AI Days 2024: A People-Centric Approach to Artificial Intelligence in India
The AI Days 2024 conference, held in Hyderabad on the 30th and 31stMarch, championed a people-centric approach to Artificial Intelligence, aiming to demystify AI and leverage its potential to solve real-world problems for the common person in India. The conference featured a diverse array of speakers from leading tech companies, academic institutions, and government, all converging on the theme of "AI for Society."
Challenging the Narrative: AI for the Commons
Y. Kiran Chandra, Founder of Swecha, opened the conference by addressing common misconceptions and fears surrounding AI, particularly the notion that it will lead to widespread unemployment. He emphasised that AI is an inevitable tool for solving societal problems and highlighted a key product launched at the conference: a Telugu language assistant designed to empower farmers and agricultural labourers, enabling them to access information and communicate in their mother tongue, regardless of literacy. This initiative underscored the conference's core message: AI should serve the commons and improve the lives of everyday citizens, from farmers to sanitation workers.
Hyderabad: The Emerging AI Capital
Chaitanya Chokkareddy, CTO of Ozonetel, articulated the vision of establishing Hyderabad as the AI capital of India. He cited the city's robust ecosystem of IT and product companies, research institutes, educational institutions, and a thriving startup scene as key factors. Chokkareddy also presented an alternative model for AI development, challenging the dominance of mega-corporations like OpenAI, Microsoft, Google, and Amazon. He advocated for a community-driven approach to data collection and model building, emphasising open-source licenses to protect data from misuse.
He shared the success story of "Chandamama AI," where 7,000 stories were collected within six hours using student power. Another groundbreaking initiative involved collecting 1.5 million Telugu voice samples through a unique concert where entry was granted in exchange for voice contributions, demonstrating how society can collectively contribute to building massive datasets at a fraction of the cost incurred by large corporations. This data was then used to develop a small, open-source Telugu speech model that runs efficiently on mobile phones, showcasing the potential for AI solutions that are accessible and do not require extensive computing power.
The UPI Approach to AI Innovation
Professor Gaurav Raina from IIT Madras presented a compelling "UPI approach" to AI innovation, drawing parallels with the success of India's Unified Payments Interface. He distilled three key ingredients for solving complex problems and driving innovation:
- Identify a Challenging Situation: Start by addressing one of the most challenging real-world problems within a specific context. For UPI, it was enabling an auto-rickshaw transaction. He underlines that the success of UPI was largely built on the Immediate Payment Service. For AI in India, identifying a challenging situation means focusing on problems relevant to the Indian populace, such as those in education, healthcare, agriculture, and financial services.
- Bring Unlike Minds Together: Foster collaboration between policymakers, government, industry, and academia. The success of UPI was a testament to this collaborative spirit, where diverse perspectives and interests converged to co-create solutions.
- Work Towards a Democratic, High-Quality, Scalable, and Affordable Solution: The goal should be to create solutions that are accessible to everyone, regardless of their socio-economic background. This often necessitates developing smaller, more contextualised models that require less compute power, making them more affordable and scalable for a diverse nation like India.
Professor Raina emphasised that by adopting this framework, India can develop at least "10,000 solutions, 10,000 dreams, 10,000 problems, 10,000 opportunities for innovation" that not only serve the Indian context but also inspire similar adoptions globally. He redefined AI as "Intelligence for the auto-wallah," highlighting the human-centric amplification that AI can provide in various sectors, from education to justice, while also touching on the architecture of AI which can be described as built around three layers, computing, data and models.
AI and Law Enforcement: Addressing Emerging Crimes
Shikha Goel, IPS, Additional DGP, Telangana, provided insights into the intersection of AI and law enforcement. She acknowledged her non-expert status in AI but underscored its pervasive influence on every aspect of life, including the emergence of new-age crimes. Goel highlighted how AI is being used as an accomplice by criminals, citing examples from airline ticket cancellations to impersonating CFOs for financial fraud. She emphasised the need for law enforcement agencies like the Telangana Police to be well-equipped to deal with these evolving threats and expressed a desire for collaboration and learning within the AI community to ensure public safety in the digital age.
Making Science More Accessible with AI
Dr. Somdatta Karak, a scientist at Centre for Cellular and Molecular Biologyand a science communicator, advocates for using AI to make science more accessible. She highlightedthe inequity of information in society and aims to bridge the gap between scientific research and the public, including policymakers and non-experts. Dr. Karak emphasizedAI's potential in content generation and platform development for science communication, particularly for creating inclusive and engaging content in various languages. She also addressedthe challenges of fighting misinformation and building science-empowered communities, suggesting AI tools can help democratize reliable information and alert experts to viral misinformation.
Dr. Jaideep Ganguly's keynote at AIDAYS 2024 covered 50 years of AI development, emphasizing that we still don't fully understand intelligence. He discussed early knowledge-based systems and machine learning, highlighting that ML is primarily classification, not intelligence. Ganguly then delved into neural networks, explaining their simplistic model compared to the complex, digital and analog functions of the human brain. He critiqued the current understanding of neurons and synapses in AI models, noting their limitations in replicating true cognition.
The talk concluded with an analysis of GPT, explaining it as a statistical machine that predicts words based on probabilities, not genuine understanding. He cautioned that GPT is prone to "hallucinations" (statistical errors) and doesn't grasp complex concepts like "African elephant" as a unified entity as it grasps that as two separate words and builds a statistical analyses based on them.Ganguly stressed that the future of AI lies in better understanding brain functions and translating that into more realistic computational models, rather than relying solely on current statistical approaches.
In essence, AI Days 2024 served as a powerful platform to advocate for an inclusive, community-driven, and contextually relevant approach to AI development in India. The conference underscored the immense potential of AI to transform lives, provided a roadmap for home-grown innovation, and highlighted the critical need for responsible AI development to address both societal challenges and emerging criminal activities.
***