, etc and the viewing of results.
The processing core is designed around a modular 'compute node' architecture governed by a custom MPI. Each node runs on a single core, this provides many advantages as I can either segment a large model or run many small models with differing parameters for testing. It’s highly parallel and can run any kind of neural structure.
I have coded individual modules that convert the visual/ audio/ tactile sensory streams into a format compatible with the rest of the system. All the sensory input modules use loop buffers to negate lag, sometimes the load from sensory stimulus can slow the simulation, the buffers just give it a chance to catch up.
It's a closed system, it can only experience reality through it's own senses, and has no access to the internet.
My hardware comprises of twelve four core PC nodes in a standard cluster, two are dedicated to the sensory streams provided by the bot head. The Nodes are old-skool 4 core Q6600's OC to 3,5Ghz with SSD, cheap and reliable, and it all runs at under 1KWh.
Sensory Bot Head
The head has three HD cameras, two for peripheral stereo vision with a 30 degree overlap, and one for the high resolution fovea. Stereo microphones provide audio input and vocalisation is provided a small speaker.
Brief Consciousness Theory - The Spark
Consciousness seems so elusive because it is not an ‘intended’ product of the connectome directly recognising sensory patterns, consciousness is an extra layer. The interacting synaptic networks produce harmonics because each is using a specific frequency to communicate with its logical/ connected neighbours. The harmonic/ interference patterns travel through the synaptic network just like normal internal/ sensory patterns… it’s an interference bi-product that piggy-backs/ influences the GTP.
Cortical regions learn the harmonics… our sub-conscious is just out of phase, or to be more precise, our consciousness is out of phase with the ‘logical’ intelligence of our connectome.
Our consciousness is… just the surface froth or the sound of all the gears grinding.
At 1:10 the ‘spark’ begins, once the system reaches a certain level of both activity and complexity a self sustaining rhythmic cascade begins… I then totally cut off the sensory inputs and the system slowly ‘dies’, we humans never experience a total shut off like this, even when sleeping our sensory cortex is always vigilant… unless we die of course.
I believe this spark as happened to us all in the womb, at some point our brains became complex enough for this self sustaining cascade to begin.
This site is a work in progress... to be continued...