Artificial intelligence is moving from systems that answer questions to systems that can plan tasks, communicate with other programmes and act on behalf of users. For cities such as Chennai, that shift creates an opportunity to improve traffic management, flood response and public services, but it also exposes a harder problem: technology cannot compensate for weak institutions, poor maintenance or unclear accountability.
That is the central tension in the assessment offered by Prof Balaraman Ravindran, Head of the Wadhwani School of Data Science and AI at IIT Madras. In an interview with DT Next, Ravindran said the expansion of AI capabilities must be matched by stronger safeguards, safety engineering and responsible deployment. His comments place Chennai’s technology ambitions within a wider question facing cities and governments: who controls systems that increasingly make decisions, interact with other systems and influence the physical world?
The change from conventional chatbots to agentic systems is important because the machine is no longer limited to producing an answer. An AI agent could set a task for another programme, communicate with it and potentially interact with the outside world. That creates more useful applications, but also increases the possibility of actions that developers did not intend or cannot easily control.
Ravindran’s warning is deliberately stark. Even one life lost because of an unintended action by an AI system would be too high a cost, he said. His proposed response is not to stop development, but to build strong guardrails around it. AI harnesses and narrow, custom-built solutions could restrict what systems are permitted to do, reducing the risk that a general-purpose model acts beyond its intended role.
This distinction matters for urban administration. A system used to analyse traffic flows or identify unusual water-level patterns is not equivalent to one authorised to alter signals, issue public warnings or trigger emergency responses without human oversight. The more directly an AI system can affect mobility, safety or access to public services, the more important it becomes to define its limits before deployment.
Chennai’s existing urban problems show why the technology debate cannot be separated from physical infrastructure. The city faces persistent challenges involving flooding, water management, traffic, pollution and urban planning. AI could help improve routing, traffic signals and traffic management. It could also support work on urban water management, wastewater reclamation and flood-related problems, areas in which IIT Madras has expertise, according to Ravindran.
But the interview also identifies the boundary of technological intervention. If storm-water drains are blocked by construction debris, technology alone cannot prevent flooding. Similarly, AI may improve routing or signal coordination, but it cannot remove the physical limits of road capacity. Public transport and supporting infrastructure still have to be strengthened.
This is more than a technical caveat. It is an institutional insight into how urban technology projects often fail. A city may possess a sophisticated dashboard, sensors or predictive model, but the value of that system depends on whether field agencies act on the information, whether maintenance is funded and whether responsibilities are clearly assigned. A model can identify a risk; it cannot by itself clear a drain, redesign a road or resolve overlapping administrative authority.
The same question applies to public trust. Synthetic voices, deepfakes and AI-generated images are becoming easier to produce, but Ravindran noted that misinformation existed before generative AI. False messages circulated on WhatsApp because people accepted claims that aligned with their existing beliefs. AI has increased the scale and quality of the problem by making convincing audio, video and images easier to create.
His proposed response combines public behaviour and industry responsibility. People need to develop a habit of verification, especially when a claim appears too good to be true. Companies should provide clear labels or watermarks for AI-generated and AI-modified content. Since malicious actors may remove those markers, important information must still be independently verified.
For city governments, this creates a communications challenge. Emergency information during floods, transport disruptions or public-health incidents can influence behaviour within minutes. If residents cannot distinguish an official alert from a fabricated message, the problem is not only digital misinformation; it becomes a governance and safety risk. Trust in the source, the verification process and the channels used to communicate will matter as much as the software generating or distributing information.
Cyber security presents a similar dual-use problem. AI can help attackers automate operations and identify vulnerabilities, but it can also help organisations detect unusual behaviour and respond more quickly. Ravindran described this as an arms race between attackers and defenders, with the advantage partly determined by resources and expertise.
That observation has a direct institutional implication for India. A city or public agency may acquire an AI tool without having enough trained staff to assess its vulnerabilities, monitor its output or respond when it fails. The presence of a system does not automatically create resilience. The capability to audit, secure and operate it must develop alongside the technology.
The interview also challenges the assumption that AI progress is defined only by the size of models. Ravindran pointed to work involving the human genome and mutations across genes, including what has been described as the Genome Atlas, as an area with potential healthcare implications. He also argued that smaller models can already perform specific tasks effectively, while AI is already used in voice assistants, navigation and logistics.
This shifts the policy question from how large a model can become to where it can be used reliably and safely. For public agencies and enterprises, a smaller system designed for a defined task may be easier to supervise than a much larger model with broader capabilities. The relevant test is not scale alone, but whether the system works under real-world conditions, at an affordable cost and with safeguards that users understand.
That practical focus is also visible in Ravindran’s assessment of India’s position in the global AI economy. The United States and China have major advantages in capital, computing power and frontier research. India should continue developing frontier capability for technical depth and strategic applications, he said, but it also has a major opportunity in AI diffusion: taking advanced tools to people and enterprises at the right price and making them work reliably in diverse conditions.
Some companies have rolled back AI pipelines because of errors or costs, Ravindran noted. The comment is significant for Indian urban systems, where budgets are constrained and operating environments vary sharply. A solution that works in a well-resourced setting may not perform in a resource-constrained region, across multiple languages or within an agency with limited technical staff.
India’s diversity, in his view, may therefore be a broader advantage than language alone. The country combines highly developed environments with resource-constrained regions, many languages and sharply different social, cultural and economic conditions. Building AI that works across that diversity could produce capabilities relevant to other markets facing similar constraints.
Tamil Nadu has an opportunity to connect that capability to sectors in which it already has data and domain expertise, including healthcare, manufacturing, automobiles, global capability centres and fintech. Ravindran said the state should not wait for AI solutions to arrive from elsewhere, but should bring AI talent together with people who understand these industries and develop applications centred locally.
The institutional gap is not only technical. IIT Madras has launched AI Studio to help turn research into companies, addressing the distance between a promising idea and a viable product. Ravindran said students and researchers may understand a problem well but lack experience in product development, customer acquisition and technical support. Deep-tech companies may also require two to three years before their potential becomes clear, creating a need for patient capital.
This matters because deployment determines whether AI becomes a public capability or remains a demonstration project. Research institutions can develop models, but cities and industries need systems that can be maintained, integrated into existing workflows and evaluated when conditions change. The transition from laboratory success to operational use requires funding, domain expertise and institutional ownership.
The labour-market consequences are similarly more complex than a simple forecast of jobs lost or created. Ravindran described a job as a bundle of skills and tasks. AI may automate some activities while leaving other parts of the job intact. In a business-process role, for example, routine information processing could become autonomous while customer interaction or supervision remains assisted.
Entry-level roles could be significantly affected, while new roles such as forward-deployed engineers and data annotators may emerge. Because the technology is changing quickly, Ravindran declined to give a precise estimate of jobs that AI will create or eliminate. He instead argued that workers should be prepared for changing skill requirements.
The evidence and arguments in the interview point to a consistent conclusion. AI can support better urban systems, but its success depends on the quality of the systems around it: physical infrastructure, public agencies, skilled workers, cyber security capacity, funding and accountability. Chennai’s flood, water and traffic problems cannot be solved by software alone, yet carefully designed tools may help institutions make better decisions when the underlying responsibilities and infrastructure are in place.
What remains unsettled is how governments and companies will define acceptable levels of risk, disclose significant incidents and ensure that human oversight is meaningful rather than symbolic. Ravindran said larger AI companies have a responsibility to self-police and be honest about system limitations, while governments worldwide continue developing appropriate norms and regulations. For India’s cities, the next test will be whether AI is deployed as a substitute for institutional reform or as a carefully governed tool within it.
