By Vidhu Shekhar
In the blink of an eye, Generative AI (Gen AI) has taken the world by storm. What began as a groundbreaking innovation with the launch of ChatGPT in November 2022 has swiftly evolved into a game-changing technology. From crafting compelling stories to making vivid visuals or revolutionizing software development, Gen-AI has emerged as a tour de force.
The rapid ascent of this transformative technology has been nothing short of awe-inspiring. Technology giants and startups alike have joined the fray, vying to push the boundaries of what’s possible. Anthropic’s Claude AI, Google’s Gemini, Meta’s Llama, Replit’s Co-Pilot, and open-source platforms like Mistral AI and Stable Diffusion have all left their mark, contributing to the breakneck pace of advancement in just over a year.
Indian startups face a critical question given the rapidly evolving technology landscape, primarily led by global entities: How can they adapt and excel? Sam Altman from OpenAI once suggested that developing foundational AI models like those at OpenAI might pose a challenge for an emerging team in India with limited resources. Can Indian startups rise to the challenge? And how?
Gen-AI: Big Engines & Domain Experts
In understanding how Generative AI can be transformed into a service model that can be leveraged by businesses, it is important to distinguish between two main types of AI services: big engine services and domain expert services.
Big engine services involve developing and deploying large-scale Gen-AI models capable of performing a broad spectrum of tasks across various domains. Typically, these services require substantial computational power, massive datasets, and specialized technical knowledge, making them a feasible option primarily for established tech giants or well-funded startups.
In contrast, domain expert Gen-AI services are focused on creating specialized models tailored to specific industries or use cases. Despite, perhaps because of the global emphasis on big engines, the specialized models are relatively underdeveloped. The scarcity of domain-specific AIs contributes to why generative AI has not yet caused a fundamental transformation in industry.
Why are the domain-specific models underexplored? The prevailing belief is that the general big engine models can be adapted or “fine-tuned” to meet specific needs. However, this is often not the case. General models lack precise command and control due to their training on extensive global data, making it difficult to predict their output.
While this unpredictability can be managed when a firm deploys Gen-AI for internal business operations, it poses significant challenges in offering Gen-AI as a B2B or B2C service. For example, Air Canada was legally held accountable when its chatbot issued discounts “on its own” without any authorization. This “on its own” way of such engines can be particularly serious in sensitive sectors like finance, healthcare, and others – where such autonomy can lead to serious regulatory, ethical, and legal consequences.
This focus on fine-tuning big engine models to achieve domain expertise, which has not yielded the desired results, is also why ChatGPT’s customized GPT store – initially touted as a game changer, has not yielded any significant service breakthroughs.
But then, if the big engine global models can’t be fine-tuned to achieve good domain expertise, how do we get domain expert AIs? For this, we need to distinguish two parts of Gen-AI models.
One part is understanding the query received – we can call it Natural language Understanding (NLU). The second part is creating the answer to the query received, or the generative part.
In the domain of Expert Gen-AI models, the key is that while global big engines can be deployed in NLU, i.e., extracting the meaning of the query raised, the generative part creating the answers must be customized bottom up and not using the global engines. The generative part must be trained solely on context and domain-specific data, without touching the global data. Doing this leads to engines that answer only as per the data fed and do not go astray in answering by extracting from the global database.
These specialized AIs require unique data sets specific to their fields for the generative part. This presents a significant opportunity for businesses, especially Indian startups, to lead in creating effective, reliable AI solutions tailored to the specific needs of different industries
The Indian Startup Opportunity
Developing these specialized AI models necessitates close collaboration between AI technologists and industry experts. India’s wealth of skilled professionals across various sectors enables the formation of dynamic, multidisciplinary teams capable of pioneering innovative AI solutions. By capitalizing on this diverse talent pool and encouraging collaborative efforts, Indian startups are well-positioned to lead in the Generative AI sector.
Additionally, India’s established base in Business Process Outsourcing (BPO) and Knowledge Process Outsourcing (KPO) provides a rich source of initial datasets for these AI models. This extensive service sector can be transformed into a robust foundation for AI development. This can be further enhanced via their deep understanding of local markets, customer needs, and the specific challenges of various industries. This knowledge positions Indian Startups to develop AI models finely tuned to the requirements of sectors such as healthcare, finance, education, or e-commerce.
By designing AI solutions tailored to these industries’ specific pain points, Indian startups can offer more than just generic AI services. They can provide solutions that truly resonate with their customers, setting themselves apart and becoming trusted partners within their domains.
These specialized services utilize in-depth domain knowledge and expertise, ensuring the AI solutions are not only relevant but also highly effective for their intended users. By focusing on creating domain-expert AI services, Indian startups can establish a distinct presence in the burgeoning Generative AI ecosystem. This allows them to capitalize on their unique strengths and the opportunities presented by this technology. With global tech giants yet to embrace domain-specific AI solutions fully, Indian startups have a unique opportunity to pave the way in this emerging field.
In conclusion, Indian startups are uniquely positioned to lead in the Generative AI space by leveraging their deep local insights and industry-specific expertise. They can leave building Big AI Engines to the behemoths of the country, like Jio, and start focusing on creating domain-expert AI solutions.
This approach not only differentiates them from global players but positions them as invaluable partners to India’s growing businesses. Imagine AI doctors providing affordable healthcare advice in rural areas or AI systems revolutionizing how countless small businesses manage their finances. The potential is limitless for Indian startups to solve real-world problems with focused AI solutions.
Vidhu Shekhar is the Assistant Professor, Finance and Fintech at SPJIMR. Views are personal.