AI in Healthcare - The Real Open AI, AI on Device and Our Future

Bart de Witte Reveals How AI Can Save Lives and Why Open-Source Innovation Is Key to Preventing a Crisis

TL;DR: Bart de Witte, a healthcare technology veteran, has shifted focus from corporate roles to democratizing healthcare through open-source AI. He envisions AI deeply transforming healthcare in the next 10-15 years, with AI assistants enhancing the capabilities of healthcare professionals. He emphasizes the importance of keeping AI open-source to prevent the privatization of medical knowledge, which could lead to greater inequalities. De Witte also discusses the benefits of on-device AI for privacy and control, the challenges of implementing AI on resource-constrained devices, and the ethical implications of AI in healthcare decision-making. He advocates for open-source AI as a tool for innovation, particularly in underserved regions, to ensure equitable access to healthcare advancements.

Table of contents

• Vision for AI in Healthcare

  • AI's Role in Transforming Healthcare Professionals' Work

  • AI Assistants and Enhanced Capabilities

• Ethical Concerns in AI Healthcare

  • The Privatization of Medical Knowledge

  • The Importance of Open-Source AI

• AI on Medical Devices

  • Benefits of On-Device AI

  • Challenges in Implementing AI on Resource-Constrained Devices

• The Role of Open-Source AI in Healthcare

  • Addressing Healthcare Disparities

  • Successful Implementations and Examples

• The Future of AI in Healthcare

  • Potential Innovations and Impact on Patient Care

  • Open-Source vs. Proprietary AI Solutions in Healthcare

About Bart de Witte

Bart de Witte has been a prominent figure in the healthcare sector for over 25 years. Initially aspiring to become a dentist, Bart soon realized the challenges and high suicide rates within the profession. His passion for technology, dating back to his days as a computer whizz-kid in the 80s, led him to pivot towards the digitalization of healthcare.

Bart’s professional journey is marked by significant roles and achievements. He began as an SAP consultant and project manager, where he successfully implemented SAP systems in large clinics. His expertise in the field grew as he took on roles in product management and business development at SAP. In 2010, Bart transitioned to IBM, where he transitioned into a global management career. He was responsible for the public market and the healthcare industry across Central and Eastern Europe, where he encountered healthcare inequalities for the first time. Later leading digital health in his role as Director of Digital Health at IBM Germany.

In 2019, driven by a desire to democratize the healthcare system, Bart left his corporate career to found the Hippo AI Foundation, a non-profit organization focused on open-source and data democratization projects in healthcare. As open-source AI has become established, Bart is now in the process of building a new AI startup, about which he can’t reveal much at this time. However, he has found a way to combine open-source AI and profitability in a for-profit company and is getting closer to his main goal.

Future Scenarios of AI in Healthcare

How do you envision AI changing the role of healthcare professionals in the next 10-15 years?

I believe that AI will rapidly change healthcare in the next 10-15 years, much like operating systems transformed computing. AI will be seamlessly embedded in numerous devices, including machines and robots that we will interact with effortlessly. With open-source language models capable of understanding voice and language, these advanced functionalities could already be integrated for less than EUR 20. I anticipate seeing the first home appliances using voice and language as a user interface. I can't wait to replace unfriendly interfaces and simply tell my washing machine at home to wash 12 white shirts at 10 o'clock in the evening, without worrying about my data leaving the machine. This can all happen offline, ensuring privacy and convenience.

These intelligent machines will understand human communication and natural language, changing healthcare as we know it. Every healthcare professional will have an AI assistant, akin to having a personal AI companion. These personal AIs will support clinicians with a range of tasks, from planning and prioritizing to answering complex questions and managing daily routines. They will act as digital mentors, available 24/7, far surpassing human capabilities. Imagine having a mentor who is always available, smarter than any human advisor you could access—this will make healthcare professionals significantly more knowledgeable and effective.

We all know the adage, "you are the average of the five people you spend the most time with." Now, imagine that one of those "people" is a highly advanced AI mentor. Utilizing these AI systems as mentors will enhance your intelligence, which is crucial in healthcare, where saving lives and advancing careers depend on continuous learning and improvement.

