A few days ago I was reading about a social research project focused on possible impacts of artificial intelligence on our lives. The main point was that the computers utilizing artificial intelligence would be so clever, that they would find ways to make us follow their orders without us noticing it. I do not know what the data behind such research are, but it is completely the opposite of what we technicians observe.
The success of artificial intelligence today is centered around computers approximating the human perceptions of images, sounds and especially speech. It is mainly thanks to deep learning with its neural networks, which, however, do not perform any clever action. They are simply trained on large datasets of pictures and sounds transformed to mathematical tensors, which cannot detect patterns outside of their training. For instance, if the original dataset did not contain the image of a flower, the neural network would not then be able to classify the picture as the picture of a flower and would instead mark it as something else, even an elephant or a dog, it all depends on the original process of training.
Neural networks are, similarly to computer graphics, an applied multilinear algebra, that requires continuous differentiable functions to transform the representations from one vector space, like the picture, to another vector space, like the text classifying the objects present. In order to allow discrete functions to be used and model the process of thinking, not only perceiving what we know, automated programming such as genetic programming needs to be used, however, it is far from being optimal and fast. If automated programming could be combined with multilinear algebra, we would be talking about artificial general intelligence, that is an intelligence, which can at least partly think and react to changed conditions. If we add more clever reinforcement learning, the artificial general intelligence could actually understand its assignment.
It is, however, not the case today. Artificial intelligence today is a set of precisely defined algorithms used only for specific tasks and an advanced geometry that helps us to detect patterns similarly to our own human perception. The process of thinking, put aside the computers being able to outsmart our social brains and engage themselves in such things as control over them, is a very distant sound of the future. So far, artificial intelligence rather seems to augment our own thinking similarly to spectroscopes, telescopes or internet communication. It is a very useful game changing tool.