Indian Al needs imagination, not imitation

True innovation happens when we take bold bets on hard, important problems — not lust by replicating existing breakthroughs like chatGPT.

To lead in Al, India must stop chasing what others have already built and instead focus on what hasn't been solved yet. True innovation happens when we take bold bets on hard, important problems - not just by replicating existing breakthroughs like ChatGPT. Remember, deep learning itself was born by challenging the status quo, not copying it.

It's concerning when Indian Al ambitions are reduced to "building our own ChatGPT." Yes, building LLMs (large language models) from scratch is essential - not to chase the past but to develop the talent and know-how needed for deeper innovation. It's a stepping stone, not the finish line. India needs a three-pronged Al innovation agenda:

A. Foundational Models for Hard, Impactful Problems

Al models today address language and images — but the real world demands more. Take robotics, for example: can we build foundational models that "understand" and replicate human actions like grasping and moving objects, not just words or pictures? Imagine a model that combines what we see, say, and do — transforming how India scales manufacturing by building automation in industries.

Another powerful area is foundational models for engineering design.

From cars to rockets, Al can help generate smarter designs faster, but it's a tough challenge - most design work is manual, with messy or nonstandard data. Similarly, there are huge opportunities to impact weather prediction, drug discovery, space modeling, Indic languages, financial systems, and archaeology.

India must lead in these bold, disruptive frontiers. That needs the courage to imagine what doesn't yet and chart one's own way in unexplored territories. The methods will require the same talent and technical depth it takes to build LLMs. Fortunately, India has a unique edge: a large, skilled, cost-effective workforce for collecting and labeling the massive datasets such models require. We should reap our demographic dividend. B. Research-led Product Design

It's a myth that Al work is limited to building big models or creating simple apps. A fast-emerging middle ground is "Al agents" - systems that can infer, retrieve, and act autonomously. But current agents are clunky, unreliable, and hard to scale. Fixing this is a huge research challenge. The first big question: how should Al agents and humans collaborate?

Should Al generate full articles or work step-by-step? Should it ask the user clarifying questions first? What feedback loops work best? Solving these questions requires cutting-edge UI/UX design and thoughtful experimentation.

Second, consider retrieval agents, where LLMs respond to queries based on documents or structured data like CRMs. Issues arise when information is scattered or when we need to mix structured and unstructured inputs. It is hard to control the foundational model responses, which go down the Alice in Wonderland rabbit hole! Third, when these agents complete a task, it is hard to verify whether the response, an image, or text is good enough. That is why most AI deployments fail. Further, if we could verify the Al output automatically, we could create virtuous feedback loops for improvement. This space - research-led product design - should be India's second focus. It's where we can create Al experiences that are not just functional, but truly valuable.

C. Be Ready for What is Coming Next Every decade, the Al stack shifts. From rule-based systems to statistical methods to today's deep learning - each wave replaced the last. But today's models are far from perfect: they're power-hungry, unreliable, and often lack reasoning. India must prepare for the next shift in Al - not by catching up later, but by helping shape it. This includes but not limited to:

World models that understand the physical world and reduce hallucinations.

Efficient learning systems that mimic how humans learn with less data.

Neurosymbolic models that mix neural networks with logical reasoning, and

New computing substrate, like low-power analog chips or even biological computing.

Industry may not fund these moonshots, but Indian academia and philanthropic capital must. These are the bets that can make India a leader in the next Al revolution.

India's Al journey must move beyond imitation and be world-class. By combining our strengths in talent, scale, and imagination - we must rise up the Al value chain.

Read Original Article