Welcome To The World Where Machines Do Better
by Karen Schwartz, on August 12, 2020 9:27:21 AM PDT
Priya Thampi, VP Product Engineering, believes that anything a human can do, machines can do better.
She’s spent the year helping teach machines how to infer.
Machines learn like humans
“Just think how we have learned to understand something, or how we have learned to infer something,” she says. “That’s the evolution of intelligence a human being has, the same thing happens with a machine, also, so slowly they start learning things, slowly they start understanding things, and in that way, they can start inferring things, and that inference is what can lead to machine learning, or we can say, artificial intelligence.”
"Anything a human can do, machines can do better."
Over her 24-year career, she’s worked with unstructured data in the service of customer experience and process improvement. She came to Infrrd in April 2019 via its founder, and as VP Product Engineering, she handles services and product-related work within Infrrd out of Bangalore with a team of 160.
Machines enable data evolution
She’s seen the data evolution happen, says Thampi, from the early web analytics that tried to think about how inference could be useful, to the implementation of recommendations and preferences that impact end users’ behavior.
She’s also seen the digitization of banks, and watched companies struggle to develop and manage customer onboarding online.
And that’s just a piece of the puzzle.
Technology fuels better decisions
The technology has marched onward, and her team has harnessed it in its current applications to successfully build a strong extraction engine and classification engine. These in turn fuel enterprises’ processes, and allow them to use worry-free data to better their decision-making.
Innovations such as classification and splitting capabilities are part of what moves AI closer to human solutions. They’re also key to giving customers the abilities they need to manage their day-to-day operations with straight-through processing, she explains.
Infrrd’s technology makes a difference
The proof is in the pudding.
So far, Infrrd’s technology is making a difference. Customers who have implemented Infrrd’s technology report that they are extracting data with great accuracy, which allows them to shift their focus off of managing data accuracy alone. Further, she says, customers believe in Infrrd’s approach – because it’s getting them results where otherwise they had been stuck.
"Customers believe in Infrrd’s approach – because it’s getting them results where otherwise they had been stuck."
Machines read better
“We had a customer, his motto was ‘if I can read something, the machine should read it better than me,’” she said. “And that’s his vision – even though it may not be 100% right, that’s a journey even I want to take, where I can say ‘if I can read it, my machine can read it better.’”
Digital document conversion as a trend has been picking up pace over the last decade, she notes. With Natural Language Processing (NLP) and Natural Language Generation (NLG), it’s possible to take data to the next level, by being able to understand the intent of documents instead of just extracting their values.
“When we look at a document the way a human does, the machine is able to understand that this document belongs to another document, or does in itself form 10 other documents,” explains Thampi. “That kind of intelligence is something we have brought in.”
Infrrd is constantly gaining ground when it comes to meeting customer needs in new and significant ways. Its machine learning capabilities have set it apart, bringing it out ahead of the competition to win bids for Fortune 500 companies’ projects. But there’s always more to learn, says Thampi.
She recalls a data extraction and classification challenge with more than a handful of document types, where the potential customer needed to know what type of document it was and to extract based on document type. To complicate matters, there were multiple documents in the same pdf.
“When you hear it now, it seems simple, because we have cracked it,” she says. “You need to split that, also almost like a human being, sitting and reading a document, to be able to say ‘the first page is a check, the second is an invoice.’”
Machine learning allows customers to make more sophisticated demands of their vendors, she says, on the extraction front and beyond. And, she adds, she’s looking forward to the evolution that lies ahead.
“We have started understanding a document’s intent, not just extracting the data’s values,” she says. “That’s a big difference.”