When chatbots are bad, they’re really bad. By now, most of us have had the frustrating experience of being automatically directed to an irrelevant FAQ page. At these times, feeling like you’re talking to a robot is all the more frustrating and reflects badly on the company.
When chatbots are good, on the other hand, they can be a huge asset to any customer support team. They’re available 24/7, they never get tired, and they can handle large volumes of requests without breaking a sweat. And the best chatbots of all are those that don’t feel like bots at all. These chatbots have been trained to understand emotions: aka emotional chatbots.
What are Emotional Chatbots?
Emotional chatbots are computer programs designed to simulate conversation with human users. These chatbots can range from simple rule-based systems to more complex artificial intelligence (AI) based systems that can understand natural language input and respond in a way that is natural for humans.
Emotional Intelligence (EI) is a set of abilities that enable a person or computer to comprehend emotions, and be able to identify (or “label”) them as a specific feeling e.g. excitement, disgust, etc.
AI enables machines to learn from experience and perform tasks more like a human. When AI is applied to EI, intelligent technologies, such as chatbots, can detect emotions and respond appropriately by successfully mimicking human emotions.
Chatbots are commonly used for customer service. While a normal chatbot is trained to retrieve useful information for the customer, emotional chatbots go a step further. The technology behind emotion chatbots strives to make chatbots more adept at responding not only to the actual question of the customer but also to their emotional needs.
This is something we do naturally as humans. Sciensio CEO, Chuck Elias, uses the following example:
“The technically correct answer to ’May I use the bathroom?’ is ‘Yes’, but the response you are
looking for when asking that question is ‘Of course, it’s over there!”
We automatically perceive that the person asking the question may feel embarrassed, or may need help finding the bathroom. And so, we respond in a way that is meant to be comforting and helpful, rather than just supplying the technically correct answer.
Now let’s look at an example of an emotionally sensitive customer query that a chatbot might assist with:
Customer: This might sound silly, but. I can’t seem to find the checkout button on your website.
Emotional Chatbot: That doesn’t sound silly at all! Sometimes it’s hard to find what we’re looking for online. Let me help you with that. I’ll walk you through the process.
As you can see, the emotional chatbot is not only able to provide the customer with the information they need but also respond in a way that is comforting and helpful. This is the power of emotional chatbots.
But do people want sympathetic responses from a computer or do we reject it because of its artificiality? One study set out to answer just this question and found that “expression of sympathy and empathy is favoured over unemotional provision of advice […] This is particularly true for users who are initially sceptical about machines possessing social cognitive capabilities.”
What’s Missing with Current Chatbots or Virtual Assistants?
The key question is: How can you create a machine that genuinely interacts with humans in a way that feels natural, and not like you are talking to a machine? There are several ways that current chatbots or virtual assistants are lacking:
Chatbots often misunderstand requests because they are not able to understand the natural language that humans use. The Global Consumer Customer Service (PDF) report found that 40% of US consumers and 50% of UK consumers still preferred talking to a human over a chatbot. The main reason cited is that the “Issue is too complex or unusual”. However, chatbots are generally preferred for quick and simple tasks.
A chatbot that implements advanced natural language processing and natural language understanding technologies (including EI) can help to overcome this issue.
Chatbots have the same answer for a query
Chatbots are easily identifiable because they have the same automated response for multiple queries. Emotional chatbots can create personalised responses that take away the feeling that you’re talking to a machine.
Difficulty in understanding accents, sarcasm and context
Chatbots are mostly not capable of understanding accents or cultural dialects to understand the right intent and in some cases struggle with the context and differentiation of sarcasm. For example, in the following sentence “I’m dead impressed with these headphones” the word “dead” means “very” in British informal English. However, the word “dead” in technology usually means “not working”. A chatbot that is not able to understand this context may begin trying to resolve the issue of the headphone not working.
Chatbots that use NLP will be able to make this distinguishment and respond appropriately.
Lack of training data
It is important to have enough training data so that the chatbot can learn from it and be able to respond accurately. People often underestimate the amount of training that chatbots and machine learning systems need in order to function accurately.
Dimension Research conducted a survey in which 81% of respondents admitted that the process of training AI chatbots with data was more challenging than they had expected.
In the long run, chatbots can save you money on customer support. However, the upfront cost of implementing and training a chatbot can be costly. You need to hire experts with well-programmed chatbots and a lot of high-quality training data to create text analytics models specific to your industry and requirements.
Introducing Symanto Brain
The latest AI technologies help with, or eradicate entirely, some of the main challenges of chatbots today.
Over the next few years, these technologies will completely reshape our expectations of chatbots as more and more companies embrace emotional intelligence AI. Symanto Brain is one such technology.
Symanto Brain is the no-code solution that enables you to create high-quality, custom text analytics models in minutes, a process that usually takes a team of costly data-science experts weeks or months to create.
Use Symanto Brain to create:
- Nuanced emotion models that detect emotions such as frustration and anger and trigger live agent takeover.
- An instant custom intent classifier with no or only a few examples, minimising the amount of training data necessary.
- An emotional chatbot faster and without costly installation.
We’ve been named Emotion AI vendor in Gartner Hype Cycle for Natural Language Technologies 2022 and our technology is used by some of the world’s largest brands.