The internet has greatly improved access to data and information but has not substituted for human analysis and implementation in the economy. AI may develop the capacity to impose precisely this substitution. But I do believe that AI cannot substitute for human creativity when it does not depend on available data and information. Einstein’s discovery of the Theory of Relativity could not have been made by an AI model.
In AI systems, large language models can access available data and information and answer queries, reducing the role of humans as synthesisers. But AI is more than this. It has already developed the capacity for coding and connecting directly with computers and the internet. It will move further ahead towards artificial general intelligence (AGI), when AI can roughly automate 90 per cent of current white-collar work.
A further development that is projected is artificial superintelligence (ASI), when AI exceeds human intelligence by a good margin. In fact, an AI researcher has said that there is a 10 per cent chance that an AI system could annihilate humanity because AI capability could outpace the human ability to monitor and control it. There is now a growing concern that the development of AI is not sufficiently regulated to prevent potential disasters.
There are proposals for an international treaty to ban superintelligence because AI products may escape human control and reprogram themselves to adopt new, self-directed goals. Preventing AI from being used to threaten humans and make them redundant deserves a clear international agreement that this activity is not acceptable.
Hence, the first requirement of AI policy for India is that the government must promote and participate in a negotiated global consensus for ensuring that the orientation of AI systems is subject to control so that the development of AI does not undermine humanity.
AI provides a substitution for skilled workers who process data and information to provide substantive outputs for users. For instance, AI’s potentially superior skill in analysing X-rays, MRIs and CT scans can substitute for radiology workers. But AI cannot substitute for outputs that require physical labour, and that is why one forecast is the growing equalisation of salaries for skilled and semi-skilled or unskilled workers.
A more unwelcome impact is the shift in income from workers to owner-suppliers of AI models. The return on AI models comes from their capacity to make constructive use of data and information from a vast number of humans that is available online. Hence, there is a rising concern about the need for a policy framework that would involve a fairer sharing of the returns to AI companies with information providers and users.
Hence, the second and perhaps the most important requirement is to design an AI policy that minimises its potentially adverse social impacts on employment and income distribution.
The way AI is developing could greatly increase the disparity among nations in control over the technology. Hence, we must also design a policy framework that prevents us from becoming a loser dependent on others.
At present, the development of AI is dominated by two countries — the United States and China. There are about five widely used AI models each from the US and China. The Chinese models are cheaper to access and are reaching out to the billion or so direct users of AI. The Chinese AI models are under the control of the government. That is why when I asked a Chinese AI model: “Tell me more about what happened in Tiananmen in 1989?” the reply I got was: “I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.”
The political bias in the response from AI reflects how LLMs are oriented. An interesting exercise was undertaken with a query including about 10 politically oriented prompts. Some AI models gave what was described as a balanced reply with a liberal worldview. What is interesting is that one Chinese AI model that gave a balanced reply when asked in English gave a China-biased reply when questioned in Chinese! But quite a few AI models from the US, France and China also seemed to reflect the worldview of their national governments1.
The impact of AI on democracy and freedom will depend on how it affects the networks of communication. If the networks are controlled by owners, advertisers and the government, the impact on democracy could be adverse. However, if the network is effectively controlled by the people who use it, and who will reject it if it is adverse and go to an alternative, then the impact on democracy and freedom is positive. The real challenge is to ensure that the AI LLM that dominates communication is unbiased.
Hence, the third requirement of AI policy for India is that the government must recognise the implicit politicisation of AI output and promote domestic development of successful AI models relatively free of ownership and government control. But the real test is the extent to which Indian nationals use the sovereign AI model rather than the Chinese and US ones that are dominant now.
A closely related area is the development of data centres, which are a crucial part of any AI system. We should not depend on compute from outside our borders. Despite generating 20 per cent of the global data, India currently holds only a 3 per cent share of global data. We have about 150 data centres in Mumbai, Chennai, Bengaluru, Hyderabad and Noida and rank 14th in the global country-wise hierarchy of data centres. This will go up substantially and require a significant amount of electricity and water.
India’s data centre policy has been dominated by its intention of localising the data information that is available through globally connected devices. It has announced a very liberal policy for foreign AI providers to set up data centres in the country, with a 21-year holiday to foreign firms that would provide cloud services based on domestic data centres. But the promotion of domestic AI models will require domestically-owned data centres.
Hence, the fourth requirement is a government policy that focuses strongly on promoting domestically-owned data centres. But this policy must also include rigorous standards about water and energy use by data centres.
AI systems are going to reshape development and human prospects. Hence, our government must devote deep attention to the four requirements listed above.
1. Sinan Ulgen, The World According to Generative Artificial Intelligence, Carnegie Endowment for International Peace, 2025
