📄 medicine

The Impact of Artificial Intelligence (ChatGPT) on Medical Decision-Making in the Healthcare System of Afghanistan

This study of 42 medical professionals in Kabul reveals that while ChatGPT is increasingly adopted by early-career physicians in Afghanistan to address resource shortages and aid patient education, its integration into clinical decision-making is tempered by significant safety concerns regarding AI hallucinations, necessitating strict human oversight and clear guidelines.

Original authors: Mohammad Reza Cov Andish, Mustafa Muradi, Ahmad Sorosh Hamdard

Published 2026-10-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mohammad Reza Cov Andish, Mustafa Muradi, Ahmad Sorosh Hamdard

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the quiet corners of a doctor's office, the tools of healing have always been a mix of human experience, formal education, and the latest medical books. For decades, a physician's ability to make the right choice for a patient depended on how well they could recall facts and how much time they had to consult their references. Today, a new kind of tool has entered the room: artificial intelligence. Specifically, large language models are computer programs trained on vast amounts of text that can hold a conversation, answer questions, and summarize complex information in seconds. These systems, often known by names like ChatGPT, act as a digital library that never sleeps. They promise to be a powerful aid for doctors, offering instant access to medical knowledge that might otherwise take hours to find. But the question remains: can a machine that generates text based on patterns be trusted with the life-and-death decisions of a human patient, especially in places where the healthcare system is already stretched thin?

This question is not just theoretical; it is a matter of urgent reality in Afghanistan. The country's healthcare system faces severe challenges, including a shortage of specialist doctors, a lack of up-to-date medical resources, and the lingering effects of years of instability. In this environment, a doctor in a small clinic might be the only medical expert for miles, forced to make difficult diagnoses without the support of a senior colleague or a comprehensive drug database. To understand how these doctors are navigating this gap, researchers from Kabul University conducted a study to see if artificial intelligence is being used in their clinics, how well it works, and whether it is safe. They focused on the practical reality of the situation, moving beyond the idea of what technology could do to examine what it is actually doing on the ground.

The researchers surveyed 42 medical professionals working in public and private clinics in Kabul. This group included doctors, surgeons, and specialists, with a notable trend: the majority of the participants were young, with nearly two-thirds having fewer than six years of experience in their field. These early-career doctors are the ones turning to artificial intelligence most frequently. They are using these tools to look up general medical information, interpret complex lab test results, and check for potential side effects of medications. For many, the technology has become a primary way to fill the information gaps left by the lack of physical textbooks or access to international medical journals.

When the doctors evaluated the performance of the artificial intelligence, a clear picture emerged of its strengths and its dangerous weaknesses. The tool was rated highly for its ability to explain medical concepts to patients. Doctors found that the computer could take difficult, technical terms and translate them into simple language that a patient could easily understand, making it an excellent resource for patient education. However, the mood shifted dramatically when the conversation turned to safety. When asked about the reliability of the advice given for actual treatment, the doctors expressed significant concern. Specialists, particularly surgeons and those who treat internal diseases, warned that the computer can make up facts. In the world of artificial intelligence, this is known as a "hallucination," where the system generates a response that sounds confident and logical but is completely false. In a medical setting, this could mean a wrong drug dosage, a fake medical study, or advice that ignores a patient's specific history.

The study found that while the artificial intelligence often followed general global medical rules, it frequently suggested treatments or medications that simply do not exist in Afghanistan. A doctor might receive a perfect recommendation for a drug that is unavailable in local pharmacies, leaving them with a plan they cannot execute. Furthermore, the computer cannot see the patient; it does not know a patient's exact weight, their physical appearance, or the specific conditions of their local environment. Because of these risks, the doctors gave the lowest scores to the safety of the recommendations. They made it clear that while the tool is useful for gathering information, it is not ready to make medical decisions on its own. The more complex the medical case, the less the doctors trusted the machine.

Despite these serious reservations, the overall outlook among the medical community remains cautiously hopeful. Nearly 60 percent of the doctors surveyed believe that artificial intelligence will ultimately improve the healthcare system in Afghanistan. Their optimism does not come from a blind faith in the technology, but from a desperate need for information. In a system where a doctor might not have access to a specialist for days, the ability to get a quick answer from a computer feels like a massive upgrade. However, the researchers emphasize that this hope comes with a strict condition: the technology must be used as a backup tool, not a replacement for human judgment. The study concludes that for artificial intelligence to be safe and effective in Afghanistan, there must be clear rules in place. Doctors need to be trained to use these tools carefully, always double-checking the computer's suggestions against their own knowledge and the realities of their local clinics. Without such oversight, the very tool meant to help could lead to dangerous errors. The path forward is not to reject the technology, but to integrate it with a human hand firmly on the wheel.

