Comparison of Generative AI and Predictive AI

Artificial Intelligence (AI) is the “science and engineering of creating intelligent machines that can achieve our goals like humans through a series of technologies.” The term was introduced in the 1950s, and there were high expectations for its rapid widespread application and significant scientific breakthroughs. However, due to limitations in computation, analysis, and data, progress in AI has been slow. Over the past 70 years, the environment surrounding these challenges has changed significantly, bringing new opportunities. Among these are two AI technologies that will be seen in nursing practice: predictive AI and generative AI. Predictive AI uses data (such as patient data) to predict future events or trends. Generative AI learns from existing data to analyze and generate new data based on that information.

Just a few years ago, researchers primarily focused on predictive AI, which uses specific datasets to find solutions for clinical and administrative issues, such as early identification of patient deterioration for timely intervention. The introduction of ChatGPT, a type of generative AI, has shifted much research effort towards generative AI. Generative AI has advantages in creativity, and the datasets it uses are much larger than those of predictive AI—billions or even hundreds of billions of data points—combining and synthesizing information from multiple sources through specific questions or prompts, thereby increasing the efficiency of information gathering in healthcare. Recently, ChatGPT has also lifted the restriction of only including data up to September 2021; it can now access the past and current internet without limitations, thus providing real-time, authoritative information and now can link with data sources. Despite their respective advantages, it is noteworthy that predictive AI and generative AI have different uses.

In many industries, including healthcare, the availability, accessibility, and analysis of structured data make predictive AI easier and more useful than generative AI. For example, predictive AI uses structured data relevant to target outcomes to make predictions that can inform clinical decision-making.

For predictive AI and generative AI used in nursing practice, the most important factor is accuracy. Every innovation has its downsides. Similarly, predictive and generative AI tools are also susceptible to biases and errors. Although these issues are being addressed, we have not yet reached an ideal state. This highlights the important role that nurses play in using AI: AI supports nursing practice, but it is nurse-led.

Researchers in North Carolina assessed its effectiveness by comparing Epic’s AI sepsis tool with existing sepsis prediction tools, systemic inflammatory response syndrome, organ failure assessments, and sepsis-related organ failure assessments. The study indicated that the Epic tool was more accurate at the highest scoring threshold but performed poorly in the timeliness of sepsis predictions. The study stated, “It seems to predict sepsis long after clinicians have identified a possible case and acted on it.”

In 2022, after changing data variables, redefining the criteria for sepsis, and adjusting algorithms for local patients, Epic released a redesigned sepsis prediction model.

The ongoing case of the Epic sepsis prediction model serves as a cautionary tale, not unique to Epic, and provides important learning opportunities in the early stages of using artificial intelligence in clinical settings. This tool is designed to support clinicians in the early identification and intervention of sepsis, which is a major factor in high mortality and costs in U.S. hospitals. The survival rate of sepsis depends on early detection and treatment. Rigorous research and publication in peer-reviewed journals are essential to ensure the reliability and effectiveness of AI tools before their release.

In this study, ChatGPT generated more detailed documentation based on word count, averaging 135 words, while dictation averaged 89 words and typing averaged 67 words. ChatGPT’s speed was reported as moderate, with an average time of 69 seconds, but this did not show significant differences compared to dictation. ChatGPT achieved the highest PDQI score of 35.6, followed by dictation (31.6) and typing (30.4).

It is noteworthy that ChatGPT also generated erroneous information in 4 out of 11 documents (36%). These errors included the addition of physical examination results not included in the patient’s history and details of two patients’ history not recorded. This information was unrelated to the patient or elements of HPI. It was created by ChatGPT, a so-called hallucination, and therefore was inaccurate, raising concerns about the reliability of ChatGPT in HPI documentation.

Generative AI can help us find new solutions to old problems, such as defining nursing workflows that we have yet to conceive. Predictive AI can enhance nurses’ capabilities by supporting data-driven decisions, such as early interventions to prevent patient deterioration. Both have the potential to derive greater value from data. As AI evolves, the distinction between these two types of AI may gradually disappear—systems have been developed that integrate both predictive and innovative applications into a single AI tool, further enhancing the value of data and the science and practice of nursing.