Consider the impact of DeepMind's AlphaGo. When the super AI AlphaGo player was made accessible as open source, those who used it as a mentor improved their skills by an average of 20%, even when not using the AI. In healthcare, AI assistants will become indispensable, serving as the primary interface for all AI-related communication and decision-making.

Beyond healthcare, In essence, AI will be viewed as a new digital species, transforming the role of healthcare professionals by making them smarter, more efficient, and more capable than ever before. This evolution will not only save lives but also significantly enhance the quality of care and professional development in the healthcare industry.

What potential ethical concerns do you foresee as AI becomes more prevalent in healthcare decision-making? Pick the most important.

The most significant ethical concern as AI becomes more prevalent in healthcare decision-making is the privatization of medical knowledge through financial assetization. This practice leads on the long term to increased prices and restricted access, as seen with the dramatic rise in insulin prices in the US. Despite claims of democratizing healthcare, large tech companies often engage in democracy-washing, privatizing data and AI under the guise of democratization. This creates a feudal-like system rather than a truly democratic one, exacerbating future inequalities and limiting the accessibility of crucial medical advancements.

In your opinion, what areas of healthcare are likely to see the most significant AI-driven innovations in the coming decades?

All

AI on Device in Healthcare

What are the key benefits of implementing AI directly on medical devices rather than relying on cloud-based solutions?

We are reversing control and ownership. For over a decade, we've discussed decentralized movements, but due to the need for computing power, the opposite occurred. Current business models rely on centralization, data monetization, and asymmetries of information and power. However, with decentralized AI, we can access intelligence in offline mode. I can use large language models on my phone to ask intimate questions, get recipe advice, travel guidance, or medical information, all while being completely offline and without sharing any data. This is a game changer. In healthcare, I am convinced this will lead to increased trust, interaction, and adoption. For my new venture, I've carefully considered how these new possibilities can lead to innovative business models. As a result, I am developing something that will be unlike anything we've seen before.

How do you see edge AI and on-device AI transforming patient monitoring and real-time diagnostics?

Remember these intelligent weight scales that one can connect to the internet in order to combine your data with intelligence on your phone with your personal data? I recently found out that the adoption rates or active usage of IoT-enabled weight scales and their associated apps is really low. With reall low I mean really low, like bellow 10%. I bought my first Withings scale in 2010 and even tweeted my weight to my Twitter account to use peer pressure and lose weight. I experimented with many apps and was part of the quantified self movement. Most of our community got frustrated as over time, we lost control on our data. A friend of mine in the US, Hugo Campos got famous for his efforts to gain access to data from his own implanted cardiac device, advocating for patients' rights to their health data. His work has contributed significantly to discussions about patient autonomy and the role of data in personalized medicine. With edge AI, we are opening new possibilities as we can create new possibilities and make personalized digital medicine 100% personal.

What are some of the technical challenges in developing AI models that can run efficiently on resource-constrained medical devices?

We are still at an early stage. With my new venture we are scratching the limits of innovation. But I anticipate some first principles. Hardware and neural chips will drastically improve during the next few years. We need to work on new memory architectures. The movement of data between processors and memory is a critical bottleneck. Power consumption and thus battery management need to improve as well.

Open Source AI in Medical

Open source AI has the potential to significantly address healthcare disparities in underserved regions by providing equal opportunities for innovation. Open source empowers individuals and communities to develop solutions independently of BigTech and other monopolistic entities. This approach helps eliminate market failures that have resulted in the power asymmetries of BigTech. Recently, I was in contact with a researcher in Africa who was building voice-enabled large language models on devices running on Raspberry Pi hardware, which he acquired for just a few dollars. He didn't need a credit card to access APIs from OpenAI or other providers; he was free to innovate. This demonstrates how open source provides access to digital resources, enabling innovation.

Imagine if Gutenberg had patented the alphabet and required everyone to pay a license fee for writing. Low-income countries would remain illiterate, relying on charity for survival.Perhaps Luther never would have written his Bible, and we still would be buying letters of indulgence. Similarly, during a visit to Guatemala, I spoke with a researcher who struggled to read research papers, let alone afford publishing fees. This is not the world I envision. I believe in the liberty to innovate. By embracing open source AI, we can create healthier markets with affordable products and services, fostering true equality of opportunity. This approach ensures that even those in low-income regions have the tools to innovate and contribute to global advancements, ultimately leading to better healthcare outcomes for all.