Technical Summary: The Impact of Artificial Intelligence (ChatGPT) on Medical Decision-Making in the Healthcare System of Afghanistan

Problem Statement
The integration of Large Language Models (LLMs) like ChatGPT into clinical practice presents a dual reality: while these models offer dynamic Clinical Decision Support Systems (CDSS) capable of analyzing vast medical literature instantly, their deployment in resource-constrained environments remains under-researched. In Afghanistan, the healthcare system faces severe challenges, including a shortage of specialists, limited access to updated drug databases, and decades of instability. While general practitioners in this setting increasingly rely on AI as a primary reference to bridge information gaps, the technology's core design as a predictive text generator—rather than a verified fact retriever—introduces significant risks. Specifically, the potential for "hallucinations" (plausible but incorrect medical advice) poses a direct threat to patient safety, particularly when used without the regulatory oversight or backup verification systems common in high-resource nations. There is a critical lack of empirical data regarding how Afghan medical professionals actually utilize these tools, their perceived accuracy, and the safety limits of such deployment in a developing nation.

Methodology
This study employs a cross-sectional, quantitative research design to gather primary data from active medical practitioners in Kabul, Afghanistan.

Population and Sampling: The study utilized purposive sampling to recruit 42 medical professionals (doctors, surgeons, and specialists) from both public and private clinics in Kabul. Participants were required to hold a medical degree and possess basic experience with AI tools. The sample skewed toward a younger workforce, with 66.7% having less than six years of clinical experience.
Data Collection: A structured questionnaire was distributed via physical copies and Google Forms. The instrument consisted of two sections: demographic profiling (age, gender, specialty, experience) and an evaluation of AI performance.
Evaluation Metrics: Participants rated ChatGPT on a 5-point Likert scale (1 = Strongly Negative/Unsafe to 5 = Strongly Positive/Safe) across four clinical pillars:
1. Clinical Appropriateness: Validity of advice for specific symptoms.
2. Usefulness to the Patient: Ability to explain medical issues simply.
3. Safety of the Recommendation: Risk of harm from following the advice.
4. Alignment with Clinical Guidelines: Adherence to established medical rules (e.g., WHO protocols).
Statistical Analysis: The authors aggregated Likert scores using sample means ( $\bar{X}_m$ ) and standard deviations ( $\sigma_m$ ) to measure agreement levels. An overall Predictive Impact Ratio (IR) was calculated to quantify the perceived future impact of AI on the healthcare system.

Key Results

Adoption Patterns: Adoption is highest among early-career physicians (under six years of experience). The primary use cases identified were retrieving general medical information (over 80% of respondents), interpreting complex lab results, and checking drug side effects. A smaller subset uses the tool for generating initial diagnostic ideas.
Performance Divergence: A distinct gap exists between the tool's utility for education and its safety for clinical decision-making.
- High Scores: ChatGPT received high marks for "Usefulness to the Patient," effectively translating complex medical terminology into simple language for patient education.
- Low Scores & Concerns: The "Safety of the Recommendation" metric received the lowest scores and the highest variance. Specialists, particularly surgeons and internists, expressed significant reservations regarding AI hallucinations in complex diagnoses and treatment planning. They noted the model's inability to account for physical variables (e.g., exact patient weight) or local constraints (e.g., drug availability in Afghanistan).
Guideline Alignment: While the model generally aligns with international protocols like those from the WHO, it frequently recommends treatments or medications that are unavailable in the local Afghan context.
Future Outlook: Despite safety concerns, 59.5% of surveyed doctors believe AI will positively impact the Afghan healthcare system by filling critical information gaps. Conversely, 21.4% hold a mixed view, conditional on the enforcement of strict usage rules, while only 4.8% predict a negative impact.

Key Contributions
This paper provides the first quantitative field survey of AI usage among active medical personnel in Afghanistan. Its primary contributions include:

Empirical Data: It establishes a foundational dataset regarding the frequency of AI use, perceived accuracy across specialties, and safety limits in a conflict-affected, resource-limited setting.
Risk Identification: It empirically validates the "automation bias" risk, highlighting that while AI is a powerful search engine, its application in critical decision-making (prescriptions, complex diagnoses) is currently unsafe without human oversight.
Contextual Nuance: The study distinguishes between the tool's success in administrative and educational tasks versus its failure in high-stakes clinical scenarios where local resource constraints (drug shortages) are not factored into the model's training.

Significance and Claims
The authors claim that while ChatGPT serves as a valuable, low-cost information tool for bridging gaps in Afghanistan's healthcare infrastructure, it is not yet ready to function as an autonomous decision-maker. The paper concludes that the medical community's optimism is driven by the severity of information scarcity rather than blind trust in the technology. Consequently, the authors argue that safe integration requires strict human oversight and the development of clear, localized clinical guidelines to prevent medical errors. The study positions itself as a necessary step toward building a national framework for digital health, emphasizing that future policy must address the distinction between using AI as a "backup tool" versus a "replacement for medical judgment."

Technical Summary: The Impact of Artificial Intelligence (ChatGPT) on Medical Decision-Making in the Healthcare System of Afghanistan

More like this