Success in machine learning is increasingly built on relationships—between businesses and AI specialists, academic institutions and corporations, and humans and machines.
Seven of every 10 Americans will get an X-ray this year, making use of a diagnostic technology first developed in 1895. Today, fueled by machine learning and critical technology partnerships, one of the 19th century’s greatest medical breakthroughs is taking on innovative and new capabilities to save lives.
GE Healthcare—a pioneer of X-ray technology since 1896—has partnered with clinicians at the University of California, San Francisco, to create machine learning algorithms capable of distinguishing normal scans from those requiring immediate intervention. The system will be able to process and then learn from thousands of scans, with the aim of using the data to detect anomalies more accurately and efficiently. The goal is to help radiologists prioritize patients needing the most attention, improving health outcomes while reducing errors and inefficiencies that tend to drive up costs.
As the most common scanning procedure, X-ray scans generate incredible amounts of data, making it an excellent source upon which to develop machine learning solutions. The value is multifaceted, inspiring multiple lines of inquiry. ”How can we do things in a way that we increase efficiency for the hospital in the capture and interpretation of images, so they get maximum quality?” says Keith Bigelow, general manager of analytics at GE Healthcare. “There are many conditions we’re working on with X-rays and deep learning that are incredibly compelling and very much attached to saving lives.”
Addressing medical and other challenges through machine learning is no longer a new phenomenon. According to a recent report from IDC, a global market intelligence firm, investments in AI systems ”will explode to $52.2 billion by 2021, more than two-and-a-half times 2018 levels.”1 The technology, which allows computers to learn based on the data they process, has proliferated, thanks in large part to the availability of powerful, scalable compute resources from businesses such as Amazon Web Services (AWS). “My desire is that our AI team at GE Healthcare focuses on leveraging a platform like AWS SageMaker and bringing to it our unmatched clinical and modality expertise to enable precision health,” Bigelow says.
For many companies, the prospect of developing machine learning solutions in-house is less than ideal, whether because of cost, technical infrastructure, internal expertise or the time it takes to develop and train machine learning algorithms.
“There’s a knowledge gap,” says David Schubmehl, research director, cognitive/artificial intelligence systems at IDC. “There’s an experience gap. And a good partner can provide a set of capabilities to help bridge those gaps.”
Bridging the Gaps
Condé Nast has been a giant in magazine publishing for over 100 years. Today, it attracts more than 150 million people to its iconic brands, including Vogue, The New Yorker and Vanity Fair. The amount of data from content generation and servicing customers is staggering. “We collect everything from historic publications to interaction data from our websites. We also manage all of the rich digital media content our production teams create within our brands,” says Paul Fryzel, a principal engineer at Condé Nast. It amounts, he says, to “billions of events” every day.
Condé Nast saw the opportunity to use that data to advance innovation in the company. By partnering with AWS, the company was able to leverage a full suite of cloud-based machine learning tools to quickly build, train and deploy learning models touching virtually every part of the company.
“We plug them into tools within our editorial platform that optimize content descriptions. We plug them into our ad products and create models we can sell to our advertisers—people’s likelihood to act in a certain way, given that they viewed a piece of content or they’ve given some behavior,” says Lindsay Silver, vice president, platform and data at Condé Nast. “And obviously we plug them into our recommendation systems, finding the best pieces of content for a specific person or in a specific context.”
Financial services companies are similarly well-suited to benefit from machine learning platforms. Intuit, with its products and platforms like QuickBooks and TurboTax, has spent over a decade developing machine learning and AI applications. It allows them to tailor, train and deploy algorithms in service of their users, whether for categorizing transactions, determining patterns and anomalies in data, or providing overdraft protection. For example, Intuit applied machine learning methodologies to develop one of their newest features called ExpenseFinder, which automatically pulls a year’s worth of bank transactions and finds deductible business expenses through their machine learning algorithm, identifying on average $4,300 in business expenses.
