Editorial: Peaking AI

In the backdrop of the AI Action Summit held in Paris, Prime Minister Modi offered to host the next conference in India, buoyed by the decision to set up the AI Foundation and the Council for Sustainable AI at the meet. The PM emphasised on the need for making the AI partnership more inclusive of the Global South and its priorities, concerns and needs. India’s presence at the summit is being seen by stakeholders as an opportunity to advance fundamental priorities for the Global South. With regard to democratising access to AI resources across the entire value chain, there are a few prerequisites.

India's domestic initiative of establishing a common computing facility with over 18,600 GPUs and 40% government subsidy is a step in the right direction. However, the scope of this exercise must extend beyond computer hardware deployment. Ensuring equitable access to data sets, data storage infrastructure, cloud computing platforms, foundation models and application development platforms are the other non-negotiables. There has to be a shift from traditional computing infrastructure to distributed computing solutions. Experts are of the opinion that while private players can support AI sovereignty, only sustained government investment in funding, R&D, and education can help build large language models (LLMs) from scratch.

We must focus on the promotion of open source AI models. The success of the AI model DeepSeek R1, which sent shivers down the spines of AI behemoths in Silicon Valley, is a case in point. The model was open sourced, came with a detailed technical paper on how it was built, and invited the global AI community to build on top of it. However, observers are apprehensive that future versions of DeepSeek's models may not remain open source indefinitely as they are developed outside India. Worryingly, in its last week in office, the Biden-Harris administration introduced a policy called the Framework for AI Diffusion.

The measure is aimed at preserving the hegemony of the US in AI technology, balancing innovation and national security, while deterring US adversaries from reaping the strategic benefits of AI. The framework places countries in three different tiers - each subject to varying levels of restrictions. The top tier is for key allies, with unrestricted access to import AI tech. The bottom-most layer comprises key adversaries like China, North Korea, Russia and Iran, to whom the export of advanced AI systems are restricted. India falls in the middle layer of limited access. Researchers have said that the framework is disadvantageous to India as it discourages development of cutting edge AI systems beyond US borders, which in turn will disincentivise investments and operations of Indian subsidiaries of leading US AI companies. As a result, India’s pool of top AI talent might be drawn away, and in turn that could impede knowledge transfer and innovation.

Another challenge is developing a framework to identify and prioritise AI use cases relevant to the Global South’s specific requirements. This could include early disease detection systems customised as per the local healthcare infrastructure, personalised learning platforms that are cognizant of varied educational contexts, productivity tools for agricultural workers that are calibrated as per regional farming practices, among others. Last, but not least, we also need to build AI-resilience, to mitigate the job losses resulting from the launch of platforms like metaGPT which simulates the entire working of software companies, AI writing 25% of Google’s code, and chatbot-led layoffs.