I recently participated in a podcast in which the concept of Artificial Intelligence came up. Basically, there are two fundamental categories of AI: Weak AI and Strong AI.
To dismiss the latter first, Strong AI exhibits all of the intellectual capabilities of a human being and does so at an exponentially faster speed and more accurately than any human is capable of. This all hinges on the (so far) handwavium of designing and building a machine that is capable of human-type imagination or cognitive leaps of intuition. The difficulty here is, we don't now how we humans do this, so the question then becomes: how do we build a machine that possesses a capability we can neither fully describe the function of nor replicate within ourselves? Unless and until we can fully describe and replicate the electro-chemical neurological processes that transpire in the human brain when we individually experience imagining a concept that is not contained within the disparate data sources we link together to inspire that intuitive leap, there simply isn't the possibility of including that capability into a device we construct.
Weak AI, however, I believe is essentially within our existing technological grasp. If you network 7 distinct Alexa units into what I term a "data orchestra", you can replicate the appearance of a Jarvis-like (from the first two Ironman films) entity. The primary unit is the Director which contains within its internal hard drive and memory the contents of a dictionary, thesaurus, and literary style guide. This unit coordinates the input from the remaining units. Each of the remaining six units has the entire contents of an encyclopedia downloaded to internal memory and one sixth of that in as great a level of detail as the data record allows. A sufficiently capable data search and retrieval system, with a robust prioritization function for assigning "relevance values" to disparate data returns, would permit the primary unit to conduct a "conversation" between the six sections of the orchestra and the inquiring human. By providing a range and depth of pre-programmed decision trees allowing the combined unit to semi-autonomously manage certain functions independent of direct human management, we begin to approach a functional emulation of what I regard as a true - if limited - Weak Artificial Intelligence.
Data security from external sources (particularly during the necessarily frequent content updates required to keep such a device current) is a critical matter to be addressed before such a device might be safely marketed, but that too is within current technology capabilities. No doubt there will be other concerns not considered here, but this does demonstrate what is possible today, I believe.