AI Without Intelligence or Artificiality

Original article was published on Deep Learning on Medium

AI Without Intelligence or Artificiality

Actually, any tool that uses mathematical methods based on vast data processing is called Artificial Intelligence (AI). However, this AI is far from intelligent or artificial.

As the current development of AI goes on, will it be possible to reach a real AI that can be intelligent and artificial? This general question contains a conceptual problem derived from not knowing adequately what we mean by intelligence. For the present analysis of AI, I will take Kant’s work, especially the criticism of pure reason, to determine what is needed to consider that something can be considered intelligent. On the other hand, I will take Simondon to build a critique on the artificial in AI. Finally, it should be noted that at present I only superficially analyze the concept of AI and propose, therefore, a more exhaustive questioning of the same concept, which will allow for the construction and design of a better AI.

A Technical Object That Senses

Intelligence, as a concept, is one of the most diffuse things in existence and is always erroneously associated with logical and pragmatic thinking, directed towards problem solving. Nevertheless, for its definition, it is important to point out the processes that make it possible and that, therefore, without which intelligence cannot be reached.

As a first point I make an analogy between intelligence and reason, from Kant’s point of view, and which is nothing more than the mental processing by which we understand the world to a great extent, and question what is beyond it to a lesser extent; that is, the reasoning of a physical and metaphysical world. This type of intelligence is only proved in humans, while in animals only a knowledge of their physical world is intuited.

In Kant, precisely in his critique of pure reason (2009), he considers that reasoning is the synthesis in consciousness between experience and reasoning, being the experience obtained by the sensibility of things. But it is in the machine where we see problems with this type of synthesis.

Firstly because the sensitivity of the machine, that given by sensors, is limited to the experience of the human so that if it visually captures something, the data it obtains will only be those that the human experience of the developer made it possible for it to see, so its sensitivity is subordinated to human reasoning.

In the experience, we have the same case, since it can only experience what is designed for it, so a machine cannot have its own subjective experience of things.

These parasitic functions of the machine towards the human, the sensitivity and the experience, make that the machine cannot have synthesis and, therefore, cannot develop reasoning.

The Artificial in the Machine

The artificial is usually considered as that which is created by man, outside of any natural origin; however, for Simondon (2008), the artificial is that which is opposed to the natural insofar as it needs it to subsist. In other words, the opposite of a complete object, concrete he calls it, the artificial supposes an initial state of the evolution of the technical object where it still needs a human to be able to carry out the activities for which it was created. Without a human, the artificial technical object is in disuse.

Simondón’s concept of the artificial helps us to visualize the conceptual problem surrounding AI, which is still evolving and therefore still abstract.

Towards a New AI

Therefore, considering the concepts of intelligence and artificiality, it can be seen that the path currently taken by AI is undermined in its concepts and the reaches it seeks to reach.

One proposal would be not to consider machine learning or deep learning as artificial intelligence itself, but rather as computer concepts. This will help us to think of a properly derived intelligence for a technical object, one that is not analogous to a human, but rather proper to an evolving object.

This way of thinking and conceptualizing a new AI will not only help its development but also to better understand our intelligence beyond the topics of logical reasoning to which it is ascribed. Understanding, therefore, different types of intelligence that there are and will be.