Bots In The Wild
Besides Pokémon reclaiming its place in the zeitgeist, bots are the buzzing tech trend of 2016 so far. The big five — Google, Apple, Facebook, Amazon, and Microsoft — have all unveiled or featured bots in their annual product conferences. Consumer startup decks have hurriedly affixed bots to their pitches, for better or for worse. The New Yorker covered conversational bots in June. The New Yorker! Bot has quickly become as common as the word app has been in startup chatter about what people are building: its a bot for ‘X’ (in this case X usually stands for a mundane, predictable task currently carried out by humans). On one hand, bots are a tech craze like many before them. As our attention spans wane and the realization sets in that bots might not change the entire world tomorrow, the hype will begin to die down. The technology powering conversational bots — which are the type of bots driving this particular moment in craze — is still in its infancy and prone to bugs, errors and usability limitations. These errors are akin to why we press zero on our phones through every automated customer service selection tree ever until we reach a person. The bot doesn’t understand the problem, doesn’t have a trained solution, or it can’t hear/read what I’ve said. An even more fatal scenario emerges when consumers go into recorded voice or bot interactions assuming inability and opt for the human without engaging the computer at all. (This is the status quo for recorded voice customer service).
So why the Bot surge? Why would major tech companies and startups decide to release these products now?
A simple answer is that a competitive market forced many companies, large and startup, to announce a bot product — whether or not its available right now — in order to keep up with their peers. Similarly, one bot-focused headline may have rushed other bot announcements or launches, ahead of a previously planned timeline. In this scenario, the buzzwords and echo chamber aren’t far off where suddenly bot is synonymous with success and everyone is building one. However, I think there may have been something strategic in the rapid roll-out of conversational bot technology that is more or less unproven in its marketed product deliverables. Bot technology may become more sophisticated and advanced due directly to its initial launch craze, that traditionally in tech only serves to artificially juice early sign-ups.
The conversational, consumer bots that have sparked this craze are early drafts of machine learning let loose in the wild. In order to deliver the efficiency that the products promise, these ML bots have to be trained largely via supervised learning: feeding tons of specific data relevant to each bot’s function into a set of decision-making conduits over and over. In theory, the machine learns the patterns that determine which input data component elicits which decisions or responses and thus can begin to function with live interactions. One of the core reasons for continuous, relevant data intake is to capture and understand outliers in the set. In experiencing to the often unpredictable syntax, edge-cases, and nuances of any particular data, the bot is equipped to handle as many unique, real-world scenarios as possible once live.
Many impossibly vast and complicated data sets exist today like the index of The World Wide Web or a human genome sequence. Human language and colloquial conversation, especially via text, may be one of the most bizarre, unpredictable, and voluminous data sets in the world. Consider: the number of ways we say hello; the endless words or abbreviations that stand for yes; how emotion (ex: irritation) and expression (ex: irony) can completely change the intended meaning of a phrase (ex: sarcasm) without any grammatical change from the literal meaning; how syntax, spelling errors, and one’s location can impact our conversations. Then consider how much we say. It’s daunting and, as long as we keep talking, impossible to master. The startups and big tech companies that very publicly announced and/or released their early beta bots into live human conversation may have been angling for a jumpstart on this training. The bot craze inevitably has and will continue to draw people into human-bot conversations. Those interactions are the relevant data sets for getting a machine to understand the semantics of human communication just like it does a line of code. Some people may be irritated with the inbox presence of bot assistants Amy or Andrew from x.ai, but you better believe that team has gathered just about every possible way that we express our need to reschedule a meeting over email.
The first wave of bots may well fail to deliver on the promise alluded to at F8 or Google IO. Though if any of these early product releases were strategic in nature, several tech companies may have taken a audacious swing to advance their technology. Early PR and consumer blowback on half-baked products may have been expected and endured in order to compile the conversation data necessary for bots to function consistently at all.