True Artificial Intelligence or a simple system of probabilities
When we analyze the so-called “Artificial Intelligence” systems, specifically those of image recognition, we actually find a probabilistic system that is based on the comparison of the image with a data set produced from a model that reduces the resolution of the images as if it were the vision of a person with visual abilities diminished to 60%, where he sees shadows. But those few shadows help the system to identify in a 90–99% many of the objects that it tries to classify.
This allows a quick algebraic comparison based on simple operations and through activation functions results in a percentage of similarity between the object displayed and the data bank pre-classified into categories, that is, a probability expressed in a percentage of approximation.
If we observe how our human vision works, our eyes are similar to 80-megapixel cameras and 86 billion neurons that are capable of storing thousands of megabytes of data and have a processing capacity comparable to several teraflops. With these characteristics, our eyes can see the details of the images and the brain can keep those details.
For practical purposes, when the autonomous vehicle drives, they do not need to see all those details. But sometimes, it is extremely useful to see the details.
It is not the same to see an armed person in the middle of the street pointing directly towards the vehicle than to see a person who has stopped for some inconvenience in the middle of the street. They are two totally different situations that the autonomous system should clearly understand.
But you do not need to go to extremes like this. If we think of the mirrors in the corners of some Italian cities where it is difficult to see the sides since the streets do not have sidewalks we realize that neither the radar, nor the lidar, nor the cameras can determine if another car comes in the crossing or a pedestrian. Only small mirrors can indicate the existence of these objects. The current technology, with 12-megapixel cameras for artificial vision, can hardly see anything in the mirrors. It can not provide any sign of pedestrians, bicycles or another vehicle crossing.
All these challenges are faced by the development of autonomous vehicle technology, challenges that we will have to pay attention to classify our vehicle really like a robot that uses true artificial intelligence.
I have been crossing thoughts with other technology developers and many of us understand that either we are using probabilistic algorithms or, if we call this artificial intelligence, really our own intelligence is made up of probabilistic algorithms since we make decisions based on our own experience throughout life. A simple system of probabilities that also includes the classification of good and bad, beautiful and ugly, for each of us.