AI for Medical Writing – Remember Clippy? Soon to be your new best friend – Morula Health’s perspective
Contrary to every other blog out there, I am not going to start with a definition of what Artificial Intelligence (AI) is – you should have a good idea by now if you are reading this blog or accessing any form of media these days!
I will also try to limit the number of times I say ChatGPT. And I am not going to reference the Goldman Sachs report on the impact AI will have on all jobs, rendering us all useless, as we may as well pack up and go home!
(https://www.goldmansachs.com/insights/pages/generative-ai-could-raise-global-gdp-by-7-percent.htmlon)
Instead, I am going to explain how we think AI will be used by Morula Health and Medical Writers in the future, why it won't take the job of our Medical Writers, and what you can use it for now. Maybe in fifty years' time I will be proven incredibly wrong—but let's see. The International Monetary Fund (IMF) is still a thing, and they are always wrong, so I'm not too worried!
AI is going to impact absolutely everything that involves data and text, and pretty much every job that involves a computer. So, basically, everything, right? Do you remember Clippy? This guy:
Microsoft has put a lot of resources into OpenAI and ChatGPT (once), and they have announced their intention to integrate the GPT-4 language into Office 365 in the form of ‘Co-Pilot’. As such, we anticipate an incredible revival for Clippy – it's sure to be monumental!
Microsoft, Google, IBM—these tech giants are all heavily involved in the rapidly advancing AI space. Unless you are planning to use one of their advanced language models, e.g., OpenAI's GPT-4 model I think it’s a lost cause to try creating any AI tool that specializes in resolving one single issue within just one industry. Morula Health simply doesn't gather enough data to make the model learn fast enough.
The majority of tasks AI will solve are going to be similar across many industries. Think about the similarities between medical writing and law, the necessity to cross-check historical reference literature and then make informed conclusions.
Developing our own AI tool would be like us deciding to create our own word processor; what's the point? Just use Microsoft Word, which is used in just about every country and industry in the world for generating text documents. It's not perfect, but it does the job.
The same is true with AI, and Microsoft is rolling it out in Office 365 in the form of Clippy - and we can't wait!
Have you ever seen the gripping film Hidden Figures? If not, it's a must-watch. This remarkable story follows three African American ladies who are mathematicians and human computers at NASA during the space race with Russia. We observe Katherine Johnson solve complex equations needed for a safe re-entry of an Apollo capsule from outer space; we also witness Dorothy Vaughan come to terms with the fact that there is a new IBM machine which can potentially replace her job as one of those "human computers". She teaches herself how to program the new computer, rendering her an invaluable asset as they need her to skilfully operate the computers daily. The movie concludes with Katherine Johnson under immense pressure double-checking the complex equations for the capsule landing which were calculated by the computer.
This was 1961; since then, we have used calculators and Microsoft Excel daily. Thank goodness for them – I don't have to do long division anymore! But the truth is that humans are still necessary in this equation: no matter how advanced these tools get, they cannot tell you what is relevant or provide context; only a human can do that.
ChatGPT (twice) has further demonstrated the imperative need for human guidance when it comes to asking and interpreting questions, evaluating whether something is accurate, and determining if a particular result aligns with desired outcomes.
For Morula Health and Medical Writers, AI should be thought of as a tool to help them with their daily tasks instead of something which will replace them. An ally or friend that not only scans content but also checks references and delivers a succinct summary of data to double-check what’s been written. Our Medical Writers, with their content and creative flair, will always be at the forefront of delivering true value; AI will be there as a tool to check it along the way.
BioGPT is going to contribute further by providing a biomedical research specific language model which will hopefully be integrated into Office 365 - Marie Clippy Curie if you like!
Soon, AI will expedite clinical study reports and patient narratives - "the low-hanging fruit," so to speak. Natural language processing allows us to collect information from sources such as Medwatch forms to generate neatly structured template text for narratives. As our industry is heavily regulated, and data is regularly structured in a specific way, it will make this software possible. From the tests we conducted however, we found that if inputted data was not of sufficient quality more medical writer's hours would be required to amend any mistakes present.
Right now, AI becomes financially advantageous when dealing with many patient narratives with the same inputs and template. Large Pharma's Phase 3 oncology studies that may have thousands of patient narratives are perfect for this AI technology; yet these clinical trials remain the exception and not the rule for the majority of clinical trials taking place today.
To wrap it up, this is an essential point for Medical Writers to take into consideration: AI grows from the data that you input. Every time a related request gets asked, AI can utilize the same information. Remember though, if confidential information is placed in an open-source tool online, then there's no going back—allowing access to such sensitive material could spell trouble like what recently happened with Samsung and their leaked code.
With my current understanding of AI, I recognize that developing a medical writing tool that is successful and reliable requires an expansive amount of information; however, this data must remain confidential. To ensure security, you will have to create a way for the system to learn without making the learned information public - in other words, keeping it within a closed system.
Even if you do manage all that, how do you deal with one of your clients' confidential documents potentially being written using another confidential document from a different client as a data source? I am not too sure how happy clients will be about that.
Disclaimer – The first draft of this blog was 100% human-written. We then enhanced and checked the blog using some AI tools we use daily for internal business operations and medical writing, but even after that it still required further editing and checking to be factually correct.
We would love to hear your feedback and what you think - please email us at info@morulahealth.com