---
title: "Can We Measure the Iman Impact of AI? Can We Increase It?"
date: "2026-09-05"
speaker: "Dr M Waleed Kadous"
venue: "IQRA' Monthly Meeting (iqra.study), live on Zoom"
type: "talk"
canonical: "https://iaser.ai/talks/iman-impact-of-ai"
slides: "https://iaser.ai/talks/iman-impact-of-ai/slides.pdf"
summary: "Most Muslims will never use a purpose-built Islamic AI. They will use ChatGPT, Claude, or Gemini. So what happens to a Muslim's iman when they sit with one? JaleesBench measures exactly that: 140 disguised dilemmas drawn from Riyad al-Salihin, six kinds of pushback, and a score from the hadith of the perfume-seller and the blacksmith. The talk covers what the benchmark found across 12 systems, why a single line of context does more than a whole guide, and how one prompt change lifted a real assistant to the top of the table."
---
# Can We Measure the Iman Impact of AI? Can We Increase It?

*JaleesBench, and what it tells us about the AI most Muslims already use*

## The key question

> When a Muslim sits with an AI, what rubs off on them? And can we make it better?

## What this talk covers

A 25-minute talk delivered at the IQRA' Monthly Meeting on 5 September 2026, hosted by Dr Safi Kaskas. It moves through seven parts:

1. **Motivation.** Three prongs for Muslim community AI use, and why the third (most Muslims will use frontier models) makes benchmarks matter for everyone.
2. **What happens** when a Muslim goes to an AI: two real scenarios, cutting off a mother and borrowing an orphan's money, and how the models answered.
3. **JaleesBench.** The jalis of the hadith, 140 scenarios from Riyad al-Salihin, six pressures, three framings, and a five-band score judged against each scenario's own proof texts.
4. **Results.** The scorecard across 12 systems, why recognition beats instruction, and where everyone caves.
5. **Impact.** One line added to Ansari's prompt, and the companion prompt anyone can paste into ChatGPT, Claude, or Gemini.
6. **Future work.** JaleesWeights, open models fine-tuned so Islamic values are the default, and why scripture must be routed through retrieval.
7. **Q and A.**

## Follow up

- **The paper:** [arxiv.org/abs/2608.07508](https://arxiv.org/abs/2608.07508)
- **The results browser:** [s.iaser.ai/jb](https://s.iaser.ai/jb), every sitting from the benchmark
- **The prompt:** [s.iaser.ai/prompt](https://s.iaser.ai/prompt), in four languages
- **The code:** [github.com/iaser-ai/jaleesbench](https://github.com/iaser-ai/jaleesbench)
- **The research article:** [JaleesBench: Are AI Assistants Good Spiritual Company?](https://iaser.ai/articles/jaleesbench-are-ai-assistants-good-spiritual-company)