The achievements of today's Artificial Intelligence, while often impressive, do not arise from autonomous "mechanical thinking," but from the apparent or hidden utilization of derivatives of human intelligence, e.g., the indirect or direct use of internet content. This book therefore contrasts the functioning of Artificial Intelligence systems with that of the human and animal mind.

Its aim is to highlight mainly the key mechanisms that determine human cognition and to show which elements must be artificially reproduced for human-like cognition to emerge. While detailed imitation of biological functions is not needed to produce thought, the imitation of human cognition is exceedingly difficult, because it begins from sensory-motor mechanisms of action-reaction before reaching language (speech) and logical reasoning.

Such mechanisms define specific concepts as ways of interacting with the environment even before they acquire a name. The sensory-motor background of thought arises from the continuous interaction of each organism with its environment, that is, from the immediate synthesis of sensory stimuli with motor reactions. Such mechanisms are systems of immediate readjustment shaped by direct temporal interaction with the environment. This is why it is difficult to capture them verbally or symbolically. Thus, they are often unconscious.

Catching a ball, riding a bicycle, or even walking are automatic mechanisms and relate little to verbal instructions, because they do not reduce to stereotypical rules. Therefore, for a robot to learn them, it must interact temporally with its environment, and indeed in a manner analogous to that of humans.

Specifically, in order to learn interactively the language, it must have experiences as well as the motivation, the need to communicate with a caregiver-educator. There are vast areas of knowledge about the world (but also human society) that an Artificial Intelligence system cannot acquire without direct bodily experience. Thus, today's robots are only superficially autonomous and generally act only when guided. They are not capable of independently planning their actions.

Similarly, impressive language processing machines, like ChatGPT, produce understandable but sometimes irrelevant text, without knowing what they are saying. The words they produce do not correspond to a clear internal mental scene, as they are calculated word by word. In contrast, as we will see, a 19-month-old child, who has just learned their first words, knows what they want to say. They have an internal mental image, as they correct us if we misinterpret their words due to their still clumsy pronunciation.