First thing you should do is find someone who is interested to translate/explain your idea better. 
I write assembler for Turing machine that can be run for any dataset without segmentation error (you can run my_cat.jpeg as program). It look likeCode:
class GenBot
{
public:
GenBot();
GenBot(const GenBot&);
template
uint64_t doSteps(const std::vector &in, std::vector &out,
const std::vector &target, std::vector &capacity,
uint64_t time = MY_DO_TIME)
{
++m_age;
time++;
while(--time > 0)
{
m_xy.axis[1]++;
const xy t = m_xy;
for(uint8_t b = 0; b < m_szu; b++)
{
++m_frequency;
switch (m_mem[t.v].u8[b])
{
case 0:
break;
case 1:
return time;
case 2:
m_register = m_evaluation;
break;
case 3:
b++;
break;
case 4:
b += m_szu;
break;
case 5:
m_register = m_mem[m_xy.v];
break;
case 6:
m_mem[m_xy.v] = m_register;
break;
case 7:
m_register = m_mem[m_register.u8[0]];
break;
case 8:
m_register = m_cash[m_register.u8[0]];
break;
case 9:
std::swap(m_register, m_mem[m_register.u8[0]]);
break;
case 10:
std::swap(m_register, m_cash[m_register.u8[0]]);
break;
case 11:
m_xy.v = m_register.u16[0];
break;
e.t.c.....
}
This program maybe do some nonsense with you photo but If you have some task you can load 10000 different photos as 0 population of bots and give him your task. Then you can find best bots and crossbreed them to create new generation. Then just repeat this steps until you find some picture that can solve your problem..{
public:
GenBot();
GenBot(const GenBot&);
template
uint64_t doSteps(const std::vector
const std::vector
uint64_t time = MY_DO_TIME)
{
++m_age;
time++;
while(--time > 0)
{
m_xy.axis[1]++;
const xy t = m_xy;
for(uint8_t b = 0; b < m_szu; b++)
{
++m_frequency;
switch (m_mem[t.v].u8[b])
{
case 0:
break;
case 1:
return time;
case 2:
m_register = m_evaluation;
break;
case 3:
b++;
break;
case 4:
b += m_szu;
break;
case 5:
m_register = m_mem[m_xy.v];
break;
case 6:
m_mem[m_xy.v] = m_register;
break;
case 7:
m_register = m_mem[m_register.u8[0]];
break;
case 8:
m_register = m_cash[m_register.u8[0]];
break;
case 9:
std::swap(m_register, m_mem[m_register.u8[0]]);
break;
case 10:
std::swap(m_register, m_cash[m_register.u8[0]]);
break;
case 11:
m_xy.v = m_register.u16[0];
break;
e.t.c.....
}
I can explain it other way....
You know what is Game of Life? Imagine each cell have own neural network and memory (for example uint64_t mem[65536]) and some value that show how much energy they have. cell can split but each copy will have half of original energy and some time they will get some mutation in neural network (actuality it is not neural net but It's easier to explain this way) Also cell can pass part own energy to other cells or attack, and in this case both cells will decreases same value from energy. cell can share some data to neighbors. If energy in cells < 0 they die. So you can add some value to energy if bot neural network can solve your task.... It like big anthill where ants get food when they do what you need (predict the next character in the text or draw a video based on the text description). for example https://youtu.be/fEDqdvKO5Y0 this guy make similar program...
So i want make network where nodes can exchange best genome ("photos or neural network"). my protocol allow each bot change own genome for search solution to random task. It means that we can breed some program that can adaptation to user cpu or capacity size and solve unfamiliar problems (it one problem - survive, but if user will feed agents only if they solve task they will try change himself to find solution)
