January 26, 2017
Machine learning will transform how software is created and evolves; business processes will evolve with it
“If you control the code, you control the world,” wrote futurist Marc Goodman. But the way that software is coded is undergoing a revolution that may have broad and deep implications for knowledge workers and programmers alike.
A recent article in Wired Magazine highlights some of the recent advances and implications of machine learning. Unlike traditional programming, where an engineer writes explicit, step-by-step instructions for the computer to follow, machine learning doesn’t require programmers to encode computers with instructions. Rather than programming computers, they train them.
Let’s say you want to teach a neural network to recognize a cat, for instance. By relying on machine learning, you don’t tell the network to look for whiskers, ears, fur, and eyes. You simply show it thousands and thousands of photos of cats. Eventually, the system figures out what makes something cat-like. If it keeps misclassifying foxes as cats, you don’t rewrite the code. You just keep coaching it.
Today, Facebook uses machine learning to choose the stories that show up in your News Feed. Google Photos uses machine learning to identify faces. Machine learning also runs Microsoft’s Skype Translator, which converts speech to different languages in real time. Self-driving cars use machine learning to avoid accidents. Even Google’s search engine has begun to rely on these deep neural networks.
Teaching computers to herd cats (and process payments)
Identifying cute cat videos or recommending Facebook clickbait to read is one thing—but what about automating document- and information-intensive processes like procure to pay or sales order processing? Now we’re really talking about herding cats.
Today, the leading intelligent document capture solutions are evolving towards a machine-learning approach. Intelligent capture platforms transform scanned document images and digital files and then try to identify relevant, actionable information. In the case of vendor invoice processing, for example, intelligent capture software would: 1) convert any “images” of words to actual text and then, 2) identify information like a vendor name, address and invoice amount for processing.
The knock against these systems in the past was that they were too costly, too complex and too rigid. The net result was that too much customization was required to set these systems up—and even then, they would only function within narrow parameters, depending on the format of the document.
Creating more flexible, adaptable intelligent capture systems that can “read” an invoice or other document type has meant programming them for adaptability. That includes making it easy to create templates to guide their process of identifying relevant data in context, and creating user-friendly interfaces to validate results and allow the system to learn from that validation process. It also means establishing databases of documents and data that the systems have already learned, so they continue to learn and adapt as they are exposed to new documents types and formats.
For any organization looking to automate a business process that involves data and documents, intelligent capture solutions offer the opportunity to eliminate the most common and most inefficient and error-prone tasks—document sorting and filing, and data entry.
Existing, adaptable solutions to handle these tasks exist. Applying machine learning to intelligent capture will do for information processing and data entry what template-driven Web sites did for html programming and Web design.
For more information about intelligent capture and business process automation, visit the Artsyl Technologies Web site at http://www.artsyltech.com/.