Chatbots are one of the most recognizable use cases for AI today. These programs are everywhere – from phone assistants to customer support chats – but it wasn’t always like this. Before the chatbot golden age we see today; these technologies went through a noteworthy drought.
So, why did early chatbots fail to deliver on their promise, and what makes today’s chatbots different? Here’s a closer look.
Early Chatbots and Why They Failed
The first chatbot emerged in the 1960s, but it would be several decades before bots were commonplace. After years of machine learning research and development, chatbot hype reached its peak around 2016.
By that year, more than 11,000 Facebook Messenger chatbots had launched, and 80% of businesses planned to use a bot by 2020. One study even suggested that people talk to chatbots more than their spouses. Results would soon fall short of these bold claims.
Chatbot mentions in company earnings calls fell to half their peak in 2018, and leading companies like Facebook had abandoned their bot projects. This decline was the result of three main reasons.
1. Misconceptions About Chatbot Abilities
First, many companies overestimated what chatbots could do in their excitement for the technology. While chatbots have many potential use cases, one bot can’t do everything. Businesses tried to create chatbots to serve multiple needs and complex tasks, which AI has a hard time with.
One of the most infamous failed chatbots, Facebook’s M, is the perfect example. M could make hotel reservations, get quotes from local businesses, send gifts, change flights, and more – at least, and it was supposed to. M’s developers quickly realized the bot couldn’t handle all of the complex requests users gave.
Alex Lebrun, the head of the project, later admitted that the challenge quickly outgrew their expectations. By 2018, Facebook had canceled the M project.
2. Reliance on Human Assistance
Many of these early chatbots also lacked full automation. According to some insiders, M had to turn requests over to human workers for around 70% of its responses. This reliance on human assistance slowed machine learning down – limiting chatbots’ growth – creating problems for users and businesses.
Reliance on human help undermines chatbots’ usefulness to businesses, as it won’t save them much work in that case. If humans are responsible for most chatbot responses, it also impacts users, who may not care to use them anymore without the novelty of talking to a robot.
3. Limited Technology
Finally, early machine learning technology limited chatbots. Companies around the mid-2010s may have been right about chatbots’ potential but not about their timeline. Machine learning takes a lot of information and training to become self-sufficient, especially with something as complex as natural language processing (NLP).
Chatbots have to understand the semantics of every word in context, not just know the words themselves, to understand natural speech. That’s a remarkably complex undertaking, especially considering how AI is best at structured, less nuanced tasks. Algorithms and datasets in 2016 simply couldn’t provide what chatbots needed.
How Chatbots Made a Comeback
After their 2016 to 2018 peak, chatbots seemed to die out. However, in the past few years, there’s been a resurgence. Chatbot usage as a brand communication channel increased by a stunning 92% between 2019 and 2020. More than half of consumers have talked to one in the past year, and only 12.8% had a negative experience.
Several factors play into this chatbot resurgence. Here are five of the most significant.
1. Playing to AI’s Strengths
One of the most important changes is that businesses have adjusted their approach to chatbots. More specifically, they’ve learned what AI is good at and started playing to that, such as the need for structured data. For example, chatbots can offer instant answers by giving users a selection of pre-set questions.
Users don’t have to use these default questions, but they give users a better idea of the chatbot’s scope. People are more likely to ask something the bot can recognize and understand instead of shooting for the moon. As a result, chatbots are often more limited in scope but offer more satisfying experiences because of managed expectations.
2. Specialization
Similarly, companies have leaned into chatbot specialization instead of trying to make bots that do it all. Giving AI a more narrow field of use makes it easier to fine-tune it to what a specific audience needs. This also helps with NLP, as bots will only have to understand a limited range of specific terms and phrases.
Chatbots have shown impressive adoption in industries with unique language and needs, highlighting this specialization. Experts expect chatbots to handle 75-90% of queries in banking and health care within the next few years. These sectors use many industry-specific terms and have more narrow use cases, which helps by giving AI more structured, predictable data.
3. NLP Advancements
It also helps that NLP technology has come a long way in a relatively short time. Part of that is because people spend more time on social media today than in the mid-2010s. Having more social media posts means chatbots have more real-world, natural language to learn from.
Transfer learning – which involves applying previous knowledge from other areas to learn more in another – also helps. Practical and scalable transfer learning NLP models emerged around 2018, leading to rapid improvements in NLP and making chatbots more useful.
4. Incorporating GUIs
Chatbot developers and businesses also learned that it helps use graphic user interfaces (GUIs). Today, many chatbots use unique visual interfaces, pictures, videos, and other visuals to engage users. This doesn’t necessarily help the bots themselves, but it makes it easier for users to get what they want.
Communicating over text, in general, can cause confusion without visuals. Applying this to chatbots both makes bot conversations feel more natural and eliminates confusion. As a result, today’s chatbots can solve users’ problems faster and with less frustration.
5. Necessity Amid COVID-19
The COVID-19 pandemic also played a role in chatbots’ recent surge in popularity. As lockdown restrictions and financial problems grew, customers had more questions, but businesses had less available staff to answer them. Chatbots were the obvious solution.
More than half of surveyed companies increased their AI adoption in response to the pandemic. Businesses could answer questions 24/7 and handle more queries with fewer workers with these technologies. Now that people have embraced them out of necessity, they’re getting used to them, which will likely lead to higher adoption in the future.
Where Chatbots Could Go From Here
As chatbots become more popular, new possibilities are emerging. Augmented reality (AR) is also growing, and the two technologies could combine to create more lifelike visual chatbots. Users could video-chat with digital assistants through AR as if they were on a Zoom call with a real person.
Chatbots are already starting to unlock new personalization opportunities. As more people talk with these bots, they can gather data about what customers like and dislike, increasing customization. Personalization could reach new levels as businesses use chatbots to learn more about their customers.
As this technology improves, NLP will also improve. Chatbots will be able to understand more complex language and complete multi-step requests, automating more daily processes. While these advancements may take several years, chatbots could finally become what people thought they would in 2016.
Chatbots Could Have a Long, Bright Future Ahead
This recent resurgence could be just the beginning of chatbot applications. If the 2016 hype peak and the ensuing drought have any lessons to teach, businesses and consumers should temper their technology expectations. Still, current trends paint a bright picture of the future, however far off it may be.
Chatbots are starting to look like what people eight years ago thought they would. As more people start to use these technologies, they’ll advance further, leading to more innovation and new use cases. The chatbot revolution is far from over.
