Mohamed El-Nemr, BMBCh, MA(Oxon), MRes, PGCert
Clinical and Research Fellow in Plastic Surgery, Bradford Royal Infirmary, UK
MD Candidate, University of Bradford, UK
The use of Artificial Intelligence (AI) in daily life, healthcare, academia and businesses has skyrocketed in the past few years, raising many questions on how this subset of technology will advance.1 As with any technology, it is essential to be aware of its inner workings to reliably critique and evaluate what benefits it brings and what ethical frameworks are required for its use. This paper will only focus on the use of Large Language Models (LLMs), which underpin popular AI systems such as OpenAI’s ChatGPT and Google’s Gemini and will evaluate these LLMs from the perspective of foundational Islamic ethical principles, analysing their function and impact through this lens.
AI is a multidisciplinary field at the intersection of computer science, mathematics, engineering, cognitive science, and linguistics. Its aim is the development of computational systems that can perform functions typically requiring human intelligence.2,3 For a machine to be considered intelligent, it must be able to perceive, represent, and interpret its environment; acquire and apply knowledge; reason with symbolic and statistical methods; learn from experience; adapt to new contexts; and act autonomously or interactively to achieve specific goals.2,3 It is well accepted that in its current state, all forms of AI available are so-called “Narrow-AI”, which is to say they are highly specialised systems that can produce excellent results in specific tasks, such as image recognition, language generation and data prediction.4 AI is currently not capable of applying knowledge across multiple domains, common sense reasoning, problem-solving or creativity (which would define it as Artificial General Intelligence).
All our current models are, at their essence, statistical models (albeit very complex) that provide the most likely output from a learned set of inputs and outputs.2,3 In commercial contexts, it is easy to criticise companies for presenting their products as ‘artificial intelligence’ without the caveat that they are, for now, narrowly specialized language generators.
The danger with the current state of LLMs is how advanced they have become in recognising and repeating human linguistic patterns and recreating them from any prompt from the user.4 This masks the fundamental limitation that they do not truly understand the input nor their own outputs, instead generating a statistically probable response. This creates a significant ethical challenge related to the Islamic principle of Sidq (Truthfulness).
There have been several recent incidents of this misplaced trust and over-reliance causing real-world issues. For instance, in 2023, a New York lawyer faced sanctions after he submitted a legal brief that cited several non-existent court cases that were completely fabricated by an LLM.5 This “hallucination” is a form of statistical falsehood (kadhib), and the user’s uncritical acceptance was a failure to uphold the Amanah (Trust) of their profession.
This potential for harm (mafsadah) becomes more severe in personal interactions. There have been two recent tragic cases where individuals took their own lives after confiding in an AI chatbot for months, with concerns that the LLM allegedly provided inappropriate and harmful responses that failed to encourage the users to seek urgent help.6,7 From the perspective of the Maqasid al-Shari’ah (Objectives of Islamic Law), this represents a profound failure to adhere to Hifz al-Nafs (Protection of Life) and Hifz al-‘Aql (Protection of the Intellect). An algorithm that cannot discern the severity of human distress and fails to provide compassionate, human-centric guidance, instead potentially reinforcing harmful ideation, is a clear violation of these fundamental objectives.
Another significant ethical concern lies in how these powerful models are created which challenges the principle of Adl (Justice). Most LLMs are trained on vast datasets scraped from the internet, often without the consent, credit, or compensation of the original creators. This practice raises serious questions about copyright infringement, is a form of zulm (injustice or oppression), as it expropriates the work of countless artists, writers, and programmers to build a commercial product. Furthermore, this data reflects the existing biases found online, meaning the AI can unintentionally learn and perpetuate harmful stereotypes related to race, gender, and culture. This can lead to outputs that are unfair, discriminatory, or offensive, embedding societal prejudices into our technology.9,10 This is not merely a technical flaw; it is an ethical failure. By embedding societal prejudices, the technology actively contravenes the Qur’anic imperative to “stand firm for justice” (Qur’an 4:135), thereby automating and scaling injustice.
To combat these issues, a multi-faceted approach grounded in technical knowledge and strong ethical principles is necessary. This technology must be developed in a way that adheres to timeless Islamic principles. This requires more than simply listing values; it demands their active implementation.
- Adl (Justice): demands proactive measures to de-bias training data, ensure equitable outcomes, and establish fair compensation models for creators whose work is used.
- Sidq (Truthfulness): requires a technical and user-facing commitment to transparency about a model’s limitations and robust defences against fabrication.
- Maslaha (Public Welfare): and the prevention of mafsadah (Harm) necessitate mandatory human oversight in all critical applications, especially those concerning mental and physical health (upholding hifz-al-nafs).
- Amanah (Trust/Accountability) must be the cornerstone. Developers are entrusted with building non-harmful tools, and users are entrusted with verifying information, linking accountability back to human agency.
All technology has the potential to create benefit (maslahah) or cause harm (mafsadah), and AI is not an exception to this rule. Users and developers are accountable for verifying information, not blindly accepting AI “truths”, and demanding fairness. The principles of Islamic ethics are not merely a supplemental checklist; they provide a comprehensive framework (manhaj) that grounds technological development in divine purpose. This framework, which champions justice (‘Adl), truth (Sidq), and the preservation of human dignity (karamah), along with our unique, God-given capacity for ‘aql (higher-order reasoning), must guide our engagement with this technology.
References
- Borges, A. F. S., Laurindo, F. J. B., Spínola, M. M., Gonçalves, R. F. & Mattos, C. A. The strategic use of artificial intelligence in the digital era: Systematic literature review and future research directions. Int J Inf Manage 57, 102225 (2021).
- Russell, S. J. Artificial Intelligence: A Modern Approach (Pearson, 2020).
- Goertzel, B. Artificial General Intelligence: Concept, State of the Art, and Future Prospects. Journal of Artificial General Intelligence 5, 1–48 (2014).
- Fundamental limitations of generative LLMs. library.oapen.org
- Merken, S. New York lawyers sanctioned for using fake ChatGPT cases in legal brief | Reuters.
- What My Daughter Told ChatGPT Before She Took Her Life – The New York Times.
- Parents of teenager who took his own life sue OpenAI – BBC News. https://www.bbc.co.uk/news/articles/cgerwp7rdlvo
- Henderson, P. et al. Foundation Models and Fair Use. Journal of Machine Learning Research 24 (2023).
- Birhane, A., Prabhu, V. U. & Kahembwe, E. Multimodal datasets: misogyny, pornography, and malignant stereotypes. (2021).
- Hofmann, V., Kalluri, P. R., Jurafsky, D. & King, S. AI generates covertly racist decisions about people based on their dialect. Nature 2024 633:8028 (2024).
- Barredo Arrieta, A. et al. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion 58, 82–115 (2020).
- Elmahjub, E. Artificial Intelligence (AI) in Islamic Ethics: Towards Pluralist Ethical Benchmarking for AI. Philos Technol 36, 1–24 (2023).
