LARGE LANGUAGE MODELS FOR CONTACTLESS AI HEALTHCARE MONITORING AND REPORTING

modern approaches to artificial intelligence (AI) represent a paradigm shift with profound implications in critical areas such as healthcare [1]. As an area of research, the term itself encapsulates the study and practice of making 'computers perform intelligent tasks' that were once the sole purview of human actors with a certain degree of domain knowledge and that

cannot be readily encoded in the design of algorithms [2]. The latest research at the intersection of AI and healthcare, as with other domains, leverages the surge in data availability due to further integration of technology into daily life and the incurrence of the 'information age' [2] [3]. As Wartman et al suggest in [3], modern medical practices must welcome the use of

'intelligence tools', and as such, research innovations that aim to solve medical problems with AI will be at the forefront of progress in the field. One of the key areas that can be improved is that of healthcare monitoring and documentation. Aiming to explore the use of a 'shared notes' system between doctors and patients, Bell et al [4] found that almost a third of patients found

errors in the doctor's descriptive notes of their condition. This kind of error is engendered by the fact that such interactions depend on the intermittent relaying of subjective experience by patients to doctors that have little objective data for creating accurate reports. Consequently, the option of using wearable biosensors has generated great interest [5], providing avenues for

data-centric bio-analytics that support informed medical decision making. One of the main issues with wearable sensors discussed in [5] relates to their cumbersome and invasive ergonomics. Research in contactless radio frequency based vital signs monitoring, however, points to the potential for a non-invasive method that can provide persistent wireless data

pipelines for training robust neural networks and other predictive healthcare applications [6] [7]. Generative deep learning techniques that are employed to produce sophisticated language models such as ChatGPT represent one of the greatest advancements in AI technology to date; that of large language models (LLMs) [8]. The capabilities of LLMs are of great interest to

researchers in a range of domains: most notably that of education and medicine [9]. Combining the language generation abilities of LLMs with other AI based advancements such as contactless healthcare monitoring will lay the foundation for a sophisticated non-invasive healthcare analytics, prediction and reporting framework that can act as an objective

intermediary between patients and doctors. The benefits of the proposed system include the focus on objective, data-centric reporting that avoids the subjective nature of doctor-patient dialogues, thus improving report accuracy. Furthermore, such a system can be extended to incorporate continuous and contactless patient-data pipelines that allow for the creation of a

real-time and predictive healthcare platform that emphasises early medical intervention rather than after-the-fact sickcare [10].