Through a Monitor, Darkly: Normalization & Machine Intelligence
Written by Ross A. McIntyre, VP of Strategy and Marc Boudria, VP of Artificial Intelligence at Hypergiant
You might not realize it, but, chances are, you interacted with a machine intelligence (MI) today. Maybe it was a direct engagement or perhaps it just facilitated a task you were completing. Nevertheless, in the contemporary world, MI’s are driving towards ubiquity (literally so in the case of autonomous automobiles). Yet, we are not typically explicitly informed when we are having a human-machine interaction; instead, many who author such software engage in a form of “Wizard of Oz” misdirection (Pay no attention to the machine behind the curtain). Sometimes, this is a simple omission, but many times it is a deliberate attempt to hoodwink the human and convince them that their interaction was, in fact, something that it wasn’t. That premeditated obfuscation is not congruent with the future as it creates an adversarial interaction insofar as the intention is to mislead. It is very difficult to progress a meaningful conversation based on chicanery or technological legerdemain. Consider: would you rather enter a situation where you don’t know whether the intelligence you are interacting with is there only to serve its masters, or one in which you are told, “Welcome. I am an AI here to go through this journey with you?”
Companies use a veritable bag of tricks to hide the synthetic nature of the bots we communicate with, including blanking the screen after each reply (to hide inconsistencies within the transcript) or employing various deflections to maintain the directionality of the conversation and ensure that it doesn’t veer off from the scripted topic. And encoding specific bias or idiosyncratic belligerence also goes a long way towards convincing someone they are interacting with a human. According to Venturebeat, 63% of humans are willing to communicate with brands via a machine intelligence; and, nearly half of all Millennials will accept recommendations made to them by bots. 75% of people want to know when they are communicating with a non-human; and, half of us find it “disturbing” when companies go to great lengths to hide that fact (e.g., Facebook).
Ask yourself, why are machine intelligent agents averse to communicating why you might be interested in a given product or service? This circumstance tends to be a requisite part of a parallel human interaction in which we come to understand the impetus or activation point for such advice. When a human salesperson utilizes their innate intuition coupled with their own awareness of similar past interactions in order to understand needs as quickly as possible, it leads them to the activation point via direct and indirect questioning. Even as a human refers to past congruent situations, she never assume that the current sale is going to be exactly the same as the last; while machine intelligences are often structured with this innate, erroneous assumption. It’s a one-size-fits-all approach that ultimately fits none.
So, how should a MI engage when the user doesn’t fall into a simple, predetermined bucket? Frequently, the human gets frustrated and tries to find another human to assist (barring very real situations where the user fury grows to the extent that she screams at the screen and rage quits). This cannot be the future of human-machine interaction. Machine Intelligence designers often lean on the Five Factor Model (FFM) that, in this context, posits that a multi-dimensional personality requires five trait dimensions: Extraversion, Agreeableness, Conscientiousness, Neuroticism/Emotional Stability and Intellect/Openness to Experience. But designing according to these elements is rarely sufficient to convince a user that a non-human intelligence is, in fact, human. And why should it? Is that actually desirable?
Humans have accepted hammers and saws as tools in their lives that exist to construct solutions to human problems (or even to construct new tools which can more effectively address those needs). Humans need to accept new tools for a new era. Moreover, companies must alter their thinking as to what a MI means to a human in their daily lives and in communicating with a brand. Would you use a hammer if it could only hit Google nails? Would you continue to use that hammer if, every time you swung it, it told Stanley tools what was being hammered? However you answer these questions, know that this reflects our current state of affairs in technology. As a physical object, the hammer comes with no “strings” or secret ancillary functions.
So, why the mummer’s farce? Is there any reason to hide the nature/origin of a communicator aside from the purposes of intent obfuscation? If the answer is only that it serves insincerity and does not exist to assist the user and satisfy her needs, it must be revisited. User-centric design is now de rigueur in software development to the extent that it hardly bears mentioning. However, when brands assume their needs supercede those of their customers, it should serve as a reminder. Good customer experience builds trust. Trust leads to affinity; and, affinity can lead to proselytization. Contemporary companies must be sure to weigh the value of honesty against the value of misdirection.