I recently caught up with Beerud Sheth, CEO of Gupshup, to discuss the state-of-the-art for conversational AI and chatbot technology. He also gives us an idea for future areas of evolution for AI chatbots. Gupshup is a premier AI platform coming off a brand new partnership with Amazon.bot and weeks away from launching an AI-infused messenger app with a focus on privacy for Android phones called Gupshup Messenger. Beerud’s vision has been to seamlessly integrate AI and encourage its power to help with everyday tasks.
insideAI News: How do you see BERT changing the NLP landscape and making chatbots much smarter by enabling computers to better understand speech and respond intelligently?
Beerud Sheth: There are two major related developments in the NLP field:
- New language models, such as BERT, are showing encouraging results in broadly modeling the English language (other languages too).
- New techniques, such as Transfer Learning, enable developers to leverage general models developed in #1 above and specialize them to custom domains.
These two, together, are a game-changer for specialized tasks such as chatbot development. Chatbots can be a lot smarter in understanding user queries and responding to them accurately.
insideAI News: Can you identify a top trend for conversational AI moving into 2020?
Beerud Sheth: The improving accuracy of AI models, which has seen a step-function improvement in just the last year or two, will certainly lead to more capable chatbots with natural human-like interactions.
insideAI News: Please pick one-way chatbots can improve a company’s bottomline.
Beerud Sheth: Chatbots will enhance the company’s bottomline by:
1. increasing revenues (by automating marketing and sales for businesses),
2. lowering costs (by automating customer support) and
3. improving efficiency (by automating internal workflows, supply chains and partner interactions).
insideAI News: What’s your take on chatbots pretending to be human? In some studies, consumers clearly indicate they don’t like this.
Beerud Sheth: Yes, in fact, there’s a name for this phenomenon: uncanny valley.
Starting with non-intelligent chatbots, humans do want their chatbots to be smarter. However, as chatbots pretend to be humanoid, consumers start disliking it. However, after a certain point, as the robot behavior becomes even more human-like, it evokes the common human-to-human empathy. (Plotting likability on a chart resembles a “valley”, hence the name).
Thus, chatbots that want to really imitate human behavior have a harder challenge of crossing over and getting beyond the uncanny valley. However, for most tasks, chatbots don’t need to pretend to be too human. Chatbots enabling transactional tasks such as ordering things, booking tickets, sharing deals, providing support can stay focused and to-the-point. They just need to understand human language better, but execute their tasks efficiently.
insideAI News: Please take out your crystal ball and give us an idea for future areas of evolution for AI chatbots.
Beerud Sheth: AI chatbots are already gaining widespread acceptance – just look at the growth of Amazon Alexa, Google Assistant and the additional chatbots created by third-party developers. Chatbots will become ubiquitous as more messaging channels such as Whatsapp, SMS, RCS add support for chatbots. Chatbots will become smarter as AI makes rapid advances and the tools for building AI systems become more advanced.
Sign up for the free insideAI News newsletter.
Speak Your Mind