Arabic AI developers struggle with unique challenges

• Problems with Arabic language processing
• Regional spending on AI to hit $12bn
• But companies lack AI strategy

As billions of dollars are thrown at developing Arabic versions of AI for businesses, the technology is improving, but it is an uphill struggle, experts have told AGBI.

Spending on artificial intelligence in the Middle East, Turkey and Africa is expected to reach $5 billion this year and rise 35 percent a year to almost $12 billion in 2028, the global technology market intelligence company IDC has predicted. 

Today, six out of 10 organisations in the Middle East identify AI as their top investment, according to IDC.

“Large language models in non-English languages are not the easiest to develop,” Vasudha Khandeparkar at the business consultancy Grant Thornton in the UAE told AGBI. 

• Adnoc signs $340m AI contract with Abu Dhabi’s AIQ

• AI, geopolitics and the Mena opportunity

• Tech sell-off deepens as DeepSeek triggers AI rethink

“There are nuances in sentence structures, variations in dialects regionally and a really rich history which remains largely under-digitised.”

In one example, researchers at the University of Jordan, in a paper published last year by Springer Nature’s journal BMC Infectious Diseases, assessed popular AI models’ performance in English and Arabic for infectious disease queries. They found that English queries showed consistently superior performance to their Arabic counterparts.

One company that may be cracking the Arabic language code is the Ontario-based AI firm Cohere, according to Khandeparkar. 

Cohere has released a new multilingual AI model called Command R7B Arabic that seems to outperform in its class at tasks such as answering questions on company materials, performing document summarisation and leveraging external tools to automate repetitive work.

The company has said it focused in developing its Arabic AI model on addressing the unique challenges of Arabic language processing, such as managing complex morphology and ensuring accurate dialectal variations. 

“When we add the complexity of architecture and the type of kit required for AI implementation at scale, we see the true scale of the challenge for businesses here in the Middle East,” Khandeparkar said. “This alone is a key factor which makes the new R7B model one to watch out for.”

Another challenge in AI adoption for the Middle East is organisational readiness, said Melih Murat, associate research director for IT services and software at IDC.

“Commonly cited obstacles include the absence of a comprehensive AI strategy, limited availability of high-quality data in sufficient volumes ad lack of an AI-ready infrastructure strategy.Others are insufficient internal guidelines and policies for governing AI deployments and gaps in employee readiness,” Murat said. 

“Without addressing these constraints, organisations will struggle to realise business value from foundation models.”