OpenAI and the UAE: ChatGPT Learns Arabic and Local Etiquette (OpenAI G42)

OpenAI no longer wants to be just a global service for everyone. Sam Altman seems to have found a gold mine in the idea of digital sovereignty. The company has begun close cooperation with the Abu Dhabi technology giant G42 to create an exclusive version of ChatGPT. This is not a cosmetic makeover or the addition of an Arabic keyboard layout. It is about deep customization of the neural network for the needs of the UAE government.

Why now? The Emirates stopped being merely an oil exporter long ago and have become one of the most aggressive players in artificial intelligence. They have money, they have ambition and, importantly, they have G42, a company closely tied to the ruling dynasty that already has experience working with Microsoft and other Western giants. OpenAI understands that to gain a foothold in the Middle East, it cannot simply offer a standard Western model. It needs something that speaks the language of the region, both literally and figuratively.

The main technical challenge here is Arabic. Although GPT-4 handles translation reasonably well, the dialects and cultural nuances of the Arab world require careful fine-tuning.

But another aspect of the deal is far more interesting: the so-called content restrictions. Western models are often criticized for liberal bias or, conversely, for excessive caution on issues considered sensitive in the US. The UAE has its own ideas about what is acceptable and what is not. Building a version of ChatGPT with "built-in filters" on a government's order is a precedent that shows OpenAI's willingness to compromise for big markets.

This move can be seen as the start of the era of "sovereign AI". We used to think there would be one big model that knows everything. Now we see that states want their own "brains" that are loyal to their laws and traditions.

For Altman, it is a brilliant business move. Selling a $20 subscription is one thing. Selling a national knowledge infrastructure to an entire state means entirely different numbers and a different level of influence.

Of course, the question of ethics arises. If OpenAI starts building special versions for every government, will it not turn into a tool of censorship? Behind the scenes, OpenAI employees are already discussing where the line lies between localization and complicity in restricting the freedom of information. But business, it seems, has already made its choice.

While the agreement is being finalized, G42 and OpenAI engineers are working to ensure that ChatGPT not only understands Arabic script but also knows what to keep quiet about in polite Abu Dhabi society.

G42's ties to China and the US add spice to the story. US regulators have previously expressed concern about G42's links with Chinese technology companies. That OpenAI (backed by Microsoft) is joining this alliance suggests Washington wants to pull the Emirates to its side in the technology war. ChatGPT thus becomes not just a chatbot but an instrument of geopolitical influence.

The bottom line: OpenAI has set a precedent by selling a "politically correct" AI to a specific state. Who is next in line for a tame ChatGPT of their own?