Artificial intelligence: capability, use and governance

Artificial intelligence means computer systems that learn patterns from data and use them to classify, predict, generate or decide: recognising speech, translating languages, reading medical images, driving vehicles, or writing text. Modern systems are powered by large datasets, powerful computing hardware and machine learning methods, and their capabilities have grown quickly, which is why governments now treat AI as both an economic opportunity and a governance challenge.

The IndiaAI Mission was approved by the Union Cabinet on 7 March 2024 with a total outlay of 10,371.92 crore rupees over five years. Its pillars include shared computing capacity, foundation models, datasets through the AIKosh platform, application development, future skills, start-up financing, and safe and trusted AI. The compute pillar addresses a basic constraint: training advanced models needs thousands of graphics processing units, which few Indian firms or universities could afford alone, so shared, subsidised access is meant to widen participation.

Uses in governance are already visible: language translation for public services, crop and weather advice for farmers, detection of fraud in tax and welfare systems, and triage support in health. Each use raises the same questions: is the training data representative, is the system's decision explainable, and can a citizen challenge a wrong decision? An AI system trained on incomplete data can reproduce bias, for example by under-serving a language, region or community that is poorly represented in the data.

Risks beyond bias include deepfakes, synthetic images, audio or video that impersonate real people, which can be used for fraud or to spread false claims during elections. The IT Rules of 2021 require platforms to act on certain unlawful content, and advisories have urged platforms to label synthetic media, but enforcement depends on detection, which remains imperfect. Labour effects are equally contested: AI may automate routine clerical and coding tasks while creating new roles in data, design and oversight, with the net effect depending on skills and on how firms reorganise work.

India's stated approach has been to promote AI development while managing harm through existing law and sectoral rules, rather than a single comprehensive AI statute. Whether that approach suffices as systems become more powerful is the open question, and it will be tested by incidents: a harmful deepfake, a biased welfare decision, or a data breach will do more to shape AI law than any strategy document.