hen it comes to international AI policy too often the narrative is focused on geostrategic competition over the core technologies, the race to build ‘sovereign AI’ by countries large and small, and how to have ‘sovereign data’ regimes that allow countries to control how data is used at scale.
To one extent or another these national security–related approaches often frustrate key aspects of development of this technology. That’s because AI requires international cooperation at fundamental levels far more intensive than previous general purpose technologies.
There are straightforward reasons why this is true. The world’s first comprehensive synthesis of current evidence on the capabilities, risks, and safety of advanced AI systems involves about 100 of the most respected scientists and researchers from more than 30 countries, chaired by Turing Award–winner Yoshua Bengio. It is the International AI Safety Report 2025 released in January. It also makes for very sobering reading.
Let’s address just one of the most fundamental challenges: Bias. While there are many drivers for bias let’s consider the big three:
- If AI software developers are primarily from one language group or cultural tradition, the biases in their worldview creep into any model built upon it. This argues for more globalization of development efforts.
- Most of the data used in “pre–training” of models is primarily in English and derived from Western cultures. That means queries are more likely to be biased against all other socioeconomic groups and to reflect biases in those cultural contexts as well. The push for “sovereign data” at the national level is philosophically contrary to resolving this as it can frustrate efforts at globalized datasets normalized to prevent bias;
- If a model’s trainers are primarily from one language group or cultural context, the model will also contain biases related to them. This suggests we need testing from people from every culture and language for best results.
The report goes into bias in depth (see section 2.2.2) but the conclusion is that reducing it requires global cooperation in every stage of development, training and testing and throughout a model’s lifecycle. Failure to address bias effectively means real people will experience genuine harm. We know this because it has happened and continues to happen.
We should spend much more public policy time enabling the international cooperation we need. Otherwise, even the countries who ‘win’ the AI race are likely to end up losing.
Images are for reference only.Images and contents gathered automatic from google or 3rd party sources.All rights on the images and contents are with their legal original owners.
