New coalition launches to tame Wild West of AI

A new coalition of over 30 leading US health systems and technology providers has launched to help build common standards and share knowledge on safe and effective use of the explosion of generative AI tools and models in healthcare.

VALID ( Vision, Alignment, Learning, Implementation and Dissemination) AI aims create a collaborative community to advance generative AI in a responsible manner to improve healthcare and research.

Speaking at launch of the initiative at the HLTH conference in Las Vegas, Ashish Atrejo, CIO and chief digital officer at UC (University of California) Davis Health, set out the challenge: “There is a huge and rapidly expanding number of models and use cases in generative AI, more than any one organisation can evaluate or assure.”

The new coalition, with founding members including UC Davis Health and Mayo Clinic, is designed to help spread and share knowledge of what different players are doing with generative AI, how they are evaluating different models and building shared knowledge.

He described the challenge as similar to the one that he faces at UC Davis Health; “A big part of my job is to share across UC ten campuses; we decided to open that to everyone.”

“We want to change the concept of open innovation. The skills and knowledge needed is distributed across different organisations; we need to get innovation from across many organisations,” said Atrejo.

He pointed to other open collective standards bodies in health as the model: “We have HL7 and FHIR as examples of how we have worked together before.  We now need to do something similar to Gen AI.

“If we don’t collaborate you wind up where we were with EMRs (Electronic Medical Records) 20 years ago, with every organisation going off on its own.” he added.

Dr Brian Anderson, co-founder of the Coalition for Health AI (CHAI), an initiative founded two years ago, was blunt about the challenge: “It’s the Wild West out there on Generative AI. We have to come together in initiatives like CHAI and others because we are going to risk causing harm to patients.”

He stressed how quickly the landscape has changed with the advent of generative AI: “An awful lot of solutions coming at lightning speed.  In just six months tech has moved very quickly.”

Dr John Halamka, President Mayo Clinic platform, meanwhile explained the new need that Valid AI was trying to fill is better sharing of limited skills and knowledge across the health system as it attempts to manage the explosion in generative AI tools and models.

“Two years ago we formed Coalition for Health AI, where the aim was to build the standards and the guardrails.  CHAI is large national standards setting initiative, and VALID AI a communications and standards sandbox.”

Cooperation as an imperative

While tools offer the potential to reduce administrative burden, Halamka said many clinicians remain understandably sceptical about the potential use of generative AI for clinical purposes.

The particular focus of VALID AI is sharing evaluations and implementation playbooks that enable generative AI models to be effectively and safely implemented across many organisations rather than bespoke for just individual organisations.

The difficulty of regulating and evaluating generative AI was described by David Rhew, Global CMIO at Microsoft, in a later session: “It’s very different to past regulation; just because you can deploy safely into one environment you can’t be sure it will be the same in further environments and the model will change over time.

Atrejo added: “If you look what Mayo is doing, each org is doing its own journey and assurance.  What is lacking is that when it comes to innovation we are completely blind to what other organisations are doing.  We have teams in different organisations evaluating the same technologies and policies of generative AI; we have to pool that knowledge.”

The only way to solve the new challenges of safely evaluating and assuring generative AI is through co-operation, “We learned during Covid that we have to co-operate and the competitive pressures between health systems has to end,” said Halamka.

“That’s not to say that we can’t specialise. Am I going to give you our playbook on how we implement a model? Yes.  Am I going to give you the specific models and data? No.”

“By embracing open innovation and radical collaboration, health care organizations of all sizes can lead the charge in creating value from generative AI in a responsible manner,” concluded Atrejo.

Umang Patel, chief clinical information officer at Microsoft, recently spoke to Digital Health news about the ability of generative AI to play a transformational role in healthcare with the right regulatory framework. Patel will be among the speakers at Digital Health’s AI and Data conference running from 30-31 October at the Business Design Centre in London. All sessions are CPD accredited, and the full programme agenda can be found here. AI and Data is free for the NHS, public sector, start-ups, charities, education and research. Commercial tickets start from £275+VAT. Register here.