Best Place To BuildProf. Ravindran on The IIT Madras Playbook for Building AI Leaders | BP2B S1 Ep. 6
CHAPTERS
- 0:00 – 2:20
From niche RL class to global AI educator: why IITM is the best place to build
Prof. Balaraman Ravindran opens with a memorable story about being recognized internationally through his ML videos, setting the tone for AI’s expanding reach. The hosts frame IIT Madras as a builder-friendly ecosystem and probe how Ravindran’s teaching became so widely followed.
- •Anecdote: international recognition from learners who studied via his videos
- •Core claim: AI won’t take your job—people using AI will
- •Podcast setup at IITM’s innovation hub; Ravindran’s role and visibility on campus
- •Early hint of IITM’s boundary-less, build-centric culture
- 2:20 – 8:30
AI education at scale: RL class growth, NPTEL reach, and the 35,000+ online BS
Ravindran explains how reinforcement learning went from a tiny, misunderstood elective to one of IITM’s most popular courses. The discussion expands to NPTEL’s massive impact and the rigor of IITM’s Online BS in Data Science & AI.
- •RL course started in 2004 with ~6 students; now routinely 130–180+ enrollments
- •Early RL adopters included non-CS students due to interdisciplinary roots
- •Online BS in Data Science & AI crosses 35,000+ learners; designed to be highly rigorous
- •NPTEL ML course reach: thousands of exam takers each offering; taped in real classrooms
- •NPTEL’s global accessibility: language/accent and foundational coverage as advantages
- 8:30 – 11:54
Machine Learning vs Reinforcement Learning: learning from labels vs sparse feedback
The conversation clarifies standard supervised machine learning as learning from labeled input-output examples. Reinforcement learning is introduced as a different paradigm—learning through trial, error, and sparse evaluative feedback rather than explicit “right answers.”
- •Supervised ML: learn a function mapping inputs (e.g., images) to outputs (labels)
- •RL motivation via skills like cycling/swimming/walking: feedback is sparse and indirect
- •Key RL idea: evaluation of actions without providing exact correct outputs
- •RL as a formal framework for learning under minimal feedback
- 11:54 – 15:31
A pioneer’s path into AI: chess, neuroscience inspirations, and discovering RL
Ravindran recounts how early AI excitement—especially computer chess—sparked his curiosity about human thinking. He describes moving from neural networks’ biological motivations toward reinforcement learning, influenced by neuroscience experiments and the need for optimization foundations.
- •AI’s boom-bust cycles; his entry was curiosity-driven, not trend-driven
- •Chess programs in the early ’90s and the later Deep Blue era as a catalyst
- •Disappointment with neural nets drifting from biological explanations toward pure math tools
- •RL’s link to neuroscience (e.g., monkey experiments) and the need for optimization/math
- 15:31 – 18:53
Multi-Armed Bandits demystified: decision-making under uncertainty (and why ‘bandit’?)
Bandits are explained as classic statistical decision problems: multiple choices with uncertain rewards that you only learn by trying. Ravindran traces the colorful term back to slot machines—“one-armed bandits”—and generalizes the concept to many real-world choice settings.
- •Bandits predate modern RL; long-studied by statisticians
- •Setup: multiple actions with stochastic outcomes; goal is reward optimization
- •Real-world analogy: choosing between movie vs restaurant without knowing payoff in advance
- •Origin story: slot machines as ‘one-armed bandits’ that ‘steal your money’
- •Multi-armed bandit = choosing among multiple slot machines/arms
- 18:53 – 23:22
Exploration vs exploitation: the central dilemma in RL (and in real life)
Using the movie/restaurant example, Ravindran explains why agents must explore to learn but exploit to benefit—too much of either is costly. He frames this not merely as a trade-off but a dilemma: when to switch, and how to do it systematically.
- •Exploration is necessary because payoffs are unknown until tried
- •Early lucky/unlucky experiences can lock you into suboptimal habits
- •Exploitation yields reward but can prevent discovering better options
- •Continuous exploration wastes opportunities for consistent payoff
- •The ‘switch point’ is the hard part—what makes RL difficult
- 23:22 – 26:15
“AI is the new CS”: why AI is pulling away into its own discipline
Ravindran compares AI’s current evolution to how computer science separated from electrical engineering in the ’80s. He argues AI has matured with deep inputs from psychology, control, economics, and more—enough to stand as an independent discipline, not just a CS subfield.
- •CS history: from math + EE roots to an independent discipline via abstraction
- •AI similarly draws from many fields beyond core computing
- •Meaningful AI work increasingly doesn’t require full-stack CS knowledge
- •AI’s maturity suggests it should exist as a discipline in its own right
- •Interdisciplinarity is a feature, not a bug, of AI’s identity
- 26:15 – 30:54
Founding the Wadhwani School of Data Science & AI: programs, umbrella model, and funding
The origin story of IITM’s Wadhwani School is rooted in interdisciplinary demand and existing cross-department AI activity. Ravindran explains how the school model unified degree programs and research centers, accelerated by alumni support from Sunil Wadhwani.
- •Early IITM push (2014–2015): recognize AI/data science as fundamentally interdisciplinary
- •Interdisciplinary Dual Degree in Data Science drew students from many departments
- •Need for structure: doing “department work” without a department led to formal creation
- •School model: degrees + multiple AI centers under one umbrella
- •Sunil Wadhwani’s ~₹100 crore support as the “kicker,” alongside standard govt funding
- 30:54 – 34:17
Building AI centers like startups: teams, KPIs, fundraising, and guaranteed attrition
Ravindran contrasts building academic centers with building startups: both demand strong teams, fundraising, and focus, but differ in products and evaluation. He highlights academia’s unique constraint—students must graduate—making churn inevitable and structural continuity a key challenge.
