Pakistani Women Innovate AI For Health And Education

Artificial intelligence is not just for fixing your grammar on ChatGPT or turning your pictures into Studio Ghibli style art. It is reshaping the world in a more meaningful way, and three Pakistani women graduating from Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi this year are a part of this shift.

Funded by the UAE government, MBZUAI is the world’s first AI university that specialises in research and graduate level education.

Three women from Pakistan are making science more accessible and developing smarter medical tools here all while proving that in the right hands, AI can do much more than autocomplete your sentences.

Pioneering AI for Early Medical Diagnosis

Twenty three year old Tooba Tehreem Sheikh lost her mother to colon cancer in 2007 because of a late diagnosis, a personal tragedy that went on to fuel her work in artificial intelligence. Sheikh wanted to develop AI systems to diagnose medical problems early on. "Doctors are humans, they can miss a lot of things, so I wanted to develop AI systems that can detect these complex, subtle problems which are often missed by experts," she told Images.

Sheikh graduated with a degree in software engineering and worked in Pakistan for a while before pivoting to research. She enrolled in MBZUAI for a master’s programme in computer vision.

Her work focused on real time AI systems for medical imaging, tools that could make faster and more accurate diagnoses possible, especially in places with limited resources. Her contributions include two models IHA-YOLO, which she described as "a lightweight, real time cell detection system" and Med-YOLOWorld, an "open vocabulary detection system designed to work across nine medical imaging modalities."

Since these systems are lightweight, Sheikh explained, they can run on the equipment many Pakistani hospitals already have. There is no need for additional high end computational resources. That makes them not only fast but also accessible and scalable in settings with limited infrastructure.

Both models take in medical images like CT scans, X rays, MRIs, and even colonoscopy footage and identify both organs and possible diseases. While IHA-YOLO is effective in detecting known classes, MedYOLO-World takes it a step further and identifies novel anomalies that it was not explicitly trained on.

Classes can be thought of as specific categories in a dataset. "Let’s say you have an abdominal CT image. You have liver, pancreas, and spleen. These are classes. Liver is a class. Pancreas is a class," Sheikh explained.

But introducing AI into healthcare does not come without ethical challenges. One major concern, Sheikh pointed out, was data confidentiality. Since AI models rely on large amounts of patient data, they can be vulnerable to breaches if not properly secured. "You need to make sure that you do not expose the patient, the name of the patient, or the details about the patient," Sheikh said.

She does not see her work as a substitute for human expertise but as tools that can aid diagnoses. Sheikh acknowledged that Pakistan still lags behind in healthcare digitisation and eventually plans to return home to establish a hospital where AI is integrated into diagnostic care.

Advancing Multilingual AI to Break Down Barriers

Ashba Sameed, 27, is an AI researcher from Karachi whose work at MBZUAI evaluated how artificial intelligence can better understand and connect spoken language with written text, using both audio and text data.

She tested advanced AI models (Spirit LM and Mini-Omni) on tasks such as summarising, answering questions, and translating between speech and text in multiple languages, including Arabic and English. Spirit LM was more attuned to emotional understanding, while Mini-Omni performed better in tasks based on instruction.

Sameed studied computer systems engineering and developed an interest in AI while working at the Research Centre for Artificial Intelligence in Karachi. After three years in the industry, she went on to pursue natural language processing at MBZUAI.

Her research aimed to improve how machines follow instructions, whether spoken or written. This research is especially useful for Pakistan, where many people speak regional languages and may have limited access to text based technology.

By making voice based AI systems smarter and more multilingual, her work can help build tools like voice assistants, automated helplines, and translation apps that are easier for everyone to use, including those who are not fluent in English or cannot read. This could improve access to education, healthcare, and government services across the country.

Sameed emphasised the need for diverse data while training an AI model. "If I use data from the UAE, and my target audience is Pakistan, there will be issues," Sameed told Images. "So it’s important to use representative data."

Sameed said MBZUAI set her up for the future, and, amongst other things, free coffee on campus is what she will miss the most.

Revolutionizing Science Education with AI

Twenty six year old Fatimah Lyba Khan pursued natural language processing at MBZUAI after studying electrical engineering in Lahore. She built SciGrade, a dataset and AI system that simplifies complex scientific texts across five education levels and 10 disciplines, including biology, chemistry, physics, ecology, and meteorology.

Think of it as a bridge between dense scientific jargon and young students, or even adults, who might not be very familiar with technical terms.

The tool uses large language models (LLMs) to simplify scientific content for five grade bands, from Grade 1 through to college level. Besides simplifying language, Khan explained, SciGrade adapts tone, vocabulary, and sentence structure for each target audience while preserving scientific accuracy. That is the key challenge she had to tackle how to simplify without compromising content.

To make sure the science stays correct, she said she designed specific prompts for each grade level and tested outputs using automatic metrics before they were evaluated by a human.

While SciGrade currently works in English, Khan said she wants to expand its capabilities to other languages such as Arabic. In countries like Pakistan, where access to formal education might be limited, a tool like this would help translate technical information into easily digestible content. So it is essentially not just for schoolchildren but for any curious learner.