Anyone who is involved in enterprise communications technology knows our industry is moving and changing faster than ever before. As a consultant for many years now, I believe that now is the most exciting time in our space since the invention of the telephone itself -- a sentiment that was echoed by my consultant peers last week at the Society of Communications Technology Consultants International's annual conference.
Us consultants are tasked with incorporating new technologies into our knowledge bases; grappling with evolving and disappearing carrier offerings; wrapping our minds around new pricing structures; sorting through vendor consolidations, mergers, and emerging startups; and, of course, advising our clients on changing local, state, and federal regulations.
One of the new technologies almost everyone is trying to come to terms with is artificial intelligence (AI). At the conference we heard from a variety of vendors on their strategies around utilizing AI in their current offerings and future products. If you believe the vendor hype, within the next five years, we should expect to see AI integrated into all technology (not just communications).
As I've been trying to deepen my understanding of AI in our space, I'm reminded of wise words from my Rabbi: "You can't know where you are going if you don't know where you came from." So in that spirit, let's take a look at this technology's history to get better idea of how AI developed into what it is today and where we can expect it to go in the future.
Defining AI vs. ML
Artificial intelligence can be defined as a platform or computing-based solution that appears to be as intelligent or exceeding the intelligence of human beings. Colloquially, the term "artificial intelligence" is applied when a machine mimics "cognitive" functions that humans associate with other human minds, such as learning and problem solving.
Machine learning (ML) can be thought of as a subset of, or one example of, AI. In essence, it's when a computer program learns from data and experience to improve its abilities, accuracy, and overall performance when it comes to completing any given task.
Granted, one could draw a fine line between the two. But the fact of the matter is, today most of what is touted as AI is actually machine learning.
In 2016 Gartner classified machine learning as a hyped buzzword that is at its peak of inflated expectations. Effective machine learning is difficult to pull off because finding patterns in data is reliant on the quality of the data and/or the process of creating that data. As a result, machine-learning programs often fail to deliver.
Personally, I believe this definition to be a bit shortsighted, as it does not take into account scientific sampling that can help ensure accurate outcomes, or beyond, such as the ability of these programs to learn to execute in an anticipated pattern.
On the other hand, if you look at the outcome of a recent Facebook experiment, perhaps it is more difficult than at first glance. A few months back, Facebook published a blog post on its chatbot program research that aimed to train two chatbots to negotiate. It turns out that there were some programmatic errors made, and the bots were not properly incentivized to communicate in the comprehensible human English language we all know and understand, instead chatting back in forth in a nonsensical shorthand of sorts that only the chatbots could understand.
The actual negotiations between the bots appear odd and not particularly useful:
Speaking of Chatbots...
Now that we are all AI and ML experts, let's turn our attention to chatbots, as these are perhaps the most impactful application of AI and ML in our industry. Chatbots are essentially computer programs that are capable of having conversations with a human or other chatbot-enabled device. They can be fairly simple (and inexpensive) or decidedly more complex. A simpler chatbot might simply scan for keywords and rely on a limited, predefined response database, which a more sophisticated chatbot might use a natural language processing system to inform its actions and responses.
Chatbots are widely used today in messaging platforms like Facebook Messenger, WeChat, and Kik for entertainment purposes, as well as for B2C customer service, sales, and marketing purposes. A few well-known businesses that have adopted chatbots for everyday business use include Domino's, Pizza Hut, Disney, Nerdify, and Whole Foods. And of course, we continue to see more vendors incorporating chatbots in their various solutions.
Chatbots in Customer Service: A Deeper Dive
Reasons that a business might use chatbots are as varied as the list is long, but major drivers are cost reduction, better customer experience, data gathering, quicker response time, and accommodating a younger and up-and-coming customer base who have made it known that they do not like talking on the phone or to live humans -- a behavioral trend that has only grown in recent years.
A chatbot can respond to any multi-media or omni-channel interaction, and offer customers the option to speak to a live agent if it is not yielding the desired outcome. This makes chatbots useful for a wide variety of customer service use cases. Additionally, chatbots can be as advanced as to provide a plethora of reporting metrics, which can be leveraged by business leaders to improve customer experiences, identify potential product issues, or assist product managers in defining new or upgraded products, for example.
From a cost standpoint, bots are relatively inexpensive when compared to a the costs of having a human doing the same job. (Chatbots don't need health insurance!) Humans can be trained and retained for higher lever skill sets, customer problem intervention, up-sells, and other better suited jobs.
There are two main types of bots used in customer service: retrieval-based and generative models.
Retrieval-based models are pretty straight forward and easy to understand. Chatbots detect certain keywords to identify context around what is inputted, then respond appropriately, referencing a repository of predefined response. This type of system does not actually carry on what most would consider a conversation.
Generative models are more complex. Under this model, new and original responses are generated by bots, which rely on more sophisticated machine translation techniques.
From the business perspective, there are pros and cons to leveraging both models. Cost is a factor, but so is the expected customer experience. Retrieval-based methods do not make grammatical mistakes, however, they may be limited in their ability to respond, and incapable of providing an appropriate predefined response (How many times have you yelled at an auto-attendant?).
Generative models are "smarter" and can provide a more "human" type response. They can refer back to the input and give the impression that you're talking to a human. These models are hard to train(as in the Facebook example), are quite likely to make grammatical mistakes (especially on longer sentences), and typically require huge amounts of training data.
Taking this discussion one step further, in addition to these models, you have open domain and closed domain chatbots.
Chatbot conversations on social media sites like Twitter are typically open domain, which means the conversation can be taken just about anywhere. In a closed domain, inputs and outputs are typically limited to the specific task the machine was designed to achieve, such as technical support or customer support like in the case of shopping assistants.
For the immediate future, chatbots won't totally be replacing humans in contact centers. There are simply some contact center functions that will never be replaced by automation. However, the technology is evolving quickly and as improvements are made, businesses will be able to turn over more contact center functions chatbots.
Summing Things Up
To put my personal perspective on chatbot technology, I'll share a brief anecdote. I receive at least three to five emails per week inviting me to join a webinar about chatbots. Truthfully, I have ignored most of them, thinking to myself that chatbots are just like IVR on steroids. What was I going to learn? That a chatbot can detect my tone and transfer my call to a live agent if my voice displays signs of stress?
In researching chatbots more, I did discover that there is more to chatbots than I originally thought. As I've learned, while chatbot technology is still not perfect and has a ways to go, it is definitely a business productivity tool every company needs to be evaluating. We will all be using chatbots sooner or later, so why wait until the competition has "out-botted" you, or your customers start complaining about long holding queues?
"SCTC Perspectives" is written by members of the Society of Communications Technology Consultants, an international organization of independent information and communications technology professionals serving clients in all business sectors and government worldwide.