Mike is an accountant at a Fortune 500 company. Before work begins, he checks his retirement fund. The virtual assistant helps him review his balance and play some “what if” scenarios with his investment options and retirement timeline. Next, he logs on to his employer’s virtual private network (VPN) to start his work-from-home day. When he logs in, he lands on the intranet homepage, where a cheery bot welcomes him by name and asks, “How can I help you?” He types in “vacation days” and the bot responds with a link that provides general information about how the company allocates time off. That isn’t what he wanted, so he types a new question. “How many vacation days do I have left?” The bot answers, “You have 10 days left in this calendar year. Would you like to schedule them now?” “No thanks,” he replies.
His company uses Google for their email and documents, and they have set up a Google Hangouts bot. “What’s my schedule today?” Mike asks. The bot presents his calendar in visible form, highlighting a few places where two calls overlap. “Cancel webinar,” he says. “Which one?” Google replies. “Financial services best practices at 10 a.m.” “Cancel financial services best practices at 10, right?” “Yes.” “Canceled.”
After lunch, Mike notices that the network has slowed to a crawl. He accesses the Help Desk page on the intranet and is greeted by another bot. “How can I help you?” it begins. Mike describes the issue. “I need to connect you to a live agent for assistance. May I schedule a callback for you?” “Yes.” The bot presents several available slots and Mike picks one that works for his schedule. An hour later, he receives a call as scheduled and the issue is resolved with the help of a human agent. Mike never once told himself that the bots he engaged weren’t “real”; he only needed to be connected to a human assistant once. The era of chatbots is here. Talent development just hasn’t caught up yet.
As we’ve seen, specialized AI succeeds in performing a single sophisticated or complex task or related set of tasks. So far, this is the only type of AI that has been built. You don’t have to wait for the far-off day when general AI is available to start using AI to enhance learning. You can begin today with a much narrower focus and still achieve powerful results. One place to get started is the chatbot.
You may have interacted with a chatbot behaving like a customer service agent, for example, or used Google’s “smart reply” to handle email more efficiently. Microsoft displays gentle contextual reminders on how to best use its Office software using AI. Many other applications work beneath the surface of our digital lives; some collect data to help auto manufacturers decide what features will be most popular on next year’s models, or tell Amazon what products we might want to buy based on our browsing history. All these examples not only make life easier for the user—they change behavior. And behavior change means that learning is taking place.
Natural Language Processing and the Illusion of Conversation
A key component of the AI experience is natural language processing (NLP). NLP allows us to interact with a computer, system, or program through speech or ordinary text, without the need to translate our instructions, requests, or input into programming languages or code.
What Is a Chatbot?
A chatbot is not a website, a learning management system, a performance support system, e-learning, or an app, but it can enhance the user experience when paired with each of these more familiar learning technologies.
I define a chatbot as a customized user interface (UI) designed to deliver content and collect information through a direct, two-way conversation with the user. The two sides of the “conversation” can be delivered audibly, via human or simulated speech, or via keyboarded text.
You’ve probably used a chatbot and not even thought much about it. Online shopping sites, financial service companies, and software help desks all employ chatbots to provide basic customer services. The fields of education and corporate training have been slower to adopt the technology, but that is starting to change. For example, the online course provider Udacity is using NLP to make personalized recommendations to learners, while some online colleges are using it to take the place of a teaching assistant.
It is important to note that chatbots are not necessarily examples of AI. Simple chatbots with limited programming do not have to be particularly intelligent to be effective. While a “smart” bot can interpret the meaning and context of a request from a wide variety of human words and styles of expression, a “dumb” bot will be limited to the responses provided in its programming. Still, a dumb bot can serve many useful purposes, but sooner or later users will arrive at the point where they get a default response like, “I’m sorry, I don’t know. Let me refer you to someone who can help.”
While even the simplest bot can be very effective in handling routine requests quickly and just in time, the true power of a chatbot lies in the bot’s ability to analyze the learner’s own language and respond in kind. The result is an experience that feels like a conversation because, well, it is.
For the purposes of this book, any reference to “chatbot” is to an intelligent, AI-powered program using conversational input and output.
When you are “speaking” with something, your brain is hardwired to conclude that you are speaking with an equal—a being capable of responding back with its own independent thoughts. The fact that we know that general AI isn’t possible doesn’t prevent us from feeling that we are having a “real” conversation. Like any other illusion, we can’t help seeing it, even when we know it isn’t true.
Although this phenomenon has serious ethical considerations for the implementation of AI in our lives, it also presents some exciting opportunities, particularly as we look for ways to use AI to enhance learning and performance, including:
- interactive coaching programs that respond to learner questions
- more realistic branching scenarios for blended and e-learning programs
- learning management systems (LMSs) that make personalized recommendations for each learner
- data analysis to identify learner reaction, achievement, and preferences in near real-time
- contextual performance and process suggestions based on actions the user is taking at the moment.
ISBN: 9781950496310 | 162 Pages | Paperback
