Why do chatbots often fail – Challenges and emerging trends


Chatbots are “chat robots” or machine agents that serve as natural language interfaces through text or voice, allowing users to ask questions or make commands in everyday language and get the needed content or service in a conversational style.

Chatbots dramatically change the way people interact with businesses online. As one of the leading messaging platform providers, Kik, claims: “First there were websites, then there were apps. Now there are bots.”

Reportedly, people are increasingly spending more time on messaging platforms like Facebook Messenger, which had more than 1.2 billion users per month in 2017. This fundamental shift in online user behavior has propelled all technology giants like Google, Amazon, Facebook, or LinkedIn and consumer service companies such as Starbucks, British Airways, and eBay to take on chatbots and reach their customers. According to Gartner, more than 50% of enterprises are likely to spend more per annum on bots and chatbot creation than traditional mobile app development by 2021.

Interestingly, chatbots aren’t new. They have been around for decades. But the real buzz about chatbots started in 2016 due to two reasons: first, the massive advances in artificial intelligence (AI). Second, a significant usage shift from online social networks to mobile messaging applications such as Facebook Messenger, Telegram, Slack, Kik, and Viber.

Despite these drivers, current chatbot applications suggest that conversational user interfaces still entail substantial challenges, in general, and for the field of human-computer interaction (HCI). Chatbots imply a change in the interface between users and technology; they imply changing user dynamics and patterns of use. This article will dig into what we see as a critical challenge with chatbots from a user-centered perspective.

Compelling conversational interface

A key success factor for all chatbots and natural language user interfaces is how seamlessly and efficiently they can support user demands in a conversational process. But this vision of a compelling conversational interface is not easily attainable. A possible explanation behind this is the difficulty in designing for open-ended conversations because chatbot capabilities seldom capture all the various ways the user wants to engage. There has also been a substantial push in chatbot development and a lack of concern for how people will use chatbots and for what purpose, potentially leaving users frustrated. Besides, we are also witnessing a rush from businesses vying to be the first to deploy chatbots in their service domain, often ignoring user needs and user experiences.

Bias and tricky interactions

Microsoft’s Tay, deployed on Twitter in 2016, is often considered as the ultimate chatbot fail. Perceived as an advanced AI-based chatbot, it was built to learn and mimic the personality of a 19-year-old through active interaction with Twitter users. But the problem was that Tay learned not only from well-meaning Twitter users but from Twitter trolls, exposing it to all sorts of hateful conversations supposedly coordinated by some trolling users in different online communities to abuse Tay’s commenting skills. Consequently, Tay mimicked many popular hateful memes online. Microsoft removed Tay from Twitter in less than 24 hours, after it praised Adolf Hitler and expressed anti-feminist sentiment in harsh language.

The lesson here is that chatbots, in general, need substantial adaptation to properly serve their task. As demonstrated in Tay’s example, the potential and non-predictable variation in user input and what constitutes a valid chatbot response represent substantial challenges.

To develop chatbots that can adapt to the specific needs and conversational contexts, there is a need for improved user and context models with specific interaction sequences to improve generative responses to inputs from users within a range of conversational contexts.

Lack of new user insights

Despite its long history, chatbots are still in rapid development, with advances being made every day. We can also expect that how people interact with conversational user interfaces will change, resulting in new user behaviors and social norms, and user expectations. Hence, more knowledge about chatbot experiences from an end-user point of view is necessary. New user insights are crucial for chatbot designers and developers.

It is important to inform these designers and developers on the desires, needs, and practices of chatbot users. Designing a new interactive technology such as a chatbot requires in-depth knowledge about why people choose to use this technology and why people stop using it. It is necessary to understand who use chatbots, their goals, the tasks they have to perform, and their context of use.

Obviously, but people have a tendency to open up more quickly to a computer or a technology than to humans since it is safer to participate in self-disclosure with a chatbot. But, we know less about what such relationships with chatbots can lead to in the long run, whether they can generate increased loneliness or depression or whether they can be balanced in a way that can be beneficial for users’ mental health.

To sum up, chatbots are not “one-solution-fits-all” technology. Most people use chatbots primarily for the effective and efficient accomplishment of productivity tasks. People typically use them to obtain assistance or information and expect effectiveness and efficiency, through customer service chatbots, in conducting productivity tasks such as access to specific content or help with administrative chores. Therefore, the main goal of a typical chatbot should be the gratification of its average user who wants to make life easier and more productive.