Conversational AI - Digital first to human first

In the last few years, most of the boardroom discussions have echoed the word “digital first”, thanks to internet initially and smart phones later. The strategies around customer centric approach have also evolved from having a CRM system in place, providing multiple channels to interact, creating a seamless experience through multiple channels and beyond. The push obviously is to move more and more interaction to digital channels and rightly so primarily due to customer preferences as well as cost of operations.

This has led to plethora of digital interfaces like IVR, web, mobile apps which created the foundation of digital first/self-service. That is about to change again with the conversation based interface e.g. Google Home, Alexa Echo, and why not? We humans are more comfortable in our natural language conversations and why not teach the machines to learn that. However to travel the path from digital first to human first there are various factors which needs to be considered

1. Deep understanding of domain/context – To have a humanlike conversation involves a deep understanding of the domain and the context. In the recent demo of Google duplex it demonstrated the understanding of the domain and context by asking about the wait time for a restaurant booking. This means a painstaking journey of training and supervising on a large set of data.
2. Conversational AI is not a panacea – Don’t try to solve every problem with AI. Considering the amount of data and training required this is best applied to solve a specific problem at least for now. E.g. the conversational/chat bots in Banking are solving specific problems. HARO & DORI from Hang Seng are multi-lingual chatbots focused on simple consumer banking actions, Ceba from CBA can assist customer with 200 Banking tasks, Clinc from USAA is leveraging tonal and acoustic data in voice to provide almost like a non-bot interactions to customers.
3.  Privacy and security – The recent incident of Alexa sending a private conversation to someone in contact list has aggravated the security and privacy concerns. This is going to be a big factor in universal adaption of the technology. The fear of being continuously listened to is not going to go away soon.
4. Don’t let the machines run on their own – It’s been experienced repeatedly that you can’t let the machine run on their own. Facebook had to shut down their chatbot after AI develops their own language, Microsoft had taken down Tay which became racist in 24 hours. Hence, it is important to have a very strong governance layer around any intelligence we building.
5. Handling bias – The machines are going to be as good or as bad as humans are. It is not fair to expect AI to be unbiased if the underlying data reflects any stereotype. So the important fact is to acknowledge and understand our own biases and try to minimize that by being thoughtful about the data strategy, having a diverse team. This is more ethical then technical.

Even with above considerations, conversational AI is moving at a very fast pace. But one thing is for sure that this is not going to replace humans at least in the near future, instead we would need substantial human effort to make it work through data, supervised training and governance. What do you think?