When one thinks of assembly-line production, their mind likely turns to thoughts of automobile manufacturing and other heavy industrial processes, rather than jumping to artificial intelligence (AI) and apps.
But according to recent Wall Street Journal reports, the Mayo Clinic is the most recent organization to use an assembly-line approach to AI development. It is not, of course, an assembly line as it has historically been known. These days, the line workers are small teams working with a common set of software tools and procedures, with an eye toward building out AI applications faster and cheaper by creating a “more consistent process to produce algorithms,” according to James Buntrock, vice-chair of the department of information technology at Mayo Clinic.
Mayo launched its AI factory last fall and is now gearing up to move into full production, with plans for 60 projects underway. Experts from neurology, radiology, pathology and other departments are working in concert with the new AI application team — which is made up of the health provider’s data scientists, machine-learning engineers and other AI specialists — to develop those concepts into usable health applications. Using agile management and a common software-development methodology, those teams break up the project into small tasks, develop various functions and features in short sprints, and adjust quickly along the way.
“Mayo teams use common data and software development platforms, including Google Cloud and Google’s TensorFlow, to build machine-learning applications,” the WSJ report noted. “The teams also use a set of computer languages, such as Python, one of the most common AI coding languages, and software-development collaboration tools. “
Those tools are getting more powerful, as Google has just announced a host of upgrades to its AI suite, including a TPU v4 chip that will soon be released to all Google Cloud customers, and will reportedly serve as a major upgrade to AI developers’ efforts.
And while Mayo is the biggest recent name to embrace the AI factory model, the approach is gaining popularity among corporations looking to build common process architecture that makes it easier for teams to share best practices and make development smoother and more efficient, Erick Brethenoux, a vice president analyst at Gartner Inc., told the Journal.
New York-based marketing firm Wunderman Thompson has embraced the AI factory model, said Cleve Gibbon, the company’s chief technology officer for North America. The firm has seen double-digit percentage improvements in both the speed of developing applications and cost savings since adopting it. “It’s a matter of building a hypothesis and having the means and the processes to actually take it from ideation all the way through to impact — at scale and at speed,” said Gibbon.
Scale and speed that firms are finding to be increasingly critical as they look to leverage AI in a recovering economy. Prime areas of interest for enterprises include building a better consumer experience (thus boosting sales) and upgrading their operations. That said, AI endeavors are often easier planned than carried out: According to Gartner’s data, only about 50 percent of AI projects make it through to the end from pilot to production — as data integration, data volume, complexity and lack of tools often shipwreck efforts as they get underway.
The theory behind the factory approach is that systemizing development will remove a lot of uncertainty from the system. “Just like a physical factory creates physical products, an AI factory should be able to deliver AI solutions in a systematic way,” said John Thomas, a distinguished engineer and director at International Business Machines Corp.