How can the technology behind tools like ChatGPT help in healthcare of the future?

Dr Pavithra Rajendran is a data scientist at GOSH DRIVE who works on natural language processing (NLP), the same type of artificial intelligence as ChatGPT. In this article, she explains what NLP is and how it could be used in healthcare. 

Natural language processing (NLP) is a type of artificial intelligence (AI) that aims to understand and interpret language. This technology has been developed over decades and has recently received a flurry of interest due to the launch of ChatGPT, an AI programme that can provide conversational answers to complex questions.

How does NLP work?

Some NLP technologies, like ChatGPT are built using large amounts of data that takes lots of computer power. These are called neural networks and large language models. These models are very advanced in their abilities to process written text and are already used widely.

The availability of several freely available NLP technologies on platforms like HuggingFace has accelerated research and development in NLP and help data scientists build new high-quality NLP systems.

However, it is always worth exploring if simpler NLP solutions can get the same results, as these take less computer power to run and the ways they arrive at answers are more transparent. As a data scientist I consider how any NLP will be used and the trade-off between a technology’s complexity and performance.  

For example, if a project does not require any understanding of sentiment in the language – it doesn’t need to work out if the user is a pleased customer or a disgruntled user, a simple rule-based approach may be sufficient, where users get pre-programmed answers to questions.

What does this mean for healthcare? 

Electronic Health Records (EHRs) are used worldwide and securely hold patient information like patient diagnoses, medications and dosage. However, often, the information is unstructured. This means data is buried in reports, patient notes, letters etc, not in a searchable database.

On way that NLP technologies can help is by extract information from these EHRs in a structured way to enable better analysis. This can really benefit the accuracy and speed of healthcare decisions, without compromising data security. This is a future we hope to see, but there will be many steps to make this happen.

What needs to be done next?

Governments, healthcare services and companies are increasingly investing in AI technologies to address challenges for their use in healthcare. These include:

• A lack of data that is sorted in a comprehensive and shared systematic way across healthcare systems. It is important to address this as AI needs lots and lots of information to provide accurate and fair answers.
• The need for better ways to understand if the technology is helping us enough to justify its use so we can start using beneficial technologies more quickly.
• Deciding ways to monitor these technologies in the real-world, just like other processes or devices that are used in healthcare.

As part of the GOSH Digital Research Environment team, I hope to help tackle these issues in the coming years so that clinicians can better utilise the huge amount of digital health information available to guide patient care.

You can find out more about the team on this website.