- •Commonality with startups: team dynamics determine success
- •Academic ‘product’ is enabling research/innovation culture—not a single market product
- •Fundraising and visibility matter even in academia for large initiatives
- •Quasi-corporate operating models can keep academic initiatives execution-focused
- •Academia has “100% guaranteed attrition” as students graduate by design
- 34:17 – 39:03
Robert Bosch Centre and beyond: IITM’s template for interdisciplinary AI research
Ravindran narrates how early network-science efforts became the Robert Bosch Centre for Data Science & AI, scaling across departments and influencing other IITs. He also outlines parallel biology-focused efforts (iBSC) feeding into healthcare AI initiatives.
- •Interdisciplinary Lab for Data Science (2015) began with network analytics focus
- •Bosch partnership led to RBCDSAI (2017), scaling from ~15 to ~38 faculty across 14 departments
- •Output metrics: ~100 projects, ~250 papers, startups enabled, many students trained
- •RBCDSAI became a governance/operations template for IITM and other IITs’ AI centers
- •iBSC: AI for systems biology, omics, microbiome, drug discovery; basis for healthcare AI CoE
- 39:03 – 42:15
Centre for Responsible AI (CeRAI): deploying AI safely in public services
Concerned about premature AI deployment in high-stakes government use cases, Ravindran champions responsible AI practices. He emphasizes that safe deployment requires expanding beyond technologists to lawyers, economists, sociologists, and other stakeholders.
- •Risk gap: proof-of-concept success ≠ safe real-world deployment
- •Example failure mode: retrieval + rephrasing can amplify misinformation or pranks
- •Government adoption in law enforcement/public services heightens responsibility stakes
- •CeRAI aims for responsible use (even if the name is shorthand)
- •Interdisciplinarity expands to include non-technical disciplines for governance and impact
- 42:15 – 54:52
Interdisciplinary reality check: branch choices, AI’s Nobel moment, and Perplexity’s founder as a student
The discussion reframes branch selection as influence rather than destiny, stressing lifelong learning and adaptability. Ravindran connects AI’s interdisciplinary power to Nobel-level recognition (Hopfield/Hinton; AlphaFold) and shares reflections on teaching a future Perplexity co-founder.
- •Undergrad branch influences opportunities but doesn’t predetermine outcomes; people routinely pivot
- •AI is transformative: at minimum, everyone must learn to use AI tools in their domain
- •Nobel discussion: AI’s role in physics and chemistry; AlphaFold reshaping protein structure research
- •AlphaFold’s impact: shift from sequence databases to structure databases; field focus changes, not ‘ends’
- •Perplexity co-founder in class: quiet, rigorous, curve-skewing; early projects led toward Perplexity’s evolution
- 54:52 – 1:02:36
CFI and IITM’s build culture: leadership vision, Research Park, and ‘no department boundaries’ mindset
Ravindran recounts his time as CFI’s first dedicated faculty advisor and how the innovation ecosystem matured into today’s large-scale maker culture. He credits visionary institute leadership and the Research Park for making applied impact and startups central to IITM’s identity.
- •CFI’s evolution: need for a dedicated advisor as projects/clubs scaled
- •Designing the new CFI building during his advisory tenure (2015–2018)
- •Build culture attributed to IITM DNA and sustained top-down support from directors
- •Director Ananth’s vision: minimize artificial departmental boundaries; curriculum by interest
- •Research Park as a major enabler for industry linkage, startups, and real-world deployment pathways
- 1:02:36 – 1:09:48
Going viral before social media: early web experiments, forums, and a proto-social network
Ravindran shares how early internet constraints shaped creativity—from snail-mail paper requests to building web pages and forums. His Tamil Film Music and “Forum Hub” projects became unexpectedly large communities, foreshadowing modern social platforms and online collaboration norms.
- •Early internet in India: first email during master’s; paper-sharing via postal mail to authors
- •Personal ‘food page’ as a pre-blog experiment with user-submitted recipes
- •Tamil Film Music page hosted abroad due to bandwidth constraints; later moved to Singapore
- •Built a full discussion-forum backend before off-the-shelf tools existed
- •Forum Hub scaled to thousands of users; seeded projects, collaborative writing, and real relationships
- 1:09:48 – 1:30:50
AI-human futures: online kinship, future of work, AGI skepticism, and IITM’s AI curriculum design
Ravindran explores how online/offline relationships are blending, alongside risks like personalization bubbles. He argues AI will change how every job is done—like Excel did for business—while challenging hype around AGI, and he closes by explaining IITM’s ground-up BTech AI & Data Analytics curriculum philosophy.
- •Relationships evolving: online communities and friendships becoming first-class social structures
- •Societal risks: misinformation, fake narratives, hyper-personalization/echo chambers
- •Future of work: AI as a universal productivity tool; jobs change more than they vanish
- •Key line: AI won’t take your job—someone using AI will; analogy to mastering office tools/Excel
- •AGI critique: ‘general intelligence’ is a misnomer; current systems far from true general problem learning
- •Curriculum design: BTech AI & Data Analytics built from scratch (not ‘CS + a few AI electives’), with stronger math + build/deployment emphasis
