The term ‘Algorithmic Bias’ is used to describe systematic errors in an AI system that tend to create biased outcomes repeatedly. Apart from flawed training data, a common cause of Algorithmic Bias is AI developers passing on their own biases to the algorithm while designing it.
Judging from the skewed outcomes incentivized by America’s recent tech policy, a clear case of Algorithmic Bias can be inferred – stemming from biases at the design stage. Just as inaccurate assumptions from AI designers can exaggerate weights on some attributes, Washington’s mistaken understanding of what it takes to win the tech war with China emphasises too much on export controls and has a blind spot on compute access.
For starters America’s basic assumption that AI chip export restrictions will slow down China’s tech development has already been proven resoundingly wrong. China’s DeepSeek AI model managed to work around export constraints and succeeded in overtaking America’s best frontier AI models like ChatGPT and doing it all at a fraction of the cost to boot.
Another design flaw in Washington’s AI policymaking is the discriminatory treatment of some of its existing and several potential allies. The latest round of tech regulations (in January 2025) adopt a three tier classification of trusted, restricted and intermediate countries. This divisive approach is unfair, as long standing US allies like Israel, Saudi Arabia and NATO partners Poland and Greece have been excluded from the trusted category. The category of intermediate countries pools together disparate countries from India, Turkey, Vietnam and Singapore which lie at various points on the spectrum in their alignment towards China and the US.
The third design flaw is America’s blind spot on compute access. Even though the trusted category (of just 18 countries) has unrestricted AI chip access, US cloud providers have been mandated to keep at least 50% of compute power within the US and a maximum of 7% of compute installations in the intermediate countries. The intermediate category of more than 140 countries faces strict quotas on AI chip access and non-US cloud providers are allowed to place only 25% of their compute power in these countries. This is detrimental to the intermediate countries as it not only impoverishes their access to critical AI chips necessary to build their own compute power, but also prevents them from accessing it at scale via cloud service providers from the trusted category. Finding themselves at the short end of these restrictions, these countries can permanently shift their allegiance to Chinese service providers.
Already, Chinese tech giants like Alibaba, Tencent and Huawei are exporting their tech and talent to neighbouring countries within Asia Pacific and other regions neglected by the US – such as Latin America and Africa. ByteDance has announced expansion plans into Thailand and Malaysia. Policymakers in Malaysia have admitted to receiving strong interest from Chinese companies into the datacentre market but maintaining a low profile “lest they incur the attention of the US authorities”.
So far the US has a strategic advantage in compute power but the policymakers’ temptation to concentrate it within the US will erode that advantage over the long term. We have already seen this design flaw and its consequences in the telecom sector. The blacklisting of Huawei Technologies in allied countries did not prevent it from being the world’s largest telecom equipment supplier (with a 30% market share). Undeterred by possible Western censure, in November 2024, Malaysia selected Huawei for the rollout of the country’s second 5G network.
To return to the analogy of Algorithmic Bias, if US tech policy can be thought of as an algorithm, then it has been poorly designed. A faulty approach to AI policy, akin to an algorithm with embedded design flaws, will lead to faulty outcomes – viz. ceding lucrative markets abroad to Chinese competitors. At the very least, design flaws lead to an erosion in the usefulness of an algorithm and continued reliance on such a system can result in outcomes diametrically opposite to the original intention. In the context of the US-China tech race it could mean witnessing entrenching dominance of China over the world’s tech infrastructure while the US watches helplessly.
Macrologue 