A chatbot is a computer program intended to engage with humans in human-like conversations. It is also known by other names like chatterbot, talkbot etc. Chatbots are the “in-thing” in the world of artificial-intelligence led technology. Both the press and the internet world are awash with stories of how chatbots are going to change the way we shop, dine, converse, and even stay in a hotel. There is a lot of hype around this technology, as is usually the case when people talk about artificial intelligence.
The scientist in me fully supports artificial intelligence when it can prove to be really useful. However, I do not necessarily agree with those who want to push immature technology down the throats of everyone. It only gives me the impression that this is being done mostly for commercial reasons, and not to really solve a problem that a society may be facing. Sometimes, it even gives the impression that a problem is being created solely to sell the idea that follows it.
Let me take one particular example application for chatbot technology: the development of “virtual assistants.” These are nothing but chatbots with a more business friendly name. The idea is instead of “human assistants,” these virtual assistants can be used to help customers with the services offered by an organization. For instance, if you wanted to shop for shoes, instead of setting different filters in a search box, all you need to do is to ask your question to a virtual assistant. Mona and Mezi are examples of virtual shopping assistants. However, most of these do not belong to any particular organization and are pretty much third-party applications. Consider Mona, Mezi, and the likes as being similar to the property brokers: They link a buyer with a number of sellers. E-commerce giants are not far behind—Amazon has its own version of chatbot named “Alexa” to assist you with shopping and many other Amazon related tasks.
Technology Assumptions Behind Chatbots
A lot of the advanced technology behind chatbots is an amalgamation of natural language processing, visual data processing, data mining, data analytics and search engine technology. Depending on the kind of services an organization offers and the various ways in which its customers have been engaging with it, some or all of these elemental technologies can be the used to design a chatbot. Now, if you examine these elemental technologies, you will realize most of these require structured data as input. Structured data obeys a specified format and follows a high level of organization. For instance, your monthly bank statement is an example of structured data, whereas your essay on your favorite pet is an example of unstructured data. A bank statement presents data in a certain format allowing one to readily extract information from it and perform analytics on it. However, your essay is a collection of fluid thoughts and a representation of your feelings and emotions for your pet. While an essay follows the rules of English grammar (or any other language used by humans), people do not always write with streamlined train of thoughts. Readers learn by making inferences, which can be speculative and depend on the reader’s background. However chatbot designers prefer structured input about which they do not have to speculate or guess the reader’s background. Such a structured input has a one-to-one correspondence with an action that a chatbot is programmed to perform.
Chatbots Still Lag Behind Humans
Chatbot technology-based virtual assistants are being pushed not only in the consumer space but also in the citizen space, where a number of services are provided by public organizations or government agencies. When users visit the websites of such agencies looking for information and help, they want their questions to be quickly addressed. One way to deal with it is to maintain a list of “Frequently Asked Questions (FAQ)”. However, if an FAQ is not updated regularly to reflect the changes in the service offerings, the FAQ serves little purpose. Now, imagine asking a chatbot a question and the chatbot refers you to a subset of the FAQ. How well does this work? In my personal experience, not well. The main reason is a number of these chatbots rely on specific words and phrases in user queries and they refer to a subset of FAQs based on these specific words without understanding the user query in totality. No doubt users feel indignant when seeing responses that do not address their concerns.
Advances in natural language processing (NLP) have not yet reached a stage where entire complex sentences can be understood by a program to extract its meaning the way a human would. At the same time, chatbot technology providers and adopters (especially business adopters) have to understand the technology must holistically integrate with those steps and measures a business may take that affect customer interaction. For instance, if a particular service is to be available only between certain dates, then the chatbot must have information about this. If a user asks a question related to that service after those dates, the chatbot should not refer to a FAQ that no longer applies. In this case, while replacing human customer service representatives with chatbot technology does save money, it may lead to customer dissatisfaction.
When a technology is hyped, it is also generally not well-grounded. It may not be weak from a research point of view, but can easily be lame from a real-world application point of view. Chatbots may be very good for assisting with shopping and dining. However, they are not yet good enough to replace human assistants for other services. At the same time, while the chatbot technology may be becoming more capable, it appears organizations are becoming dumber by using it in areas for which it is neither mature enough nor its integration with the operating environment (for instance, the business environment) is fully understood. No doubt that the same press that praised its potential is now beginning to talk about the “burst of a chatbot bubble.” The future of chatbot technology may be holding some promise but it is yet to be there.
