This is where fast-evolving LLMs hold the promise for game-changing possibilities, everything from enhancing clinical decision support to automating administrative tasks, yet their integration into medical systems brings up a very fundamental strategic choice: the trade-offs of an open-source versus closed-source LLM framework.

Openness as a Strategic Imperative for LLMs

LLMs have often been critiqued as “black boxes” due to many concerns over reliability and transparency. Medicine, however, is one domain in which a decision may question the safety or life of a patient; hence, these concerns become amplified. Open-source models are easy to scrutinize, and the developer can view and adapt the architecture, datasets, and update mechanisms for their needs, furthering trust and innovation. In contrast, closed-source models retain these details in-house, making it very difficult for healthcare organizations.

For example, inconsistencies in the behaviour of ChatGPT, as cited by user observations in 2023, underlined some of the risks of opaque development processes. Such models can change without notice and have an effect on their reliability in medical contexts​.

Performance Dynamics: Open-Source versus Closed-Source Models

While closed-source LLMs like GPT-4 have shown state-of-the-art benchmark performances, their open-source counterparts are fast catching up. Open-weight models like Llama 3.1 (405B) and OLMo 7B Instruct show competitive results, especially on specialized medical tasks. Such flexibility of open-source frameworks allows fine-tuning to meet local medical guidelines—a key feature for precision medicine​.

Benchmarking and Adaptability

Without universally accepted evaluation methods, it is still challenging to benchmark LLMs for medical applications. While closed models might have slight advantages in certain domains, open models can quickly catch up through collaborative development. Human expert reviews are the current gold standard to ensure the effectiveness of these systems in supporting medical decisions.

Regulatory Landscape and Its Implication Over LLMs

Legal frameworks governing the deployment of LLMs also have a bearing on the open-versus-closed debate. For example, the AI Act of the European Union classifies LLMs as GPAI and, depending on their rating of systemic risk, may apply transparency requirements to the models. Open-source models under free licenses often meet such criteria better than closed models. However, overly stringent regulations can have the unfortunate consequence of stifling the development and diffusion of open-source innovations​.

Benefits of Open-Source Models in Health

The reason open-source LLMs meet healthcare priorities is for the following reasons:

  • Transparency: Open models enable inspection down to the minute details, hence addressing safety and ethical concerns.
  • Customization: It allows the developer to tune the models for specific clinical settings or guidelines.
  • Cost Efficiency: Open models alleviate dependence on high-cost proprietary solutions.
  • Community Collaboration: Contributions are open to researchers, clinicians, and technologists in open ecosystems.

Challenges and Future Directions

Of course, open-sourced LLMs have their own set of drawbacks, including the need for heavy computational resources and their possible misuse. That’s why some proponents of the closed-source model cling to hope that unhindered, open models can lead to such nefarious uses: just think of disinformation campaigns or, worse, even bioweapon development. The way ahead is poised on balanced regulations that ensure innovators are supported while reducing these risks​.

The proposition of living labs by the authors might finally create that bridge between regulation and innovation by providing a collaborative environment where all stakeholders, including patients, would engage in the co-development of transparent AI tools, identifying actionable deployment guidelines within clinical settings.

Conclusion

The choice between open and closed LLMs carries heavy implications for the future of AI in medicine. While it is obvious that closed systems have immediate benefits in constrained settings, the lack of transparency and ability to adapt limits their viability in the long term for medical purposes. Open-source models better align with the values of accountability and innovation critical to medicine.

As LLM capabilities continue to evolve, a balanced approach that leverages the strength of both models, underpinned by robust regulation and community collaboration, unlocks their full potential in reimagining healthcare.

Acknowledgement: This story is inspired by the article “The path forward for large language models in medicine is open” by Lars Riedemann, Maxime Labonne, and Stephen Gilbert, npj Digital Medicine. The original article can be read here under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Reference:

Riedemann, L., Labonne, M. & Gilbert, S. The path forward for large language models in medicine is open. npj Digit. Med. 7, 339 (2024). https://doi.org/10.1038/s41746-024-01344-w.

Dr. Prahlada N.B
MBBS (JJMMC), MS (PGIMER, Chandigarh). 
MBA in Healthcare & Hospital Management (BITS, Pilani), 
Postgraduate Certificate in Technology Leadership and Innovation (MIT, USA)
Executive Programme in Strategic Management (IIM, Lucknow)
Senior Management Programme in Healthcare Management (IIM, Kozhikode)
Advanced Certificate in AI for Digital Health and Imaging Program (IISc, Bengaluru). 

Senior Professor and former Head, 
Department of ENT-Head & Neck Surgery, Skull Base Surgery, Cochlear Implant Surgery. 
Basaveshwara Medical College & Hospital, Chitradurga, Karnataka, India. 

My Vision: I don’t want to be a genius.  I want to be a person with a bundle of experience. 

My Mission: Help others achieve their life’s objectives in my presence or absence!

My Values:  Creating value for others. 

Leave a reply