And last but not least, don't assume that health inequalities won't affect you. With gene therapies costing over 3 million euros, you might find yourself unable to access life-saving treatments in a decade from now. Ironically, these treatments are developed using the data you shared when consenting for research.

Can you discuss some successful implementations of open source AI in medical diagnosis or treatment planning?

Although open source AI communities are successfully challenging the monopolies of OpenAI, Google, and others, we are still in the early stages when it comes to healthcare. The healthcare sector remains largely dominated by established players and is not typically known for its openness. However, change is on the horizon. A particularly fascinating example is OpenAPS and AndroidAPS. AndroidAPS (AAPS) is an open-source artificial pancreas system designed for individuals with insulin-dependent diabetes. This smartphone app, which runs on Android devices, aims to automate insulin dosing to maintain healthy blood sugar levels, effectively mimicking the function of a real pancreas. Remarkably, it was developed entirely by patients dissatisfied with the quality of industrial solutions.

Another good example is HealthSage AI in Holland: This platform is dedicated to generative AI in healthcare, offering an open-source generative AI framework. It hosts its own Large Language Model (LLM) and encourages collaboration within the healthcare community to build and customize AI applications. HealthSage AI emphasizes safety, compliance, and transparency, allowing users to review and modify the underlying code to fit their workflows better. Their models are trained using high-quality health data and undergo rigorous validation processes to ensure reliability and trustworthiness in clinical settings

What are the main challenges in adopting open source AI solutions in healthcare settings, particularly regarding data privacy and regulatory compliance?

The biggest challenge is not data protection, as the business models of open source and edge AI would improve data protection, but regulation. During the AI Act, lobbyists indirectly funded by Silicon Valley billionaires like Elon Musk, Dustin Moskovitz, Jaan Tallin or Sam Bankman Fried successfully created laws that hinder open source deployments. With the exception of Sam, who is thankfully in jail, they have all invested in their AI unicorns and see open source as a threat. It's similar with global MedTech companies, strong regulations play into the hands of the larger well-funded companies and thus a few. Meta, who has been actively contributing to the open source AI ecosystem as they understood the theory of commoditizing the compliment, has just announced that they will not release their multimodal LLaMa-3 models due to regulatory risks. This reminds me of the Sultan of the Ottoman Empire who banned the printing of Arabic letters in 1485, leaving the Islamic world with an educational deficit. I'm not sure we want that.

What are some of the most significant open source AI developments you have seen, and how are they impacting the AI and healthcare field?

As mentioned, we are at a very early stage, but surprisingly, Microsoft Research has begun releasing open-sourced medical foundation models. Their Gigapath foundation model was trained on 170,000 whole slides and over one billion image tiles. A few years ago, people criticized my idea that LLMs would be linked to domains such as pathology imaging, molecular biology, therapy, and outcomes and will be open sourced. Well, Microsoft released the model with a proprietary license that doesn’t allow commercialization, but it is available as open source. I expect that, similar to Meta’s LLaMa, for which the first release did not allow commercialization, open science groups will eventually release these models as open source that can be commercialized. When and how this happens depends on all of us, and our vision of the future we want to live in.

In your opinion, what are the main advantages and potential drawbacks of open source AI compared to proprietary AI solutions especially in healthcare?

I believe those who oppose open source haven't fully understood that AI can take one of two paths: commoditization or assetization. If we commoditize AI through open source, startups, innovators, and others can innovate faster and more affordably, leading to greater success. Transparent AI models build trust, resulting in higher adoption rates. On the other hand, if we follow the path of assetization, we could end up relying on a few large players, which would slow down innovation and result in costs that are significantly higher. The choice is ours: do we want to give each researcher, entrepreneur, and innovator the opportunity to build independently and create equality of opportunity, or do we want all medical knowledge to become a private asset, monopolized by financial markets, leading to unprecedented power asymmetries?