Balancing the Promise, Progress, & Problems of AI

There’s a lot of hype—and hysteria—when it comes to artificial intelligence (AI). Some expect that AI will improve nearly every human endeavor, leading to greater efficiency and accuracy. Others fear that AI’s expansion will eventually reduce the need for human workmanship. Could the truth lie somewhere in between?

AI’s potential was explored during a few sessions at the recent Advanced Manufacturing Minneapolis, particularly the panel discussions, AI & Advanced Sensors in Medical Manufacturing and AI: The Hype & the Hope. The first panel featured Dave Thiede, vice president of business development at Infinity Robotics; Wes Doty, sales director of North America at Mech-Mind Robotics; Justin Grammens, founder and CEO at Recursive Awesome & Lab651; and Rodney Landers, field applications engineer at Arrow Electronics. The second panel included Bill Betten, director of solutions at S3 Connected Health; David Schwietz, managing principal at Rockpoint Ventures & Consulting; and Tom Waddell, CEO at Waddell Group.

What Is AI?

Grammens of Recursive Awesome & Lab651 helped set the stage by defining AI. “What AI really means is allowing computers or systems to perform complex calculations and to solve things that are not explicitly programmed,” he told attendees. “AI is now allowing these systems to infer . . . when out in the environment and make intelligent decisions.”

AI may also be operating closer to applications, or what could be called the edge. “Most people think of AI as some big computing environment where you are sending tons of data to do training and everything happens in the cloud. And that’s certainly true and does happen, but that data is also minutiae,” said Landers of Arrow, suggesting that such volumes of minutiae could lead to overload. “So, for me, AI is driving intelligence to the edge, driving some level of intelligence even down to sensors . . . so sensors can [detect] when something is different.”

Grammens sees AI supporting personalization for patients. His firm is working with a company developing a wearable “that needs to be trained at the edge” through wearer input in order for the unit to learn over time. “As these products come out in the field, there is a level of personalization that’s happening for patients,” he said. 

When it comes to embedded AI, Landers said that “we are starting to see manufacturers come out with AI accelerators built into chips to do some of that work on the device.”

Advancing edge AI’s potential involves an understanding of TOPS (tera operations per second), which essentially means “the speed at which an AI can do its job,” Landers explained. For instance, Nvidia Jetson started with 100 TOPS, “which is pretty fast,” he says, and Qualcomm has announced its own capabilities. So “if you want to be battery powered, then TOPS per Watt is a thing to be concerned about,” he said.

Gains in edge AI could help doctors utilize real-time information during procedures. “They can’t go out to a server in the cloud and wait 5 minutes for information to come back—it has to be at the edge, and it has to be fast,” Thiede said. “Some of this will enable advancement in surgical automation.”

Landers agreed, pointing out a new ablation approach using AI “to determine what type of material is being ablated. . . . This is AI literally at the edge.”

Grammens mentioned TinyML, which allows users to “run very small models at the edge” that are “energy efficient.” And “you’ve got security built in—if all your data is at the edge and your inference is at the edge, you don’t need to send anything over the Internet, which is awesome,” he said.   

Other Medical Advances

“The jury is out a little on the value of AI and what it’s going to do,” said Bill Betten when kicking off the panel discussion, AI: The Hype & the Hope. 

He went on to tell the audience, however, that to date, “there are 522 AI-enabled applications approved by FDA, with over 400 in imaging,” Betten said. Not all are true AI, he said, but something is helping them, and a lot of these are “just really good data analytics and image processing—getting information and looking for a clot or cancer.”

Waddell pointed out Stryker’s AccuStop system that gives haptic feedback if a physician cuts into the wrong part of a bone, for instance. “If doctors are tired and cut in the wrong spot, the AI will say no,” he says.

Manufacturing Applications

“So, what AI means to me is faster ROI, faster deployment, and profitability,” says Wes Doty of Mech-Mind Robotics. “What used to take sometimes days and days to deploy, such as a vision inspection process or a palletizing process, can be done in minutes or an hour. That ability to learn based on information already gathered is profitability for me and my customers.”

Doty also shared hope for AI to step in where machine vision has struggled a bit. He described a customer with 12 manufacturing lines in which operators are picking up parts from baskets and putting them on conveyor belts all day long. “It would take them days to set up and optimize each SKU for the camera to pick it up in the optimal position for a robot. But with AI they are able to do so in a couple hours,” he said.

Landers added that what Doty was describing is called “regenerative AI—it is regenerating itself and getting better.” 

In response to the attendee question, “Will AI bring down the cost of vision inspection?” Doty said, “1000% yes!” He compared AI’s evolution to what has happened with flat-screen TVs. “AI gives you the ability to learn on the fly and vision will be as effective if not better than people,” he added.

“With better AI, you don’t need as high a resolution picture . . . so you know your cameras can be cheaper,” Landers added.

Data Matters

During the second panel, Schwietz said that “organizations are holding on to data, especially in healthcare, because companies are trying to monetize it—that’s the new gold rush.” But you have to be mindful of biases, he said.

Betten sees significant efforts to gather data. He points out that Mayo Clinic and Google and others have one of the largest databases in the world. And “the latest noble effort” he said is a new repository of healthcare records being built by b.well, Walgreens, and Samsung, he added.

Trustworthy data is critical to decision making. “I’m a big believer that if you are going to feed a medical system that’s going to make medical decisions about me, I want to trust the input of data from a medical device,” Betten added. “So, from my perspective, data integration, fusion, and trust that the data are good are paramount, particularly wearables.”

In terms of AI challenges, Waddell answered an audience question about FDA’s acceptance of AI by saying he expects there to be a limit on real-time database use. “They will be approving AI but [the AI] won’t learn until you do the next validation” of the database.

“Configuration management is still the order of the day,” Betten added. “You need to know what’s going in and you need to be able to control it.” He emphasized the importance of “transparency of the black box” as well as physician understanding of what the device is doing. “If so, it can be a great clinical decision support system. If they have no idea how that decision is made, they can’t use it.”

Future Outlook for AI

AI still has some bad press to contend with, however. And there’s President Biden's executive order to consider. 

“I know about the White House regulation—if you don’t put a check box on a product to say you’re certified and using the standard, you are suspect,” said Grammens. “If you aren’t transparent about who you are, you are suspect.”

But there’s also great potential. Grammens says that eventually “we will be able to talk to AI instead of pointing and clicking—at some point everyone will have an AI buddy in their pocket”

David Schwietz said he recently used a generative AI tool to create Microsoft Visual Basic code for a project. “Within in four seconds I had a code that I could cut and paste, and I did have to modify some things. But that would have taken me the better part of a day to figure out and remember the legacy code,” he said. “I am excited for the efficiencies in the software development world. People are worried about AI taking over jobs—I’m not so sure that’s going to happen any time in the next decade, but it will make a lot of jobs easier.”

Betten sees the potential for efficiencies, too. “With the shortage of doctors, I think the first places where we are going to see the promise of AI is not in diagnosing anything . . . but what it can be used for is to listen in on conversations with clinicians to generate a report and help patients understand things and search for information.” 

Schwietz added that there are “so many problems in the clinical space and challenges and opportunities for data to solve them.” For instance, “doctors spend 40% of their time on notes and records, often after hours.” He pointed to physician burnout and staff shortages and information overload. “In healthcare, information doubles every 72 days,” he said. “Can you imagine clinicians keeping up?” He expects to see rapid advancements in this area.

Many more developments are yet to unfold. Landers summed it up this way: “We are right at the beginning—this is the early adopter stage right now. We are not even at the early majority. In the Gartner Hype Cycle, we are way at the top of the hype cycle.”