Bart de Witte: A Passion for Open-Source Medical AI

Bart de Witte discusses all things of medical AI and offers a glimpse of what’s in store at the #LINO23 Artificial Intelligence and Medicine Panel Discussion, which he will moderate on Tuesday, 27 June.

How is artificial intelligence (AI) used in medicine today?

AI has made significant contributions to medicine across various domains. Of course, the low hanging fruit has been medical image analysis where we have tonnes of images to train AI models that help us to make better informed decisions when looking at X-rays, CT scans and MRIs, and help radiologists to diagnose diseases and identify abnormalities.

For a few years, we’ve been using chatbot technologies. Symptom checkers where patients will chat with a bot are helping people to navigate healthcare systems better because we can do better triage. Chatbots are being used as well by people who have questions on their health insurance or, for example, by physicians who have questions where they need to have more information about certain drugs.

Furthermore, AI is increasingly becoming a fundamental tool in accelerating drug development processes. One notable application is the utilisation of AI in designing next-generation mRNA therapeutics.

Could you provide a couple of key examples where AI could be applied to positively impact the practice of medicine in future?

There are many areas where we cannot augment a physician’s capabilities because there’s no physician available at all. For example, Niger is a country with 25 million inhabitants and only three registered pathologists. I believe there is an opportunity for using AI to diagnose or give access to diagnostics where people didn’t have access at all, particularly in lower- to middle-income countries. These countries often have their resources outstripped by high-income countries, making AI an invaluable tool in their healthcare arsenal.

Also, physicians spending so much time behind the keyboard instead of spending time directly talking to patients is currently a huge problem. But because we now have access to these large language models, we are seeing the first implementations where physicians can eliminate the keyboard and really have a conversation with the electronic health records, improving the interaction with the systems of record, and allowing them to do it in a very natural way by talking. I said seven years ago in a keynote that we will move from paperless hospitals to keyboard-less hospitals, and I think the current progress of generative AI will allow this to happen.

Is this already happening and what are the challenges?

Pilot-wise, yes. The problem with large language models is that until you reach a certain level of accuracy, you cannot implement them. But if you compare the large language models today with four years ago, we have made huge improvements. Moreover, synthetic data is a valuable tool, and has been used to get access to more biodiverse data, for example, and create a better population distribution in the training data so you have less discrimination.

What is the aim of your Berlin-based non-profit, the HIPPO AI Foundation?

If physicians and clinicians fail to embrace open-source AI developments, they may face a fate similar to pharmacists in the late 19^th century. As the pharmaceutical industry grew, pharmacists lost control over their professional knowledge. Likewise, physicians could be relegated to simple users under the dominion of Big Tech’s control over medical knowledge. This would eventually lead to transgressions of norms and values, making it challenging to extract public or social value from the data these systems capture.

In order to prevent the monopolisation of future medical knowledge and resulting inequalities, I established the non-profit HIPPO AI Foundation. Inspired by Hippocrates, our foundation is dedicated to fostering and accelerating the development of open-source medical AI. We have assembled a community of 1000 researchers worldwide and are actively developing tools within a framework known as regenerative AI. This approach embraces principles like data solidarity, open-source licensing, open access publishing and collaborative community engagement. Ultimately, our aim is to collaborate with the industry to enable the practical application of open source solutions. By doing so, we can reduce their research and development costs while expediting the process of clinical and physician controlled standardisation.

Who have we got on the Artificial Intelligence and Medicine panel and what can we expect?

We have a Turing Prize winner, Shwetak Patel, and the Nobel Laureates Aaron Ciechanover, Avram Hershko and Michael Levitt, all of whom have worked in this specific field of medicine and have used machine learning, so it’ll be interesting to see how up to date they are. The conversation I want to have with them will involve reflecting on the easy access they had to data in their careers and then asking how we make sure that Big Tech companies or industrial conglomerates do not close off that data so that young scientists have the same opportunities they had, how do we assure that young scientists even have a profession?

I also want to have a conversation about turning data into what we call non-tangible assets. This is the same principle that made insulin in the US 1200% more expensive than 20 years ago, because it was being turned into a financial security. Last year, 100,000 diabetics died in the US, and part of that is because they didn’t have access to insulin anymore because it became so expensive. It would be incredibly foolish for humanity to permit our patient data to be converted into non-tangible assets. The question then becomes: how can we prevent this from happening?

What are you most looking forward to at #LINO23?

I’m looking forward to the conversations and I’m extremely excited with the exchange between these different generations. How can we avoid being exploited by financial markets when we talk about healthcare? How can we decrease inequalities when it comes to AI? How can we make AI work for us? These are all important questions whose answers are going to have consequences for the younger generation.

Attending an event filled with highly skilled intellectuals discussing these topics is undoubtedly going to be the highlight of my year. It will be fascinating to observe how these problems are perceived by different people and, more importantly, whether we can find solutions together and approach them with an open mind.