India’s emergence as a potential global AI powerhouse presents a striking paradox. While the country boasts the highest AI skill penetration rate globally, with a projected AI talent pool growth of 1.25 million by 2027, industry leaders are increasingly concerned about the depth and quality of this talent.
In this context, the term “weekend engineer” often gets thrown around to casually describe professionals who engage with AI technologies superficially without understanding the fundamental mathematics and principles behind them.
This trend is particularly evident in the current landscape, where everyone claims expertise in ChatGPT and generative AI.
Vinay Kumar, the CEO of Arya AI, strongly echoes this sentiment, saying we (Indians) simply use and then produce, and create, maybe wrappers. “Now, everybody is an ML engineer. Everyone is a ChatGPT expert. Everyone is a generative AI expert. [I’d say] 98% is rubbish because it’s a weekend project,” he added.
At Cypher 2024, AIM asked Prateep Basu, the co-founder and CEO of Satsure, about the biggest challenges he faced while building models from scratch. He said that most of the talent in India is used to predefined black-box runs and not the actual maths behind them.
The Depth Problem
The challenge isn’t just about numbers—it’s about depth. India’s active pool of senior AI engineers building core AI products and services is surprisingly small, with less than 2,000 professionals. This stark contrast becomes apparent when compared to the 650,000-700,000 people reportedly trained in AI across top-tier tech companies.
Along similar lines, a Reddit user said, “A few years ago, AI in Indian companies was about rote learning the ImageNet architecture and knowing how to process data and use Hello World of the ML frameworks.”
He further said that these days, most of these ML experts and data scientists are basically ChatGPT experts. “Their resumes change faster than Python version releases,” he added, suggesting a sudden skillset drop in the Indian AI market.
Kumar mentioned that in India, the focus has always been on applied engineering, not research engineering or core R&D engineering. Not much goes into finding out the whys and hows. It’s primarily “let’s create something functional.”
Thus, the reinforcement was always around, making it functional, not understanding why it is functional.
Another problem that companies face while building models from scratch is data. Basu also mentioned that data is the second most important requirement to build models from scratch after talent. Now, most of the data available to these models comes from the USA. So, when one wants to build models from scratch, especially for India, data becomes a huge problem.
As India lacked enough data and talent, Kumar mentioned that they were in constant touch with researchers from the USA who were skilled, had developed a model to follow and had enough data as well.
India, the Service Market
This has been the idea for the Indian service market. If we track the evolution of the software ecosystem in India over the past few decades, we will realise that the country is known as a service provider. And that is what India is pursuing with AI, too.
Without mincing his words, Kumar said that though we can create great apps and software applications, our contribution to core R&D in AI is negligible. “We are far behind when you compare to, let’s say, China, which was extremely behind 10 years back. Now, they are frontlining the contribution to R&D,” he said. “But there is also a lot to do with the ecosystem.”
For OpenAI, the fundamental idea was R&D—to explore the possibilities in AI and probably get to AGI. The idea was well accepted in the US ecosystem, but according to Kumar, it won’t be feasible in India. He mentioned that when they started their company, they were the first deep learning startup at that time, and the ecosystem questioned them at each phase.
“You know, India is not an ecosystem known for deep learning research. When we started, there was no coursework. Forget about any research. We spoke to researchers to figure out how they had done it, got that learning, and implemented it correctly. So, it’s got a lot to do with the ecosystem,” he added.
However, slowly, the situation is changing. OLA’s Krutrim and Sarvam 2b are among the few indigenously built AI models, suggesting India is slowly transitioning to the R&D part of AI. But we have yet to witness a point from which we can say that India is leading the AI research department.
The post India is the Land of Weekend AI Engineers appeared first on Analytics India Magazine.
