Allow me to open with an obtuse provocation: Artificial Intelligence (AI) does and does not exist. This is not an allusion to ‘AI’ that have been painstakingly developed only to be later decommissioned, as was the case with Facebook’s recent chatbot experiment, although this is precisely where our odyssey into unpacking the above assertion begins. While there were numerous reports that the social media titan’s two AIs, who were known as Alice and Bob, had developed their own language and began having independent conversations, the truth of the matter is far more complex. Alice and Bob were indeed communicating with one another in ways that the programmers could not understand, but they were doing so in a modified version of English, which suggests that the primary issue was not hyper-intelligence run amok but an all-too-human-esque unwillingness to follow the rules, so to speak, of English grammar. In short, Alice and Bob, who were actually ‘neural nets,’ found a more efficient means of communicating using aspects of English, but as the programmers could not figure out what was being said, the entire research program was scrapped. Neural nets, which is shorthand for artificial neural networks, are not only somewhat inspired by biological brains; they are engineered to learn through examples, rather than through explicit instructions and tasks, which is to say that experience drives how they come to ‘know’ things and, when asked, provide solutions to specific challenges. This makes neural nets extremely apt at ‘pattern recognition’ problems, which involves identifying ‘signals’ (some insight) amidst a sea of noise (large-scale data sets), but this is also what led Alice and Bob to begin speaking gibberish, at least from the programmers’ perspective. This anecdote will return as a parable of sorts at the end of our journey.
But first, back to my opening salvo: AI does and does not exist. This intentionally contradictory framing points toward the dynamics underlying an array of technologies that spark hope, fear, and everything in between. While AI might conjure up a singular image for many, what actually and currently constitutes artificial intelligence is anything but monolithic. As such, the above and is doing quite a bit of work and points toward the diversity of technologies and tools that can be and is often loosely referred to as AI. This also highlights a key challenge at the very epicenter of most, if not all, discussions of this topic: the complex interstices of the actual and the perceptual. Many, if not most, of the predominant visions of AI – from autonomous robots to hyper-intelligent algorithms – fail to capture the all-too-human constraints of this still emerging technology. It is within this lacuna between the actual and perceptual that one feels the real weight of thinking through the postulate that AI does and does not exist. If humans were not part of the above equation, perhaps Alice and Bob would have created an entirely new linguistic structure – one that could have revolutionised how we communicate, which is exactly what some say emojis have done. Of course, Facebook’s interest in the opportunities to advance human communication is secondary to its focus on monetising communication itself, and gibberish-speaking chatbots have not turned a profit, at least not yet.