Thread by @uras_salvatore

@svetlyak40wt Ok, I sent you an invitation on github on the backend project, that contains under neuron/ what little I have implemented. @spacebat: could do the same if you let me know a github id.
Following: a short, yet long, discussion of the concept.

@svetlyak40wt @spacebat THREAD
The concept: living beings have a nervous system, and "real" intelligence. I think nature is complex, of a chaotic, fractal kind of complexity; still I think it's, well, natural: by this I mean that no natural phenomenon, including intelligence, ...

@svetlyak40wt @spacebat ...needs to be explained by sovrannatural causes. Of some phenomena we have a complete mathematical understanding (or so we think: eg the laws of motion); some other we can't "integrate", but we can approximate them to an extent with numerical models (think of weather forecasts).

@svetlyak40wt @spacebat But things can be mathematically defined and NEVER extrapolable: think of the Mandelbrot set.
Intelligence and learning are incredibly complex phenomena. But so is life, yet the Miller-Urey experiment demonstrated that natural forces can spontaneously...

@svetlyak40wt @spacebat ...produce the biochemical elements of life.
A human nervous system consists of a number of neurons in the order of 10^11. Each neuron can have up to several thousands of connections: the number of possible combinations is astronomical. The information isn't stored...

@svetlyak40wt @spacebat ...in neurons themselves, but in the structure of the network: a neuron isn't a "drawer" that contains memories, but a node of a network. That's why a stroke destroys our learned abilities, and simply growing up new neurons won't solve the problem: those abilities were learned...

@svetlyak40wt @spacebat ...with an evolutionary process.
The little I know about this process is (deeply semplifying) that synapses are selected by a somewhat darwinian process or random creation and decay, or survival, according to a feedback. So what if we could recreate in a simulation...

@svetlyak40wt @spacebat ...the very basic properties of a network of neurons, including input, output and a feedback system? Would such a system, in due time, show a trace of "real" learning?
So far the elements of the theory.
The implementation (what little there is of it) is in Common Lisp. ...

@svetlyak40wt @spacebat Neurons are excitable objects, with a baseline potential and a threshold - when they receive one - or more - exciting signal, such signals sum, adding to the neuron's potential for a short time span, after which the excited state subsides and the neuron sontaneously returns to...

@svetlyak40wt @spacebat ...its base state. Since the process must be asynchronous, a thread is used for each neuron. This probably is a severe limitation if the simulation runs on ordinary hardware, and I don't know enough about GPGPU to tell if this could be a good use case.

@svetlyak40wt @spacebat If a neuron's potential exceeds its threshold, the neuron "discharges", ie generates an exciting potential that travels through its outbound synapses to reach connected neurons. A neuron can have inbound synapses from many other neurons, ...

@svetlyak40wt @spacebat ...so the interactions are incredibly complex.
This part is implemented, tested, and works.
What I haven't implemented (and I don't know whether and when I will):
...

@svetlyak40wt @spacebat * a specialization of neurons into a "receptor" class, that receives its input potentials by sampling and translating external events (could be hooked to static data, like a file; could be connected to a peripheral like a microphone or a camera)
...

@svetlyak40wt @spacebat * another specialization into an "effector" class, that translates action potentials into data or, in a future, behavior of a peripheral (think of a simple robot)
* a feedback system on generated output
...

@svetlyak40wt @spacebat * a system of random decay and generation of synapses, reinforced and directed by the feedback system.

Given these elements, my little first Miller-Urey experiment on learning would be to give the simulation a simple static input (say a barcode?), ...

@svetlyak40wt @spacebat ...a "desired" output for the feedback system, then let it run and see what happens.
Should it prove to work, endless possibilities for experimentation and practic application would open.

@svetlyak40wt @spacebat As you can see from the code, I am very far away from realizing the experiment. Bear with me - in the meantime I had a daughter and I lost her, I went through a very unfair and traumatic dismissal, I discovered I am autistic... Plenty to deal with. Thanks for your kind interest.

@svetlyak40wt @spacebat @threadreaderapp : please unroll