Building chatbots: Be patient and grow your assistant
What's on the outlook for chatbots? And what challenges do you face as developer when building one?
For a long time, chatbots have been the number one example of putting artificial intelligence (AI) into practice. Although there are much more sophisticated AI applications, the simple chatbot interfaces make the technology tangible and accessible to the average user through smartphones and messenger apps. However, a massive breakthrough of chatbots is still to come. So far, interactions remain too linear to simulate the feeling of a fully fledged digital conversation partner. What’s on the outlook for chatbots? And what challenges do you face as developer when building one? We discuss this with Tim Groot, AI developer at e-office and speaker at our latest Developer Night.
“Organizations often step in with too high expectations and they want to build in a huge amount of functionality in a short time,” says Groot. “In addition, the time required to train the bot with data is underestimated. While this is an essential part of the ultimate success.” Groot uses booking a flight as an example. The entire booking process can be complex, while individual components such as changing seats or adding luggage can easily be done with a chatbot. “The rule is that you should start simple with a limited number of variables. This considerably reduces the chance of an impossible request being submitted to the chatbot. Otherwise, you will get answers that cannot be tied down, which leads to irritation for the user.” Groot advises taking enough time to feed a chatbot with data after which his tasks are gradually expanded. “This transforms its role from an ‘FAQ in a chat screen’ into the virtual assistant that many of our customers have in mind.”
Predicting coffee preference
Groot himself was looking for a simple chatbot application for his employer e-office to use to introduce visiting customers to the power of their default machine learning classifiers. Curious to see whether the stereotypes about latte macchiato or black coffee drinkers were true, he drew up a list of 16 apparently unrelated questions (“How intelligent do you estimate yourself?,” “Are you an outdoor person?,” “Do you often dream?”). With the answers, stereotypes were examined. For example, black coffee drinkers have a no-nonsense mindset. Espresso drinkers may rarely keep their opinion to themselves or seem to be late for an appointment more often. While cappuccino drinkers would be more adventurous and creative.
To test the hypotheses, the data set was first split and an algorithm was selected. The patterns within the data were then identified and tested. The string data was converted into binary variables with one hot encoding to insert in a machine learning model in Watson Data Studio. With just 80 answers, Groot managed to get the predictive power of his model above 50%, a percentage large enough to make the proof of concept successful for him. “In essence, this application improves the decision making of the chatbot based on machine learning, rather than predefined answers and links. The algorithm, therefore, ultimately determines which questions have sufficient predictive power to be included in or removed from the questionnaire.”
Groot sees a lot of potential for the application of chatbots within departments such as customer service simply because of the enormous amount of time that can potentially be saved. “In a next phase, there’s potential to apply the same algorithms within other domains, such as optimizing organizational processes.”
Do you want to get started with building a chatbot? Choose a code pattern here and build your own chatbot.