09/27/2016 at 14:22 •
Current artificial intelligence doomsdayers assume that massive mathematical machines will digest billions of points of data and become functional sentient units creating a singularity and ushering in the end of existence as we know it for us fleshlings. I beg to differ. I'm no AI expert, I could not for the life of me tell you what a neural network is, but I fail to see how probability formulas can link completely unrelated data and give meaning to it. Apart from being an incredible pattern matching and abstraction machine we as humans give meaning to our experiences through our senses and our perception of space and time. The brain digests this sensory data and develops semantic links and execution routines in relation to this massive influx of data that coincide at a certain instance in time. As we experience events and phenomena that stimulate our senses we develop relevant execution routines which we store in memory. The more we experience, the more we fine tune these execution routines or state machines and the more “knowledgeable” or experienced we become for a certain set of input stimuli that occur at an instance in time.
Take for instance walking on the sidewalk in a New York winter. You suddenly come upon a patch of sidewalk that's highly reflective. Your memory kicks in… the last time you waltzed nonchalantly unto that highly reflective patch on the sidewalk your derriere learned an object lesson through negative reinforcement. Your developed “slippery ice avoidance” state machine kicks in and you walk cautiously around it crunching through the mound of shoveled snow pile on the curb. You're now back to your regular walking to work, or wherever you intended to go, state routine. The example shows the linking of completely unrelated phenomena to create an execution routine. Walking, a reflective slick looking surface, cold temperature. How the brain abstracts and stores this data, maps it together and associates it with a learned execution routine and also fine tunes that routine to react more efficiently is a critical part of intelligence.
Mathematical image processing can do image segmentation, pattern grouping and frame to frame motion analysis but how do you store the action of walking after we have processed all the images, extracted all the necessary parameters and determined there is a moving human in the frame. Semantic linking can be used to create meaning. We can store relative motion of connected geometry with other unrelated data and call it walking. e.g. detected face, flesh tones, repeated swaying of related geometry at certain distribution in moving area of the frame, translational motion of entire figure, stepping sounds. The sensory input of pressure at regular intervals on your feet. All these can be used to determine there is a high probability there is a walking human.
In the previous example though, all this may not be needed as you would have a conscious knowledge that you decided to walk and your brain would be in a walking state, you wouldn't need to continuously observe yourself walking to know that. It does illustrate though that intelligence or consciousness does require multiple sensory inputs and linking them in an instance of time to form memory and execution routines to address them. We are reaction creatures and we learn by positive and negative reinforcement over time. If we are to create artificial intelligence of any sort it would have to include these principles.
If for some weird reason you found yourself reading this article share your thoughts and tell me what you think.