Part 1 : The Approach required to create real Artificial Intelligence

1.1.  Introduction

A biologically plausible approach that has the potential to create 'generic' Artificial Intelligence is described in this paper.

This paper describes how intelligent behaviour can be achieved by:

  • incorporating timed delay in neural signalling
  • encoding stimuli in the form of weighted neural connections
  • creating neural circuits for specific functions
  • enabling automatic creation of timed-associations between different types of stimuli
  • enabling embodiment, motor-actions and autonomous subsystems in the creation of an Artificial Life Form.

A simulation of an 'Artificial Life Form' in a virtual environment is illustrated, wherein its behaviour is controlled purely by its 'brain' composed of interconnected neurons.

1.2.   Approach required to create 'real' Artificial Intelligence

“I often refer to biology as setting a set of “constraints”. At first, the constraints make a problem more difficult to solve. Many machine learning and AI people avoid studying the anatomy of the brain for this very reason. But if you keep digging deeper into the biology eventually a solution emerges, and when a solution appears that satisfies the constraints you know you have the correct answer.”
                        - Jeff Hawkins (Numenta) on building a biologically inspired AI

It is widely accepted that contemporary techniques such as backpropagation will not lead to Artificial General Intelligence.[1] Any algorithm that attempts to create ‘generic’ artificial intelligence should be based on what we know about the human brain and should satisfy known biological constraints.

The following facts should be taken into account when formulating an algorithm for generic intelligence:

  • Information is stored in the brain in the form of weighted connections between neurons.
  • The connection strength between neurons may strengthen or weaken, enabling 'remembering' or 'forgetting'.
  • Associations are formed between different types of stimuli by means of neural pathways connecting different brain regions.
  • Any AGI algorithm should be able to explain empirically noted phenomena such as synaptic plasticity, sensory integration, false memories, deja vu, synesthesia, blindsight, phantom limb, etc.