Artificial Brains: An Evolved Neural Net Module Approach

This Chapter shows how the first author and his research team build artificial brains [9]. An artificial brain is defined to be a collection of interconnected neural network modules (10,000–50,000 of them), each of which is evolved quickly in special electronic programmable hardware, downloaded into a PC, and interconnected according to the designs of human ‘BAs’ (‘Brain Architects’). The neural signaling of the artificial brain (A-Brain) is performed by the PC in real time (defined to be 25 Hz per neuron). Such artificial brains can be used for many purposes, such as controlling the behaviors of autonomous robots.

The remaining contents of this Chapter are as follows. Section 2 places the current Chapter in context by describing related work. Section 3 provides an overview of how we evolved our neural network modules. Section 3.1 describes briefly the evolution tasks we used to calculate the speedup factor using the Celoxica board compared with a PC; Sect. 3.2 provides some details on how we performed the evolution on the Celoxica board; Sect. 3.3 gives a brief description of the ordinary genetic algorithm we used to perform the evolution. Section 4 briefly describes the characteristics of the Celoxica board. Section 5 presents the experimental results. Section 5.1 explains the so-called ‘IMSI’ (Inter Module Signaling Interface) – that is, the software used to allow modules to send and receive signals between each other. Section 5.2 answers the question “How Many Modules?” – in other words how many modules can an ordinary PC handle in an artificial brain?