Kickstarted new year with an interesting experiment involving two outstanding teams to develop chatbots. They were given the same set of API’s , Python code ,configuration files and the infrastructure.With no addition /deletion of single line of code, the final outcome was amazingly different. Team which understood users & their day-to-day conversation was able to create more training data and model a superior chatbot than the one who couldn’t go beyond few regular bridled conversations. This instance is one classic example which heralds a tectonic shift in design process of an orchestrated monolithic mainframe/legacy systems to unstructured, predictably unpredictable AI world
The impressive and exponential progress in chatbot frameworks in last few years is just the baby steps in an assiduous exodus. Focus is more on predictions such as categorizing the “intents” and flow of conversation, area of personalisation is yet to gain the real momentum. A combo of Personalized & Predictable model will be the ultimate game-changer. To reach this vantage position, chatbot design need to eclipse three broader areas of contextualization.
Enumerating them in the reverse order of maturity
a- Behavioral Trends : Besides focusing on pure-play direct responses to customer queries, can Bot add intell from previous behavioral trait of the customer? For instance responses cannot be of same type for an impulsively extrovert and an overtly introvert personality . This trait details should be derived from previous touch-points and harness for creating a better customer engagement
b-Personalization: A well formed Bot should have a channel to continuously learning and refine the user profile from public domain. For instance, can bot personalize responses to a Yankess fan differently from Red Sox? Even in case of simple transactions such as garment selection, instead of prompting for size details in each new sales,can the bot remember,recommend and re-prioritize items fitting user affinity? Challenge shrouding personalization is real and convoluted in world of non inter-operable systems , nevertheless the barriers are slowly but certainly vaporizing
Behavioral and personalization traits are closely related yet different, as the former utilizes in-house data while the latter focuses on building a broader profile from external systems
c- Factoring in non-verbal cues : When majority of bot interaction occurring over smart devices, can it harness the facial expressions and adjust its responses? 76% of real world conversation is shaped by non-verbal cues, so a successful bot blueprint cannot afford to miss on such an essential item. This nexus of NLP and Image processing systems will propel the conversational AI to next orbit in detecting and effectively managing depressions
When word processor was gaining prominence, multi-purpose computer annihilated it completely, similarly emergence of Tablets cannibalised the online-reader market. In parlance, single purpose utility bots have a smaller shelf life. Federated model connecting a Data-Lake/Mart will ensure scalability, sustainability as well as value for dollars. Scope of bot design shouldn’t be limited only to Human-Machine combination rather it should cater to machine-machine interaction as this will dovetail into a robust “Personalisation” strategy
Standing at cross roads of Science and Technology, AI is presenting never-before possibilities to re-imagine our lives making it more likable and lovable. Changes we are about to witness in coming decade will be much bigger than those which impacted us in last 3 decades put together. To cater and sustain this tidal wave, we need to blend syntax of technology with the semantics and sentiments of human emotions !!
My other contents on this topic :