A Japanese tech company has trained an AI to give love advice to troubled hearts.
NTT Resonant, which operates the Goo web portal and search engine, created a system called Oshi-el to answer people’s relationship questions, like a virtual agony aunt. The researchers chose to focus on this genre of query as “non-factoid” questions are difficult for AI to address.
“Most chatbots today are only able to give you very short answers, and mainly just for factual questions,” says Makoto Nakatsuji at NTT Resonant. “Questions about love, especially in Japan, can often be a page long and complicated. They include a lot of context like family or school, which makes it hard to generate long and satisfying answers.”
Nakatsuji and his team trained their algorithm using almost 190,000 questions and 770,000 answers from the company’s Oshiete goo forum. Based on this data, they came up with a generic structure for answers that includes a sentence showing sympathy, a suggested solution to the problem, an additional comment and a note of encouragement.
The Oshi-el artificial intelligence then selects and combines appropriate sentences to use from a database based on the words used in the question. To combat the ambiguous nature of certain words – “relationship”, for example, could refer to a romantic situation or a business partnership – they use the category or title of each question to give the AI more context.
Money for love
For now, the answers still come across as a bit scripted, but they make sense. “I can see this is a difficult time for you. I understand your feelings,” says Oshi-el in response to a 30-year-old woman who finds herself stuck in a love triangle (the response has been translated from the original Japanese). “I think the younger one has some feelings for you. He opened up himself to you and it sounds like the situation is not bad. If he doesn’t want to have a relationship with you, he would turn down your approach. I support your happiness. Keep it going!”
It might work for love advice, but this approach is limited. It wouldn’t be able to write an essay, says Di Wang at Carnegie Mellon University in Pittsburgh, Pennsylvania. To be able to write a more comprehensive answer, an AI would need to actually “understand” the question. “We don’t even have a clear definition of ‘understanding’ [in the context of AI], so an AI can only grasp the very shallow surface of things,” says Wang. “But I think in this case people don’t care whether the advice is correct or not. You can say whatever as long as it sounds good.”
Eventually, Nakatsuji wants to be able to generate the advice word by word. “But it’s hard to get money for love,” he says. “If we develop it for [questions about] travel, we will be able to monetise it through hotels or restaurants.”