For Intuit, centralizing machine learning initiatives with AWS has value beyond the products and services the company provides. It helps foster innovation and allows Intuit to deploy AI and machine learning techniques at speed and scale. “AWS gives people within Intuit a common platform to share and collaborate with data in a secure environment,” says Ashok Srivastava, senior vice president and chief data officer at Intuit. “For example, Amazon SageMaker gives us the platform and infrastructure we need to apply our sophisticated artificial intelligence and machine learning technologies.”
Solving the Big Challenges
Increasingly, companies that offer cloud-based machine learning platforms aren’t simply providers, but partners. “Amazon has been doing machine learning for more than 20 years,” says Swami Sivasubramanian, vice president of Amazon AI. “We continued to hear our customers ask, ‘How do we get that same level of experience—of the thousands of engineers that Amazon employs—to solve some of my challenges?’”
That was the genesis of Amazon ML Solutions Lab, where AWS engineers and scientists engage with customers to help them think creatively about machine learning and the problems they might solve through it.
“Not only are their developers now empowered to author and host machine learning models on top of the data that they have in the cloud, but now they can also collaborate with them as scientists,” Sivasubramanian says. It’s a concept working equally well for large companies with strong technical capabilities (Johnson & Johnson and the Toyota Research Institute, for example) as well as smaller organizations.
“Amazon SageMaker continues to resonate with enterprises as well as CIOs, because they see it as the single biggest tool enabling their developers to unleash, build, tune and host machine learning models on top of their data,” he says. “With the ML Solutions Lab, it has been a unique combination for them.”
That those models can be scaled at significantly lower costs provides another critical path towards innovation. Andrew Moore, dean of computer science at Carnegie Mellon University, believes the rise in access to resources and expertise has transformative potential. “Plenty of fantastic, strong companies don’t have a large AI staff. They’re thinking they need to do a $10 million or $20 million project to build capability. If this was 2013, they’d probably be right,” he says. “Now it isn’t—and services like AWS are a part of that. There’s open-source software. There are cloud providers. Suddenly, we’re not talking about $20 million projects anymore. We’re saying, ‘Use this and this, and you can get this out the door quickly.’”
The result is an acceleration of innovation. “There are huge disruptions happening in the sciences,” Moore says. “It’s possible we can make the pace for scientific research go much faster.”
The proliferation of strategic partnerships, and their variety, helps provide the fuel. “There are going to be more and more partnerships between academia, government and industry,” notes Adam Wierman, director of the information, science and technology initiative at the California Institute of Technology, another school partnering with AWS to drive knowledge and applications of AI and machine learning.
At the academic level, that means infusing AI into disciplines less typically associated with computer science.
“The phrase we have at Caltech is CS [computer science] plus X, or taking CS and using it to create new disciplines to help projects, regardless of the other field,” Wierman says. “We imagine that intersection being where the exciting new scientific and engineering disciplines are created.”
Outside of academia, Wierman believes those partnerships are equally critical to driving innovation and the ability to stay competitive in a rapidly evolving landscape. “Companies that don’t have an open view on creating these collaborations are going to fall behind,” he says.
Moore agrees. “It’s essential that we’ve got academics and entrepreneurs pushing those envelopes,” he says.
In the 19th century, IDC’s Schubmehl notes, when someone built a house or barn, the community was called in to help. “It took 30 or 40 people to do all the hammering, all the sawing. To put everything in place, get the bracing and so forth,” he says. “Today, we have power tools. Nail guns. Power saws. Lifts to help put things in place.”
Machine learning, Schubmehl notes, functions similarly. “Machine learning and AI are power tools for our minds,” he says. “We’re knowledge workers.” Armed with better data and decision-making tools, people within the enterprise become stronger and more efficient, with more capacity for creative, innovative experimentation.
“Machine learning works in sync with you, providing the capacity to do things faster, better and more capably,“ Schubmehl says.
To learn about AWS machine learning and AI, please visit here.