Language and conversation is part of what it is to be human. A large part of human brain function is devoted to processing, creating and understanding communication in both verbal and nonverbal forms. It’s no surprise then that we find it easier to communicate through natural language rather than typing, swing, and clicking. However, the natural language of humans is not the language of computers.
Artificial intelligence is helping bridge the gap between humans and machines by empowering machines to have natural language capabilities. Not only is it making it easier to engage machines in useful dialogue, it is even helping to facilitate human-to-human conversation by enabling advances in machine translation as well.
Chatbots are software functionality that is designed to receive conversational input through text or voice and then generate a response that is also in natural language. (Disclosure: I’m a principal analyst at Cognilytica). Chatbots are best used in situations where a back-and-forth interaction is required. These scenarios include customer support and customer service, information acquisition, interaction with devices where physical input is inconvenient or impossible, or through voice enabled intelligent assistants. AI-powered chatbots are seeing widespread adoption and increasingly being used in a variety of ways.
History of chatbots
Eliza chatbot
To understand chatbots as we use them now it is important to understand their history. Chatbots date to the very beginnings of AI. Famous mathematician and grandfather of modern computing Alan Turing not only established many of the fundamental tenets of modern computing, but also designed the Turing Test in 1950 as a way of determining machine intelligence. The goal of the Turing Test is to determine a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human by means of a conversation that a human has with an unknown entity. According to Turing, you can determine that a computer system is intelligent enough if an ordinary human can’t discern whether the responses are coming from a machine or a human. In essence, one test of intelligence is the capability of a chatbot. Nowadays, some very advanced chatbots (and even the recent Google Duplex demo) are seeming to pass the Turing Test.
The first chatbot to really put the Turing Test to the challenge was Joseph Weizenbaum’s ELIZA program in 1966. Using typed commands and displayed interactions, ELIZA was modeled after a “Rogerian” style therapist. It would ask questions and lead the conversant into particular directions. ELIZA was surprisingly good at fooling people into thinking they might be interacting with a real person and not a computer. The kicker though, is that Weizenbaum didn’t claim that ELIZA was actually intelligent. Instead, he created a unique way for the program to generate open-ended questions and remember answers from the human respondents, turning those answers back to the person as if it was responding in a way a person would. He actually wrote ELIZA to debunk some of the hype and ideas of AI by revealing the “magic” of how it works.
Since ELIZA, many more chatbots have been created. Many chatbots created since have followed the same basic model as ELIZA detecting keywords and having various responses that can be combined in different ways, use previous answers for future responses, and have a basic flow of conversation. However, ELIZA wasn’t necessarily meant to show intelligence. It displayed how just cleverly using keywords and programmable sentences can fool users into thinking they’re having a somewhat intelligent conversation. Clearly, real intelligence isn’t just a matter of detecting keywords and pre-defined sentences. This is where more intelligent conversational systems need to come into play.
Chatbots: current and future use cases
Chatbots are currently being used in a variety of different contexts such as online customer support, phone interactions, information retrieval, or assisting with online commerce or tech support. Increasingly, we’re also seeing the increasingly popularity of intelligent personal assistants.
Because chatbots are fairly easy to deploy, companies find them a great first use case for AI within their organization. Businesses ranging from banking, finance, retail, and many others have AI enabled chatbots to help enhance customer engagement, collect basic customer information, answer product questions or general company questions such as the hours of operation and other similar information. Because bots can have an always pleasant tone and don’t need to sleep or take breaks, companies are able to keep them deployed 24/7 to engage and interact with customers.
Many governments are also seeing the benefit of chatbots. Agencies such as the USPS use chatbots to help customers with a variety of tasks such as tracking a package. U.S. Citizenship & Immigration Services use AI powered chatbots to help users answer a large variety of questions about various services, hours, and forms.
Businesses are also turning to virtual assistants as another way to engage with and interact with customers. Brands such as Tide, Patron, Purina, Campbell's, and many others are creating skills on popular voice assistants as another channel to engage with customers. These skills can help provide step-by-step instructions for how to tackle every type of stain, recipes and shopping lists, or other helpful information that brands wouldn’t be able to otherwise push to customers. There are many assistants on the market, with some geared more towards personal use and some straddling both business and personal use. As these voice assistants become more frequently used, the concern is really becoming how the information that they gather is being used, utilized, and shared.
A cautionary chatbot tale
Just like with everything, there are examples of chatbots that have gone horribly wrong. A famous example of a chatbot going rogue is Tay, which was designed by Microsoft in 2016. it only lasted for about 24 hours before it was pulled down. The program was designed to mimic the people it interacted with. Within 16 hours of being released, various internet trolls had taught the interface to essentially spout off Nazi rhetoric and anti-feminist speech forcing Microsoft to quickly shut Tay down, but the damage had already been done.
What the entire incident revealed is that the people who are creating these various technologies really need to be vigilant and careful about what they're producing and how much influence they want external users to have over the AI. Developers and companies need to be aware of how things can be used in ways they were not intended or how things can go terribly wrong once the technology is released.
The bottom line is that AI chatbots are best used in scenarios where human to human verbal and language interaction is required and necessary in order to complete a task. People demand increasingly intelligent interactions from businesses and governments and want instant and immediate responses. Because of that, it’s only a matter of time before bots become the norm for all customer-facing organizations.
