7 Things To Expect From AI In Healthcare This Year

The past year was all about artificial intelligence , with a particular focus on its integration into healthcare in our universe. At The Medical Futurist, we have extensively explored how AI is reshaping the healthcare landscape, outlining what to expect and how to prepare for these transformative changes.

As we move into 2024, it's time to continue our forward-looking journey. This year promises to be a blend of consolidation and revolution. Some trends we've previously identified are beginning to solidify and integrate into the fabric of healthcare systems, while others are just starting to unfold.

In this spirit of exploration and anticipation, we present our seven predictions for the 2024 AI in the healthcare scene. These predictions span from the emergence of specialized AI platforms in healthcare to teaching prompt engineering in medical education, indicating a year of significant progress and exciting developments.

1. Generative AI platforms in healthcare

Generative AI has sparked a lot of interest among healthcare companies, particularly in patient management and chatbots. However, it hasn't yet gained substantial trust.

With major healthcare and tech companies launching generative AI platforms for patients, I expect the trust factor to improve throughout 2024. This will likely happen as more people use and become familiar with these platforms.

We've previously covered the (generative ) AI topic, noting that not many innovations mature so rapidly as generative AI has in its first year. This technology is set to transform not only doctor-patient communication but also how patients perceive health issues. In multiple articles , we have discussed the generative AI revolution , analysing many aspects of this remarkable breakthrough technology and how to navigate its implications .

2. Medical large language models will replace ChatGPT in healthcare

We’re expecting to soon see generative AI algorithms that were designed specifically for medical purposes. Such tools will likely have GPT-4-like algorithms under the hood and are set to take the place of general-purpose models like ChatGPT in the healthcare industry.

We have discussed in detail that new AI medical chatbots, such as MedPaLM by Google/DeepMind, will soon surpass the waiting times for a doctor. These models are improving rapidly, and the risk of missing care due to healthcare capacity shortages is soon expected to outweigh the risks associated with algorithmic errors.?

MedPaLM is currently the most commonly known algorithm that was designed for healthcare use. It is a 540-billion parameter PaLM model, which was trained on multiple medical Q&A datasets and can answer healthcare-related questions with impressive accuracy . While not yet available for public testing, it is already available to a select group of customers .?

Recent studies and publications, such as those in Nature and JAMA discuss the potential and limitations of AI in medical diagnosis and treatment.

3. Multimodal large language models (M-LLMs) for hospitals

A critical advancement on the horizon is the arrival of Multimodal Large Language Models (M-LLMs) in healthcare systems. These M-LLMs are poised to act as "central AI hubs", helping doctors and hospitals to step into the AI era.

The significance of M-LLMs lies in their ability to process and interpret multiple types of input data simultaneously, including text, images, audio, and video. This capability marks a significant departure from current medical AI systems, which typically handle only one data type, such as text or X-ray images.?

Medicine inherently requires a multimodal approach, much like human interaction. To diagnose and treat a patient, healthcare professionals must synthesize information from various sources: they listen to the patient, read health files, examine medical images, and interpret laboratory results. This integrated approach is beyond the capability of any current AI technology.

The difference between existing large language models and M-LLMs can be compared to the difference between a runner and a pentathlete. A runner specializes in one discipline, whereas a pentathlete must excel in multiple disciplines. In this analogy, current LLMs are runners, excelling in unimodal tasks, while humans in medicine are akin to pentathlon champions, adept in multiple areas.

We are currently at the dawn of the multimodal era: Med-PaLM’s multimodal version reportedly has advanced capabilities in answering medical questions. Meanwhile, Google’s new Gemini was announced as? "natively multimodal" .

The integration of M-LLMs into hospital systems represents a monumental shift. These will be the ultimate interface between physicians and AI, enabling more comprehensive and nuanced AI assistance in medical diagnostics and treatment.?

4. AI digital twins

AI digital twins, less politely known as deepfake avatars, are rapidly evolving. An interesting example was when Synthesia created the digital Dr. Mesko , demonstrating impressive progress. Since then, there have been significant advancements, such as HeyGen’s AI-based translation service , which not only translates recorded videos into other languages but also synchronizes lip movements to the new language.

While these digital avatars might still seem somewhat uncanny to friends and colleagues and are not yet suitable for patient-facing roles, their potential in healthcare is substantial. One of their most promising uses is in creating educational materials. This technology provides a faster, more cost-effective method for producing content, especially for healthcare professionals seeking new knowledge.

Moreover, these advancements will greatly enhance access to content and care for patients in their own languages.

5. Over 1000 FDA-approved AI-based medical devices?

This year, the number of FDA-approved AI-based medical devices may surpass a thousand. This significant milestone reflects the rapid advancement of AI and ML (Machine Learning) in healthcare.

We recently discussed the current state of nearly 700 FDA-approved AI-based medical devices . The emergence of AI and ML technologies, adaptive algorithms, and generative AI are reshaping the medical landscape. These innovations promise to change how medicine is practiced but also present unique regulatory challenges.

Unlike traditional medical devices, AI and ML technologies can evolve and learn, potentially performing differently in real-world applications than in pre-market testing. This evolution could lead to improved patient outcomes, but it also introduces new risks that require careful management.

The FDA has historically been at the forefront of regulating novel technologies in healthcare, setting global standards. As of our last review, the FDA's database included 692 AI-based medical devices, and as of October 19, 2023, none of these devices utilized generative AI, artificial general intelligence (AGI), or were powered by large language models.

Given the pace of development and the traditional pioneer role of the FDA, other countries are to follow the US regulatory framework.

6. Influx of AI tools for patients

We will see a surge in AI tools designed specifically for patients, encompassing applications such as analysing lab tests, summarizing wearable data, and providing nutritional advice.

Already existing examples include algorithms like Abridge, which translates complex medical terminologies into plain language for patients, and innovative skin-checking apps such as Miiskin , Cube , SkinVision , aysa or Skinive , offering early detection services for skin cancer.?

But the field will extend, and we will see many more applications, like apps helping patients make sense of their lab results .?

7. Forward-looking medical universities will start teaching prompt engineering

The discipline of prompt engineering is gaining such importance in this new realm of AI that it is expected to become a subject taught at forward-looking medical universities.

Prompt engineering is both an art and a science, essential for interacting efficiently with generative AI models. It involves the strategic formulation of inputs to elicit the best possible responses from AI. If you are not yet familiar with the concept, we summarized our 11 prompt engineering tips , outlining the basics of how to give effective prompts to guide AI models like ChatGPT in producing desired outputs.?

This skill is not just about using tools effectively; it's about understanding the capabilities and limitations of AI, such as the risk of generating answers from unreliable sources or failing to provide sources at all. As AI models are becoming indispensable tools, the mastery of prompt engineering is becoming a crucial skill - for medical professionals as well.

As AI continues to grow and evolve , becoming adept at prompt engineering is increasingly vital. This skill empowers users to maximize the potential of AI, turning it into a tool for expanding knowledge and ideas rather than just a means of solving problems. And as it is becoming so fundamental, it should be integrated into the medical curriculum.