Made in Kazakhstan: Building an AI for a Nation
On a cold November morning at Al-Farabi University in Almaty, students gathered in a drafty lecture hall, many still wrapped in their coats. The setting was more reminiscent of a forgotten Soviet-era classroom than a venue for cutting-edge technology. But amid the peeling paint and rickety seats, some of the country’s most ambitious young researchers had come to discuss Kazakhstan’s latest steps into the world of artificial intelligence. The star billing came from the Institute of Smart Systems and Artificial Intelligence (ISSAI) at Nazarbayev University in Astana. Last year, the institute released KazLLM, its first Large Language Model (LLM), to much fanfare, inspired by a philosophy of building AI systems that understand the country’s language and culture rather than borrowing second-hand from Silicon Valley. But can Kazakhstan keep pace in the global AI race? And despite the government’s efforts to back local products, can it convince the population to use them over Western alternatives? Recent developments The Institute’s founder, Doctor Huseyin Atakan Varol, was keen to stress that steps have been taken to develop Kazakhstan’s native AI ecosystem over the past twelve months. “Since the release of KazLLM last year, we have witnessed what I would describe as a ‘Cambrian explosion’ of generative AI development,” he told The Times of Central Asia. “The KazLLM project enabled us to create the team and amass the know-how to build a new generation of multilingual and multimodal models tailored to Kazakhstan’s needs.” Among these, he lists Oylan, a multimodal language–audio-vision model; MangiSoz, a multilingual speech and text translation engine; TilSync, a real-time subtitle and translation engine; and Beynele, a text-to-image generation model. All these models have been fine-tuned to better reflect Kazakh culture and linguistic norms. “In short, we are building AI made in Kazakhstan, by Kazakhstani youth, for Kazakhstan –models that understand the language, culture, and needs of the people,” said Amina Baikenova, ISSAI’s Acting Deputy Director of Product and External Affairs, in an interview with TCA. [caption id="attachment_39061" align="aligncenter" width="1600"] The old lecture hall at Al-Farabi University, Almaty; image: TCA, Joe Luc Barnes[/caption] Much of this progress stems from the enthusiasm of a generation of students, whom Kazakhstan has invested heavily in training. Indeed, the country has become a magnet for young researchers from across Central Asia. “After completing my bachelor’s degree in Kyrgyzstan, I was looking for opportunities to build my research career. That’s why I moved to Kazakhstan,” said Adam Erik, an ISSAI student from Bishkek. “Kazakhstan has become a scientific center of Central Asia.” Erik believes strongly in building local language models. “There is a thing called bias in data sets,” he said. “Models from the U.S., China, or Europe are incredible, but they’re trained mostly on Western culture and literature. Local solutions are still necessary.” These sentiments reflect a common frustration among researchers: even the best global AI systems stumble when asked about Kazakh idioms, rural social norms, or local history. The data used to train the world’s most powerful models rarely includes more than a sliver...
