Orange partners with OpenAI, Meta on custom African language AI model

French telecommunications giant orange The partnership was announced on Tuesday. microsoft– Supporting OpenAI and Facebook owners Meta Build custom artificial intelligence models designed to provide a deeper understanding of African languages.

Orange is collaborating with OpenAI and Meta to create custom AI built on their respective Whisper and Llama open source AI models (openly available systems that can be adapted to meet specific needs). He said he is developing a model. This model can understand West African languages ​​that most conversational systems cannot understand.

Many of the big data AI companies are now training their algorithms based on those originating from the United States. This means that when it comes to different regions such as Europe, the Middle East, and Africa, the model may lose important context such as culture and language.

According to Steve Jarrett, Orange’s chief AI officer, this means that these models can have difficulty understanding text- and voice-based communications composed of less commonly expressed languages. It is said that it means

“With open models, you can do what’s called fine-tuning, which means you can introduce additional information into the model that wasn’t included when the model was originally trained,” Jarrett told CNBC. said in an interview. “Add recognition for regional languages ​​in West Africa that no AI currently understands.”

Orange plans to initially deploy an AI model in early 2025 that incorporates Wolof and Poular, two regional languages ​​in West Africa, spoken by approximately 16 million and 6 million people, respectively.

Wolof is a language spoken in Senegal, Gambia, and southern Mauritania, while Pular is primarily spoken in Senegal.

The open-source AI model will be made available externally by Orange under a free license for non-commercial uses, including public health and education, the company said. Orange plans to expand its custom AI model efforts to eventually cover all 18 West African countries.

Jarrett told CNBC: “We operate in West African countries, and many regional languages ​​are spoken in our contact centers, and our current AI model does not reflect what people are typing or saying. I can’t understand it,” he said.

According to Orange’s head of AI, major large-scale language models such as OpenAI’s GPT, Meta’s Llama, and Anthropic’s Claude are not specifically trained on data originating from the African region and therefore cannot meet the needs of Africans. is not very suitable.

The move capitalizes on a concept that is gaining global attention, known as “sovereign AI.”

The term allows individual countries and regions to control the core technical infrastructure on which AI systems are built by localizing data storage and processing to ensure that their particular languages, cultures, and histories are represented. refers to the idea that we need to strengthen our

Orange is also looking to localize data processing and hosting of OpenAI models in European data centers. This will provide early access to OpenAI’s latest and most advanced AI models, which will help build new applications such as AI-powered voice systems for customer service, Orange said.

Jarrett said Orange is committed to using AI “responsibly,” and given the environmental concerns associated with the technology’s enormous energy requirements, it “always responds to every problem at scale.” It does not use a language model (LLM).”

In addition to using AI systems to improve customer service, Orange is also using its technology to improve its mobile network, a core part of its business.

“On the network side, we’re using (AI) to not only optimize how we plan the network, but also how to properly operate the network,” Jarrett told CNBC.

“The amount of data from all network equipment is so large that AI systems can identify patterns in the data to identify and predict failures before customers notice them.”