The term “Artificial Intelligence” transitioned from Science Fiction to Product Marketing with very little time in-between to consider what exactly qualifies as “AI”. While tackling this definition has been done ad nauseam for hundreds of years, I believe that the modern definition of “AI” and those from Science Fiction are worlds apart. This is my attempt to bridge that gap.
Fundamentally, I believe that “Artificial Intelligence” is decided by the observer, not the creator of the entity being observed. You can not declare that you have built AI – others have to verify that claim through interaction. Corporations wishing to sell their products have conflated “Machine Learning” and “Artificial Intelligence” to such an extreme degree that we are now lowering our expectations for what a AI actually is.
My definition of “Artificial Intelligence” is based firmly on the Turing test, as the Turing test matches Science Fiction’s definitions, and I believe that we should defer to the futurists who envisioned AI and not the capitalists who benefit from selling a poor imitation of AI to gullible customers.
The Turing test is stated as follows:
A test of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human
This to me is as simple and direct a definition of “Artificial Intelligence” as one could ever hope for. Unsurprisingly, it is a definition no corporation wishes to use.
With this definition of AI in mind, how do our contemporary consumer-facing “Artificial Intelligence” fair? Has anyone, anywhere even been even remotely confused that Siri, Cortana, or Alexa were anything more than voice interfaces into a software system? Describing these toys of convenience and entertainment as AI is tantamount to defining an airplane as a space exploration vehicle.
What of Deep Blue, Watson, AlphaGo, and the like? Certainly they have a far more legitimate claim to being Artificial Intelligence than the collective of voice assistants? Still, how long would an adult of reasonable intelligence have to interrogate any of these systems before they learned that they were not a human? Certainly minutes, and not hours. That’s not nearly good enough to claim to be an “Artificial Intelligence”.
But what of secret government projects where limitless resources are poured into simulating real humans for any number of nefarious reasons? Speculating on this is no different than Science Fiction, so it is best to reserve judgement until such a thing reveals itself to the public.
There is a sad complication in this simple challenge of “exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.” Humans today are now dismal at communication, and that communication rarely if ever has a modicum of intelligence. The communication-intelligence bar is now so low that a chatbot that identifies keywords and respond with a handful of emojis can fool most adults for quite a while. If and when it does, is it now artificial intelligence? To the person being fooled, yes – I claim it is. Whether of not you think something is intelligent is not based on how it was created, but by how many people and for what duration are fooled into thinking that it is.
Tragically, our lack of communication skills has tipped the balance in the Turing test so heavily in the machines favor that there is no need for sophisticated computer models. We have lowered the bar of what human communication is to such an extent that even a rudimentary script could fool us. The mistake of Siri, Cortana, and Alexa is attempting to emulate a college-educated professional rather than a blackout-drunk spring breaker flunking out of freshman English. Simplify the language, lower the bar of the Turing test, and let our low-attention span emoji-trading LOL/ROFL/LMAO society do the rest. Stop focusing on academic definitions of Artificial Intelligence, and focus on fooling people. When it comes to AI, that is the definition of success: fooling a human. A pity that today we are so easy to fool.