Las Vegas - Executives from major U.S. tech companies gathered at this year's HLTH conference, pledging to lead the next wave of innovation in healthcare AI.

AI products launched by companies such as Google, Microsoft, Amazon, GE HealthCare, and NVIDIA claim to help healthcare systems solve a range of problems, from reducing documentation time to optimizing operating room scheduling. Tech executives said they have moved beyond the early stage ofpoint solutions-focused AI tools, and are now offering platform-based solutions that can be tailored to the needs of individual healthcare systems.

"We are truly at an inflection point," said Sally Frank, Global Lead for Health and Life Sciences at Microsoft.

However, as tech companies market their products to healthcare systems and providers, executives are divided on how to responsibly introduce new technology to the industry - and whether clinicians are equally ready to embrace these new tools remains an open question.

"Any evaluation we do at the foundation model level is just a first draft," said Greg Corrado, Senior Director of Research at Google and co-founder of the Google Brain team. "It really needs to be led by healthcare systems that are willing and able to do research on the ground, and not every healthcare system can do that."

Tech companies emphasize platform-based approach

Tech companies are positioning themselves as partners to healthcare institutions in AI development, offering deep expertise in product development and testing.

However, tech executives at HLTH also cautiously noted that they rarely tell healthcare systems directly how to apply AI.

"Google is not a healthcare provider, and we don't want to be one," Corrado said.

The executive said that AI use cases should come directly from healthcare systems. He compared tech companies to semiconductor chip manufacturers - they produce resource-intensive, high-value raw materials - in this context, large language models - which the healthcare industry then applies.

"We develop technology that should enable healthcare institutions to envision and build their own future in this space," he said.

Google has partnered with institutions such as the Cleveland Clinic and community health systems to pilot various technologies, from healthcare-specific cloud platforms to documentation tools that can search electronic health records.

Other tech companies, including Microsoft, have launched similar collaborations, encouraging healthcare systems to customize their own AI solutions.

Kees Hertogh, Vice President of Product Marketing for Health and Life Sciences at Microsoft, said this commitment opens the door "wide open," going beyond "out-of-the-box" tools that address "direct use cases."

Earlier this month, Microsoft announced it would make it easier for healthcare systems tobuild their own AI tools directly. The company is also using generative AI to help healthcare systems organize unstructured data as well as imaging and medical image data, enabling customers to "build their own copilots, their own AI agents," Hertogh said.

Bill Fera, Principal at Deloitte, believes healthcare systems are likely to prefer such tools over earlier products.

"I think as people become more familiar with building AI themselves and understanding how to operate it, the hyperscalers and platform players will dominate this space," Fera said in an interview. "There will be a shift from applications to proprietary platforms."

Robust testing requirements raise questions of access

In healthcare, tech companies say the current focus is on helping healthcare systems manage information overload. Their ability to process vast amounts of data in written form and images can help reduce the time providers spend reviewing medical records, orschedule patients for critical surgeries, executives said.

But different AI applications require different levels of human oversight.

Currently, tech companies at HLTH agree that all healthcare AI should have human oversight. Some companies are focusing first on administrative AI tools because they require less oversight. For example, at GE HealthCare, scheduling tools may require less oversight than tools used to assist in cancer treatment.

"These are areas where we think AI can be introduced very quickly because it doesn't directly involve clinical decision support," said Abu Mirza, General Manager of Digital Products and Senior Vice President at GE HealthCare. "It's not directly tied to decisions about someone's surgery or other matters. This is where AI can truly get started."

However, when AI approaches clinical decision points, more rigorous testing must be conducted, said Google's Corrado.

AI "hallucinations" - where models generate answers not supported by source text - have drawn widespread attention.

But Corrado believes omissions are equally important. Omission refers to AI failing to cite relevant information in its answers. From some perspectives, omissions are harder to test because they require a comprehensive review of often lengthy medical records.

However, executives from Microsoft and Google noted that not every healthcare system has the resources or expertise to rigorously test cutting-edge technology.

According to Hertogh, Microsoft hasconnected healthcare systems seeking to use AI, enabling more advanced systems to provide resources and knowledge to smaller systems before deployment. The consortium has more than 15 hospitals and healthcare systems, including Providence, Advocate Health, Boston Children's Hospital, Cleveland Clinic, CommonSpirit Health, and Mount Sinai Health System.

Microsoft provides voluntary testing guidelines for users of its AI products.

"We provide guidance and technology so you can build visibility and transparency into the system... It's a bit like, 'Hey, this is how we build it, this is how we think about responsible AI,'" Hertogh said.

According to the executive, Microsoft also benefits from the information exchange. While Microsoft offers its perspective on how to implement AI, the exchange is two-way. Interacting with leaders from some of America's top healthcare systems helps build trust in Microsoft's products and establish credibility, he said.

Hertogh believes this broad collaborative spirit is relatively new. He thinks it can help smaller healthcare providers access new technology and act as an "accelerator" for broader adoption of generative AI in healthcare systems.

However, Google not only recommends that healthcare systems test its AI - the company almost mandates it.

Corrado said clinical feedback is an integral part of Google's testing process, so far, the company has only worked with healthcare systems that have a certain level of "maturity" in AI and share common values regarding model testing.

"I think healthcare systems that want to buy ready-made solutions - maybe they should wait. Wait for things to mature and stabilize," the researcher said. "I don't think there is a sufficiently mature evaluation framework in any (healthcare) context that allows you to deploy technology directly... Nothing is monolithic. You have to put it in a real environment, evaluate how it performs in actual use, and adjust accordingly."