Decentralized AI in Healthcare: Preparing for ‘GOD AI’ and Its Agents

SANTA CLARA, Calif. —The god of healthcare is coming. And he’s bringing his agents, millions of them. At least, this was the vision of Ramesh Raskar, Associate Director, MIT Media Labs, as he laid it out at the most recent TiEcon conference here.

Don’t worry. Professor Raskar isn’t asking any of us to convert to a new religion. But it may be just as transformative to the way healthcare is delivered and the way healthcare IT organizations operate. When he talks about the god of healthcare, he’s actually talking about an AI that serves as an honest, impartial broker. “We already have these gods,” said Raskar, “Who is the god of traffic? Google Maps!”

In Google Maps, we choose to give up privacy (by telling it who we are and where we are) in exchange for the incredibly convenient direction service. In healthcare, the same concept is needed, but it is a much more complicated task. The privacy issues are even more complex, and the services delivered require deeper access to data, more complex simulations and wisdom, and more stakeholders needing access to all of it.

That’s where the agents come in. But before we get there, let’s take a step back and look at the problem.

In another panel at TiEcon 2024, called “AI in Healthcare Delivery,” two practitioners summed up the problems faced by healthcare today. Dr. Ganesh Krishna, a pulmonologist at PAMF/UCSF said, “There are three main ways to look at healthcare: access, utilization, and delivery. In all three aspects, there is room for streamlining the process. There is a shortage of professionals. There are access problems for social and economic reasons, and healthcare is expensive. For those reasons, use of AI to streamline the process and decreasing costs and time is vital.” He further pointed out that every minute where AI frees a doctor from administrative tasks is another minute that he can provide care and comfort.

Moving from sick care to healthcare

Daniel Kraft, founder and physician-scientist for NextMed Health and Digital Health, summed it up this way, “Healthcare is still very much in a sick-care model, and it is very reactive. The new model should be taking advantage of all these signals we put out into the world (from your smart watch, digital biomarkers, social media, and every other aspect of our digital footprint) toconnect the dots in our lives to help move from sick care to healthcare.” In Dr. Kraft’s model, for instance, our voice assistant could note early signs of Parkinson’s disease or neurological disorders. When we Facetime or use Teams, it could track our vital signs to predict strokes.

In other words, the future model of healthcare is data-driven and personal. And that’s how we get back to agents. Professor Raskar is predicting a near future where all of us take advantage of a series of agents or co-pilots in our personal lives, just as we’ve started to do in the office. Imagine, he suggested, planning a birthday party for your daughter where you ask an agent to do a lot of the heavy lifting. The agent might access your social media to find pictures of your daughter to see her age, her likes and dislikes, and create a list of friends to invite. Once it has a list of likes and dislikes, the agent might look for a choice of location for the party or decorations to purchase. And then based on your feedback and approval, it could actually book the location and order the decorations.

In each case, the AI doing the planning needs to access a series of other AIs. It needs to interact with the social media agent. It needs access to store inventories. It needs to interact with the reservation bot at the location. It is a series of agents talking to other agents. This is incredibly convenient, but it requires significant work on the part of IT departments. Data needs to be made available and removed from silos. Agents need to be able to interact in a secure fashion. Privacy and permissions policies need to be maintained. Now imagine this on the scale of healthcare where patient privacy must be balanced with healthcare access, and all the actions need to be monitored to make sure patients are receiving proper care and everything is accurately billed.

Professor Raskar’s ultimate vision includes a digital twin for every person on earth as well as a series of smaller agents working in concert as a sort of Global Orchestration of Decentralized AI—GOD AI. But we do not need to agree with the entire vision, or even expect it to be adopted as suggested, to see that something like this is going to happen. We already have machines talking to each other in nearly all sectors. This is just a larger scale and one tied to people (whether you want to call them patients or customers) where the same type of service will be expected.

Key steps for IT departments

And, of course, Professor Raskar’s talk did not encompass how the IT department should prepare for this future. But, in our view, some steps are fairly clear.

1. Data hygiene is king. If you have not completed your data journey, it is increasingly imperative that you have integrated data that is clean and ready for AI to digest.
2. Embrace zero-trust networks. AI agents will geometrically expand network traffic and the number of unverified sources looking to access your network. Zero trust is by far the best way to deal with this issue.
3. Enhance your privacy protection. AI-based healthcare will rely on ever-increasing data sets to do its job. All of that data from wearables and other digital footprints can be used to create large population models that can lead to healthcare breakthroughs. However, separating the personal data from the useful data for the model isn’t easy. There are multiple models for this including a federated data approach and what MIT calls a “split knowledge” approach. Adopting one of these approaches not only prepares you for AI-driven healthcare but can actually save you from embarrassing breaches.
4. Build more APIs. Another topic discussed at TiECon was that a major thing holding back AI was a lack of APIs. One example discussed was that despite the power of chatbots and avatars, customer service agents were not going to be soon eliminated. The major reason is that many systems are set up to receive only human input through a visual interface. APIs are needed to make sure agents can talk to each other without the need for human intervention.

Much of the promise of AI-driven healthcare, especially a “GOD AI” of medicine, is a few years away. But if you aren’t planning for the coming of “GOD AI” and its agents, you’ll likely find yourself praying later for help.

By Dave Wagner, Avasant Senior Director of Research