# Standardized clinical assessments and advanced AI-driven instruments used to evaluate neurofunctional deficits, including within biomarker based framework, in Parkinson’s disease - human intelligence made vs. AI models - systematic review

## TL;DR

This paper compares human and AI capabilities in conducting systematic reviews of Parkinson’s disease assessment tools, finding that humans still outperform AI models like ChatGPT.

## Contribution

The study evaluates the current limitations of AI in performing systematic literature reviews and highlights the superior performance of human intelligence in this task.

## Key findings

- Human intelligence outperformed ChatGPT 4.o and ChatGPT Scholar in conducting systematic literature reviews.
- AI models provided inconsistent and often inappropriate responses when queried about Parkinson’s disease assessment tools.
- The paper emphasizes the need for continued improvement in AI capabilities for systematic review tasks.

## Abstract

Considering the extensive development of artificial intelligence (AI) facilities, like Generative Pre-Trained Transformer (ChatGPT) 4.o and ChatGPT Scholar, we explored their abilities to conduct a systematic literature review. Using as a specific domain, an attempt to frame/methodize clinical assessment instruments used to evaluate neuro-functional deficits in Parkinson’s disease (PD) – including framed through the ICF(-DH) paradigm – for the above-mentioned comparison between human intelligence (HI) and AI, this paper is as well, a follow-up regarding the most actual subject matter of the AI’s capabilities evolution in this respect. As well-known clinical-/paraclinical-/functional evaluations, using assessment quantitative (as much as possible) instruments, are basic endeavors for rehabilitation, as they enable setting of appropriate and realistic therapeutic-rehabilitative specific goals.

Within the actual work, we have first achieved a narrative synthesis of the main molecular mechanisms involved in PD pathophysiology, underpinning its clinical appearance and evolution. To fundament our knowledge on an up-to-date information regarding the clinical-functional evaluation tools practiced in PD, we systematically reviewed the literature in this domain, published in the last 6 years, through a PRISMA type method for filtering/selecting the related bibliographic resources. The same keywords combinations/syntaxes have been used contextually, also to dialogize with ChatGPT4.o and ChatGPT.

Scholar Applying PRISMA type methodology (HI achieved), we have selected, matching the filtering criteria, 24 articles. Interrogating the two AI above-mentioned models, we obtained quite difficult to be availed/useful – comparative to our HI obtained – outcomes. Thus, when interrogating ChatGPT4.o, ChatGPT Scholar repeatedly, they provided - partially diverse - inappropriate related answers, including ones pending on the interrogator’s IP, although they claimed to have this capacity.

We consider, regarding their capabilities to achieve systematic literature reviews, that neither ChatGPT 4.o nor ChatGPT Scholar still cannot succeed this (yet, they keep improving lately). Additionally, we have consistently extended, including within a narrative related literature review, our ‘dialogue” with these two AI facilities regarding their availability to enhance the related evaluation instruments accuracy on neurofunctional assessments within biomarker-based frameworks. So, our research aimed basically to emphasize the main topical data regarding these two important paradigms of knowledge (based on HI and on AI) acquirements – considering the impetuous development of the latter – and thus, possibly to contribute inclusively at improving the actual performances to achieve Systematic Literature Reviews through the PRISMA type method – for the moment still better served by HI.