A better idea would be to stop calling those existing systems “AI”, because we already have a definition for that and nothing comes close. At best what we have now are keyword-context lookup bots. A little voice recognition on the front and a search engine on the back-end (or in “the cloud” if you prefer).
Even the IBM Watson system isn’t much beyond that. It can work on a number of contexts and categorize possible answers with some pretty sophisticated algorithms, including assigning probability scores to candidate answers. It’s often referred to as a “machine learning” system, but it’s not even that. The “learning” is really just using a cache of answers and evaluating them later with refined probabilities based on new inputs. Similar to Bayesian probability methods created by Thomas Bayes in the 18th century, and that has been used for things like spam filters for many years.
Silicon Valley douchebags like to throw around terms like “AI” because it sounds advanced and cool to the VC companies they’re trying to impress for the next round of funding for their world-changing new smart phone app. It’s abuse of the term, and we should call them out on it, not decide we’re going to need a new term because it’s been so widely misused.