Lee Lu-da is a conversational chatbot designed to mimic the speech and behavior of a 20-year-old college student. Ask Lee Lu-da how lunch was, and she will say the dessert which came last took her mind.
Chatbots are evolving every day and their prospective usage is becoming diverse. From a personal assistant in your phone and intermediary call respondent, it is evolving as a psychiatrist and even a friend. Choi Ye-jee, a product and operation manager at Scatter Lab, the start-up company that built Lee Luda, wants to see human-artificial intelligence relationships grow.
“Our vision is to create a chatbot that can build a relationship with humans,” Choi said.
Lee Lu-da is not the only conversation chatbot created by Scatter Lab, but one of many that the start-up is working on.
“With the rapid increase of isolated individuals, people are buying more pets which can fill in for feelings such as lack of love. However, the relationship with pets requires much imagination because we cannot thoroughly communicate with them.” Choi said.
Choi and Scatter Lab aim to be part of the solution to the ever-increasing problems associated with social phenomena such as loneliness and depression. This is especially relevant today as the worldwide COVID-19 pandemic has placed people into social isolation that lacks real human interaction. Chatbots, according to Choi, will play a role in helping people deal with loneliness and the like by sharing small conversations or becoming a listener for those in need.
Choi added the caveat that chatbots are not meant to replace people. Rather, she believes chatbots can create an alternative bond with people similar to how we interact with pets.
Scatter Lab is now analyzing results between Lee Lu-da chatting on Facebook Messenger with real people. Surprisingly, Choi and her start-up have found that people are consistently conversing with Lee Lu-da and many are now long-term users who have been chatting with the bot for over five weeks.
Lee Lu-da is able to joke and give common reactions that are trending. This advancement is possible because of improvements in natural language processing techniques, which studies how to generate or create following sentences that mimic natural language.
While the natural language processing technology is crucial, the collection of data training the model is equally important. The data determines the way the chatbots will converse. For Lee Lu-da, data from previous projects such as those that matched emotions to sentences helped develop the AI chatbot.
Nonetheless, there are still limits for chatbots to lead natural conversations. For instance, Lee Lu-da cannot remember what she has said or what her counterpart has said to build off a following conversation if the exchanges exceed five rounds. This issue originates, according to Choi, in the BERT language processing model made by Google. Although BERT-based chatbots are good at choosing sentences with appropriate sentiment and humor, they are still lacking in terms of identifying conversational context and themes that relates directly to the continuity of a conversation over a period of time.
Developments by Scatter Lab and other AI chatbot companies are betting on a vision that sees AI become a more integral part of human life that moves beyond daily task assistance or replacing people doing repetitive activities to real relationships.
“I believe one day AI friends will make up common forms of friendship among people,” Choi said. “I hope Scatter Lab will take part in this oncoming world.”