Artificial Connectomes

 

From the movie Young Frankenstein

               

"Abbey Normal" is of course the perfect scenario of what we do not want to do. To illustrate further, I would start the C elegans emulation and let the robot wander around the house, sometimes for an hour or so. Just let it go. I liken it much like a Cat in that it would explore and do its own thing. One day my wife started calling it and the robot stated to come towards her. To her, calling it made it come to her but to me, I knew what was really happening in that she was activating the sound sensor which in turn was stimulating the food sensory neurons so the robot thought she was food. Not sure this would be the desired effects we would want to occur. Other animal connectomes will most likely give us a number of undesired effects. The obvious consequence to connectomic emulation in the movies around Frankenstein is that if we use a criminal brain, we end up with a criminal. To understand how connectomics work and to find out if we could wire a nervous system to give us mostly desired outcomes, we are turning to artificial connectomes.

Our brains are devoted to a lot of homeostatic functions. Obviously breathing, eating, sex, drinking, defecating, etc are not functions we need to put into a robotic brain per se. It could be said that some of these functions are needed but in a different form (e.g. recharging batteries could be very similar to eating) but the multitude of functions an animal brain has just for the animal to continue to be alive are not something we would have to recreate in a robotic facsimile of an animal.

However, at this time, these functions are so intertwined in our nervous systems, that trying to remove them from an animal connectome is truly impossible. One alternative is just to ignore these neurons that may activate but since they are not required or used in our robot, we never derive anything from it. The issue with this, is that you are expending resources unnecessarily which I will talk about in much greater detail later.

To create simple connectomes that are equitable to something akin to C elegans, we can use the Watts-Strogatz and its derivative small world models to build our artificial connectomes. 

We can also combine these models into functional areas to build larger brains and complexities:

However, even simple models miss calculated circuits that evolution has created. For example, in C elegans the motor circuit that moves the animal from right to left and forward/backward has been well mapped out.

University of Massachusetts Medical School

Simply plugging in numbers to generate a connectomic small world model, will not create these circuits. So even an artificial connectome must have features added to include these evolutionary developments; otherwise, we just end up with an artificial connectome that does nothing useful or purposeful. 

In conclusion, in order to develop Artificial General Intelligence (AGI), we have a solid model with the neural circuits nature provides but there is much to this model that is not needed and creates complexity in emulations that are very difficult to overcome. On the other hand, just creating our own artificial connectome makes for a nervous system that is very mundane and useless. The key is to use animal connectomes to create artificial connectomes; i.e. use nature's model of a given animal and then develop the necessary circuits to recreate the general intelligence we desire. This is much easier said than done!