Conversational AI has exponentially seeped into almost every industry by changing how humans and machines interact and perform everyday tasks. 

 

However, not all conversational AI initiatives are created equal. Results have shown that those with the highest success rate are more strategic in their development and deployment than their rush-to-market peers. 

 

In a pre-event interview with former Forbes editor-in-chief David Churbuck, Aravind Ganapathiraju, VP of Applied AI at Forbes’ AI 50 company Uniphore, discusses the latest AI strategies for companies to deploy conversational AI and its future innovations.

 

Keep reading for his insights and watch the whole interview here.

 

 

Having been in the speech area for over 25 years including having done a Ph.D. thesis about using one of the early machine learning breakthroughs called support vector machines for speech recognition, Dr. Ganapathiraju described himself as “a dinosaur in the industry”. 

 

Aravind Ganapathiraju is currently the VP of Applied AI at Uniphore. He explained that, unlike core research, Applied AI is the bridge between research and product prioritization of ideas to cater to different use cases and developers would have to look at model efficiencies, model efficacy, and new features that are required to make machine learning successful other than the actual model. 

 

Prior to that, he was with Genesys for a decade working on Automatic Speech Recognition (ASR), Natural Language Processing (NLP), and Computer-assisted Instruction (CAI). He has led several teams that have successfully developed cutting-edge AI solutions to drive consumer and enterprise deployments. 

 

 

Aravind mentioned that the big transformation that’s currently happening is end-to-end neutral systems. In the past, it used to be a hybrid of statistical and neutral models. But now the need for that has almost vanished where a lot of systems, especially the Alexas of the world, are truly end-to-end neural systems with very little need for such knowledge of linguistics and sciences. 

 

Another strategy that has been implemented lately is using transformers in your production systems, which Dr. Ganapathiraju said that if your business is not using or planning to use it, you're behind the curve already.

 

He clarified that since transformers are encoder-decoder networks, they can help companies learn about a particular domain or concept by consuming tons of data in a completely unsupervised fashion. Upon obtaining knowledge of the language provided by the base transformer, you fine-tune it for your particular use case, which saves a huge cost for your corporation.

 

 

 

Aravind believed that the innovation would become feasible more from a use case standpoint. The next wave of innovation from his perspective is being able to deploy models and removing the need for any further fine-tuning. 

 

The reality is deploying in a complex use case can take months nowadays. Once the deployment time is decreased to a couple of weeks, the need for big professional services costs is no longer necessary and the end AI-powered products do not require any further fine-tuning, that’s when we’ve achieved sophistication with deployments.

 

At the upcoming Worldwide AI Webinar, Aravind Ganapathiraju will be discussing data challenges in applied AI research.

 

Reserve your spot at the event to learn more from him: https://wow-ai.com/event

 

Watch the whole interview here.