How to help chatbots to select the only one answer that fits to the conversation?
Chat is becoming most popular form of communication between people and businesses. And companies react to the growing trend quickly. More than 80% of businesses want to automate their chatting by 2020.
Chatbots for a while seem to be the right solution to scale personal chatting of brands. But soon it became obvious that chatbots for now can’t replace humans in complex conversations. Most of the chatbots today follow simple rule based models and understand human languages only in their most basic forms.
The problem is in the complexity of our communication. Very small nuances can change the context of a sentence. AI is not yet on the level where it can be 100% sure that it is giving the right answer.
It’s easy to reach certain level of confidence and rank possible answers. The hard part is to select the only one right answer, among the many possible ones.
It’s a bit like in case of running a marathon. Runners say, that the real marathon starts at 32km (20 miles). For most marathon runners it’s relatively easy to reach the 32km sign (20miles), but then “the wall” comes and makes every new km a hell of a pain.
Machines can do the “first 32 kms” by clustering and ranking answers based on historical data. But the item that the algorithm ranks to the top of the list is usually not the good answer (see: Facebook chatbots operate with 70% failure rate). Today the algorithms can not fully modell our conversational behaviour, so they struggle to overcome the wall.
Imagine that you couldn’t browse movie recommendations, but a “chatbot” would take over your Netflix and tell you what to watch next.
Of course it would save you a lot of time by eliminating browsing, but I would stop watching Netflix after a few bad movies. And conversations are even more complex than movie recommendations.
So does it means that machine today can’t help to automate complex conversations with customers?
BUT IT DOES!
But not the way chatbots approach it.
Instead of replacing humans, technology could help humans to be more productive.
Humans can use their emotional intelligence to overcome the wall, and select the right answers from the options provided by the machine. Machine works hand in hand with humans. It recommends best matching answers based on historical data, humans select the most appropriate one.
With this approach AI technologies can augment human capabilities resulting in increased productivity level. Humans can answer repetitive messages quicker, so they can focus on more complex conversations. This extra time for special care means a lot today, when quality of customer care plays bigger role in shopping decisions than ever.
At Chatler we believe that the road towards chat automation starts with human and AI cooperation. Chatler augments the capabilities of human chat agents, by recommending responses for user requests, and learns from their interactions. Once our algorithm gets confident in repetitive topics it takes over the conversations.