Natural language processing (NLP), the ability for computers and humans to communicate in a more human fashion, has come to the masses via Large Language Models (LLMs), a chatbot powered by artificial intelligence (AI) and developed by OpenAI. The tool proved so popular, it attracted more than a million new users in days—a benchmark that took the most active social media platforms months to reach.

While ChatGPT works as a replacement search engine or writing tool, the underlying technology, generative AI, has far more applications. Generative AI tools will transform jobs and industries, and will revolutionize the way healthcare is delivered and experienced.

Key Differences Between Current Healthcare Apps and NLP

Generative AI is a type of “unsupervised learning,” where the AI generates new data, images, and text that are similar to existing examples in its data set. Unsupervised learning is a form of AI where the machine finds patterns and relationships among the data independently. The intelligence created by generative AI models is impossible for any human or decision-tree framework to replicate.

Decision-tree models, like the ones that power patient engagement and clinical decision support, are currently used in many healthcare applications. They are hard-coded, however, limiting application options to the knowledge embedded in the branches created by the application developers.

Existing generative AI models have been trained on consumer internet data and lack the specificity required for healthcare. To generate accurate insights, the healthcare industry needs a generative AI model trained on healthcare data.

AI Model Can Make a Patient’s Journey More Efficient

Generative AI models can make a patient’s journey far more efficient and convenient. Patients could interact directly with generative AI-based chatbots to obtain accurate answers for lower acuity symptoms as a first-line triage. The chatbots would respond with targeted questions and then make highly personalized recommendations. As patients’ first interface with the healthcare system becomes more convenient and less stressful, they will access care more frequently. Chatbots can also guide patients through their care journey, improve post-discharge adherence, and help them manage chronic diseases, leading to better clinical outcomes and lower costs.

AI-based models trained on data from multiple health systems can significantly lower the cognitive burden on physicians as they work to diagnose and treat patients. With generative AI-based models, physicians could quickly review summaries of prior visits without having to review a patient’s past medical records. This technology is able to create accurate care pathways for a multitude of disease states so physicians can recommend effective care for patients based on their specific conditions. This could significantly reduce clinical variability among physicians by increasing adherence to guidelines-based treatment. In this way, doctors would be able to see more patients without experiencing burnout or degradation of care.

Generative AI Model Can Reduce Administrative Burden, Costs

In addition to improving clinical outcomes, a generative AI model can reduce the administrative burden and costs on hospitals and practices, while also improving accuracy. For example, it can replace the labor-intensive process of manually creating clinical summaries, coding, and billing. Using advancements in automatic speech recognition technology, provider-patient communication can be converted into clinical summaries.

This new process would eliminate human errors in the revenue cycle, improve cash flow, and reduce record-keeping costs, streamlining the entire revenue cycle. Generative AI models can also facilitate other administrative processes, such as eligibility verification and prior authorization.

Although the transformative potential of NLP in healthcare is clear, there are barriers to achieving wide-scale adoption. First, it’s imperative to train the model on data sets covering diverse populations or the outcomes will be biased. Second, the expense of training the LLMs is steep. While these costs are expected to come down as the technology grows, there’s a need for innovative commercial models to enable wide-scale adoption.

One of the most challenging barriers to overcome is the reluctance of physicians to trust the model’s diagnoses and clinical recommendations. Regulators can help by validating these models. The FDA has already created smoother regulatory pathways for clinical AI algorithms through the Software as Medical Device classification.

It’s just a matter of time before the FDA begins regulating generative AI models and their applications for clinical decision support.