Chatbots are the new "it" things in technology today. They're computer programs that are designed to carry out conversations with users via text or speech. Famous bots like Alexa, Siri, Google Assistant, and Facebook Messenger are beginning to represent the next big computational platform. Bots are user-friendly, they afford tremendous opportunities for actionable insights, and they are ever-evolving, ever-learning, ever in beta.
So why aren’t they able to deliver on their promises? Why aren’t they replacing apps and websites?
1. The World Isn't AI-Friendly
First, the world is not that AI-friendly yet. There aren’t enough AI-integrated system platforms or infrastructure. Consider a company that is looking to introduce bots as customer service agents. The bots would have to be integrated with their existing system infrastructure in order to truly work and take actions on the behalf of users. Often, such an overhaul of a process presents a large cost that the customer is not yet willing to take on — especially if they haven’t seen successful implementations of chatbot solutions. Moreover, the majority of bots are based on iterations of if/else commands and pattern matching. They don’t leverage AI or natural language processing (NLP).
Imagine a digital assistant bot. If user communicates their intention, the bot is programmed to respond to said intention. However, if the user’s intention is communicated in a way the bot can't understand, the bot may not be able to respond in an appropriate manner. We’ve all seen something like this:
NLP-powered bots can understand "appointment" as a synonym of "meeting" in addition to understanding the intention behind such a statement. Advances made in the field of AI and NLP (augmented by synonyms, spellcheck, and so forth) have rectified the situation, but we’re not fully there yet.
2. Not All Bots Are Context-Aware
Second, not all bots are context-aware. While some bots can recognize and understand the context of a conversation, most bots don’t. Context is what makes a conversation as close to a human interaction as possible. Context is what makes us comfortable enough to want to carry on a conversation with a bot. Context awareness and training also take a lot of time and entail deep learning with neural networks. Imagine a bot that follows a decision tree-like structure. I ask for restaurant recommendations, and the bot returns a list. I could say, “But I don’t want to eat pizza,” and the bot will probably say it doesn’t understand because it doesn’t “remember” that it already provided restaurant recommendations and we’re talking about food. Conversations can be unpredictable and messy, and while no one expects a bot to fully understand every contextual clue, we do expect a certain amount of context awareness that not all bots possess just yet.
3. It's Not Them, It's Us
Third, it’s not them, it’s us. I’m not an evolutionary biologist, nor do I claim to know anything about psychology, but I made an observation based on things I’ve read, conversations with people, and time spent thinking about this topic. As sophomoric it may sound, humans have technological anxiety. Sherry Turkle highlights this anxiety in The Second Self; technology and technical devices almost become a second self, embedded deep in our consciousness as we grow ever-dependent on them.
This anxiety exists on a spectrum. On one end, we have ardent believers of technological singularity, in which people believe that AI systems’ evolution can surpass human evolution and transcend the world. On the other end, we have people who simply dislike or mistrust interactions with bots; those who may only want to chat with a human on the phone and not a machine; those who may not want to ask Alexa to order some toilet paper, but want to order it online themselves. Changing customers’ behavior is a challenging prospect. When you’re asking someone to change the way they do things in order to fit a new product or service into their routine, it can take a while before they can adopt said behavior. I read an article in college about how superfluous the iPad was because it would have to force people to not read a paper on their laptops or phones or print and do it on the iPad instead. Now, the iPad is pretty ubiquitous.
As developers, engineering managers, and technology enthusiasts, we embrace the chatbot. As a society, however, we’re still at the Early Adopters stage of the innovation adoption lifecycle. Perhaps we will, like we did with the cloud, move onto to the Early and Late Majority stages soon enough. Or maybe chatbots are just not that great in real life. But I suspect it’s the former.