A unique DALL-E-generated image combining ChatGPT's perception of its appearance and its ideal physical embodiment.

Introduction and Background

Around 2015, while running my medical writing business (CoreMed Communications), I came to realize that the rate-limiting step in my work was the time it took me to write. As a medical writer who dealt with providing standard responses to physician questions based on published clinical trial data, I recognized that there is a formula to this process. Consequently, it should be possible to automate at least the initial creation of standard responses. This realization set me on a path to explore the potential of AI in pharma, specifically AI med info, to streamline the medical writing process.

Discovering ChatGPT

In early 2023, I began experimenting with prompts in ChatGPT, an AI language model. I quickly found that writing was something it easily excelled in. However, I also discovered that it had some weaknesses in providing accurate data and facts. ChatGPT tends to give a “best guess” when it comes to medical information and sometimes even fabricates references that appear surprisingly realistic. This led me to consider the potential of training the AI to write responses based only on summarizing the data it was trained on.

Defining the Parameters

To create a useful AI model for my purpose, it needed to fulfill certain requirements:

    1. Only provide information it was trained on.
    2. Present the information in a customizable, logical flow.
    3. Adhere to a style guide, such as the American Medical Association (AMA) Manual of Style (11th edition), as well as user-specified style guides.
    4. Output the summarized text into a pre-specified Word template.

Acknowledging the Limitations

While AI has immense potential in revolutionizing the medical writing process, there are some limitations to consider:

    • AI models struggle with creating tables and graphs. However, it might be possible to generate outputs that can be easily inputted into tables or fed into graphing programs or Excel.

    • Data checking is crucial since AI may infer and add its own information or mischaracterize something. It is not possible to control every bit of output.

    • AI models rely on pulling and preprocessing text from PDFs, Word documents, and websites. If the text extraction is not possible, AI won’t be able to summarize the content.

Experiments Conducted So Far

I have tried a few different approaches in my journey:

    1. An AI model that provides answers to both patients and doctors using only the prescribing information for a product. This approach worked, but I noticed the model would occasionally include information outside the product information.
    2. An AI model that checks a document to ensure alignment with the AMA Manual of Style and other pre-specified style guidance. This worked too, but it wasn’t as exciting since programs like spell check and Grammarly have been around for years.
    3. An attempt to create a program that extracts key sections of a study and feeds them into a Word document that can be summarized by AI in a subsequent step. This required numerous parameters to identify key sections, but AI seemed to be a better fit for this task, as OpenAI has no problem identifying study objectives, results, conclusions, etc.

What’s Next?

My latest experiment is focused on creating a custom AI model trained on specific data, style guides, and templates to automate writing standard responses for medical information. This idea was inspired by a leaked Google document discussing “personal AI” from the open-source AI community. I have already started working on this solution, and I am excited about the potential it holds for enhancing the role of artificial intelligence in the pharmaceutical industry.

As I continue to explore the possibilities of AI med, AI in pharma, and machine learning in pharma, I invite you to join me on this journey. By sharing my experiences, challenges, and successes, I hope to engage you in what I am doing and encourage you to follow along as I navigate the world of artificial intelligence in the pharmaceutical industry.

My mission at Med Info AI is not only to streamline medical writing processes but also to foster an open dialogue with professionals who work in medical information within the pharmaceutical industry. By keeping the conversation focused on innovation, best practices, and the responsible and ethical use of AI, we can work together to shape the future of medical writing in the pharmaceutical sector.

Stay tuned for more updates, experiments, and insights as I continue to advance my work at Med Info AI. I welcome your thoughts, questions, and feedback, as your involvement is crucial in creating a supportive and collaborative community centered around AI and medical writing.

In conclusion, the journey to leverage AI in the pharmaceutical industry is filled with challenges and opportunities. As a dedicated professional in the field of medical writing, I am excited to share my experiences and discoveries with you as we delve deeper into the potential of artificial intelligence for transforming the way we create and deliver medical information. Together, let’s shape the future of AI med, AI in pharma, and machine learning in pharma, one step at a time.