Will artificial intelligence (AI) make it possible to get good answers to your questions from chatbots? Or will chatbots sometimes provide articulate but wrong answers—also known as “botsplaining”?
Chatbots have been a channel for performance support and customer support for at least a couple of decades, as a form of FAQ (Frequently Asked Questions) files delivered to users of equipment and software. When a chatbot is well-developed, it can be a reliable (if incomplete) guide for new users of software or equipment. Unfortunately, when a chatbot’s developer does not conduct a good analysis of user needs, the result can be a source of user frustration. The truth is that even when a chatbot is well-developed, most people prefer to get help from another person.
Problems with chatbots and how to fix them
Why do businesses use chatbots? There are several reasons, most of them having to do with cost. First, a chatbot may appear to be more economical to develop and deploy than eLearning. Second, a chatbot is available “24-7”. And third, a chatbot online can respond to many people at the same time. Is there another way to provide users with the support they seek?
One way is to give users a choice: chatbot or live support? Make it possible for a human to pick up where the bot leaves off. The frozen holiday turkey providers realized this decades ago, as did the providers of support for software and pharmaceutical products. In both cases, developers added dial-up “help” numbers that customers could access from around the world. The problem came down to cost. Support staff expect to be paid for their work.
The alternative could be to make chatbots work better, without taking humans completely out of the solution.
Types of chatbots
There are two types of chatbots. Declarative chatbots are task-oriented. They respond within a limited domain to questions or requests from customers by posting scripted replies. Scripts can offer a choice to transfer a call to a human or offer an opportunity to transfer the call at a later time. The defects in this choice are obvious: people will still prefer to speak to a human.
The other type of chatbot uses newer technology such as AI to provide more human-like responses.
Declarative chatbots rely on a database of pre-written responses developed from frequently-asked questions, often ones provided by help desk staff. This database can be expanded upon through frequent reviews and updates to include and select more nuanced choices provided by humans or by software platforms such as natural language processing (NLP), natural language understanding (NLU), and even machine learning (ML). Keeping humans in the loop also helps to tweak the chatbot responses, a critical factor in the success of the chatbot and user acceptance. With continued development chatbots can become digital assistants, sometimes referred to as conversational chatbots. These learn to deliver increasing levels of personalization as they gather and process information.
In 2022, chatbot technology took a definitive step forward with the development of large language models (LLM). These are sometimes said to be context-aware but it is important to remember that the AI software developed on LLMs does not “know” what it is doing, and it is still necessary to keep humans in the loop.
One example of how AI can get it wrong was Galactica, an artificial intelligence developed by MetaAI. Galactica was shut down in December 2022 when critics suggested an early release produced pseudoscience, was overhyped, and not ready for public use. On the other hand, GPT-3, an autoregressive language model from OpenAI that uses deep learning to produce human-like text, seems to be faring much better than Galactica. An important lesson learned is the need for keeping humans in the loop and carefully vetting the sources used to train an AI application to eliminate the incorporation of human biases.
Develop your own chatbot
Making a chatbot is cheaper and easier than creating a cross-platform app or building an eLearning app. You can learn to develop chatbots by using online courses and tutorials. Most online educational sources and tools, from YouTube to paid courses, offer useful guidelines for creating declarative chatbots. Most require learning some programming skills but there are a few examples of no-code approaches. Just be sure to start with a solid understanding of user needs, test thoroughly, and make changes to fix the places where your home-grown app failed.