1) No product that has been called “AI” as of today is intelligent. 2) “AI” that is not machine learning will never achieve intelligence. 3) Deep learning may not be the tool to build intelligent systems.

Intelligence is the ability to generalize.

The initial goal for “AI” that was set when the term was coined in the 50s, was to achieve human like intelligence.

Today, “AI” is most often defined as a superset of machine learning. This means that a rule based algorithm doing something (not necessarily well) that is usually done by a human, is an “AI” (e.g. a rule based chat bot).

However no system to this day has been capable of real (i.e. global) generalization (vs local generalization). All machine learning systems are trained on specific tasks, and can do pretty well on similar task. For example GPT-3 can predict well the word that would most plausibly come after a sequence of words, and through that is able to generate some python code (=local generalization). But none of the systems is able to achieve something that is completely different than what it is trained for.

Humans are born with hardware that makes it good at learning stuff, and even, we suspect, allows them to already “know” stuff straight out of the womb. For a system to be intelligent, we may need a specific hardware, as of today the closest we have to this concept are innate priors embedded in the NN’s architecture.

An intelligent system will most certainly require data to be trained on, just like humans feed their brains with sensory stimuli, that is literally the definition for “machine learning”.