Artificial intelligence (ai) is all set to fast-forward India’s digital economy, enhancing governance, social inclusion and push for an exponential multi-sectoral growth that includes data-driven intelligence and insights. Within the realm of AI, Generative AI (Gen AI) is a major leap from traditional AI models that are primarily leveraged for classification or predictive tasks. Gen AI can don the creativity hat and create novel content across text, images, software code, audio, video, and 3D models by developing deep predictive power through continuous training on large data sets.
Government’s Aims
The Indian government is leaving no stone unturned to put India on the global AI map. The government is boosting AI adoption through many initiatives, such as the INDIAai portal for R&D, an applied AI research centre in Telangana for industry-academia collaboration, Bhashini for vernacular language digital services, and international partnerships like the US-India AI Initiative for knowledge sharing and joint R&D efforts. Further, because AI could create a tectonic shift in the digital economy, the Reserve Bank of India (RBI) and National Payments Council of India (NPCI), too, have entered the fray and are mulling over introducing ‘conversational payments’ on the UPI, which is expected to use an AI-powered system to add an extra layer of security. When launched, this will be a game changer in the payments landscape.
In fact, India’s banking, financial services, and insurance (BFSI) sector is one of the key beneficiaries of this AI revolution. Critical use cases include fraud detection and prevention, compliance and risk, customer servicing, and cybersecurity.
Similar trends can be extrapolated for the Indian banking sector where banks are enthusiastically conducting Gen AI proof of concept (PoC) to test productivity gains and transform customer experience even in the public sector. Fraud detection and prevention is viewed as the most critical use case for Gen AI in financial services. Around 76% of financial services executives say this is the area where Gen AI is likely to find most application in the near- to mid-term. With its ability to consume large data sets to base its predictive outcomes on, Gen AI can be used to identify patterns and anomalies indicative of fraudulent activity. In the Indian context, there are multiple evidence-based examples that complement this global trend.
For instance, one of the leading Indian public sector banks is expected to build new AI models that can help generate actionable insights for accurate business forecasting, facilitate hyper-personalised offers for customers and enable data-led insights for fraud detection, prevention and mitigation. Also, the Department of Telecom (DoT) is expected to introduce an AI-powered platform to prevent frauds using disconnected mobile numbers by real-time flagging to both banks and the Unique Identification Authority of India (UIDAI).
About 68% of KPMG’s survey respondents said compliance and risk is the second priority area for Gen AI in their companies. It can be used to automate labour-intensive compliance tasks, such as SEC filings, bringing a higher degree of accuracy. In risk management, Gen AI could be used to better simulate different risk scenarios and stress test investment strategies and portfolios because of its ability to trawl humungous unstructured data sets. Contextualising for the Indian banking sector, the RBI recently announced its use of AI to gain deeper insights into the operations of supervised entities. The tools that RBI uses encompass early warning system, stress testing models, vulnerability assessments, cyber key risk indicators, phishing, and cyber reconnaissance exercises, targeted evaluations of compliance with know your customer (KYC) and anti-money laundering (AML) norms, and data analytics.
Customer servicing and marketing/sales is tagged as the third most popular Gen AI use case in the banking sector. Financial services companies have already adopted AI to improve customer-facing processes, such as help desks and robo-advisor services. About 66% of financial services executives surveyed say their organisations are likely to use Gen AI to enable more sophisticated chatbots and virtual assistants, and 62% predict it will be used for customer services and personalisation. In fact, a majority of the top Indian private sector banks are in advanced stages of adopting private large language models (LLMs), trained on internal data sets, to leverage Gen AI, in order to drive efficiencies across specific functions and processes.
Another important use case of Gen AI is cybersecurity. Due to the fast-improving stature of India across various performance dimensions, it is featured among the top 10 destinations facing cyberattacks. According to the Data Security Council of India (DSCI), cybersecurity defenders detected 400 million instances of malware spread across a network of 8.5 million endpoints in 2023. In cybersecurity, Gen AI trained on enormous data sets, including both malware-related and synthetic data, can predict cyber threats, simulate security scenarios and pinpoint anomalies—providing a richer, real-time defence strategy. CISOs need to be particularly cautious and proactive in their AI cyber strategy as AI can equally equip the hackers as well with Ransomware-as-a-Service (RaaS) models to notoriously lower the threshold levels for launching malicious attacks.
While the Indian banking sector has started seeing green shoots of Gen AI applications and use cases, enterprise-scale adoption will be a bumpy road full of challenges and bouts of adrenaline spikes.
Hallucinations: Gen AI hallucinations can lead to the generation of factually inaccurate or misleading information, which can adversely impact decision-making, communication and customer interactions leading to reputational risks if not controlled and governed appropriately.
Data security: The most critical responsibility of a bank is to protect Personally Identifiable Information (PII) data of its customers and partners. While most LLM models are stateless and do not store the prompts, it is imperative to either create private LLM models or cautiously use open source LLM models in air-tight containers in the banks’ infrastructure.
Infrastructure: Running LLMs require high-performance servers or cloud computing resources with Graphical Processing Units (GPUs) or Tensor Processing Units (TPUs) to support overwhelming computational demands of model training which can have cost/budget implications.
Copyright infringement and data leakage: Lack of enterprise-wide AI governance controls and guardrails can lead to employees ingesting confidential strategies and sensitive data sets into LLMs, which can lead to leakage of proprietary information, which is already raising red flags in terms of the misuse of proprietary data. The need to use large training data sets simply cannot be done away with, if we are to utilise the capabilities of LLMs. So, an appropriate governance model precluding proprietary third-party data needs to be put in place.
Indian policymakers will play a key role in developing guardrails and a robust AI regulatory framework to realise the digital value of Gen AI. But there is a need to focus on putting together guardrails without either complicating or stifling the innovation required to push the cause of Gen AI. Tenets such as fairness, transparency, accuracy, consistency, data privacy, explainability, accountability, robustness, monitoring, updating, and the imperative of human oversight have been much debated in regulatory circles and could soon be formulated as policy. Further, The Telecom Regulatory Authority of India (Trai) presented a report a few months ago stating that an independent statutory authority may be deputed, and use case-based risk-regulatory framework may be formed. With respect to the concerns on usage of proprietary data, new upcoming AI regulations may mandate organisations to ensure bias-free AI model training.
Further, there could soon be customer protection-oriented regulations related to possibility of deep-fakes and synthetic content as well as instructions for transaction on B2C/B2B apps. It will be imperative that AI models undergo sandbox and stress testing before release. While the recently passed Digital Personal Data Protection Bill is a positive first step in ensuring AI readiness, it will be interesting to see how India transitions from established policy position and recommendation phase to that of implementation.
To sum up, it is portended that India will soon establish a leadership position in the global rankings of Gen AI adoption. And, in this endeavour, the banking sector too is well-poised to use Gen AI and leapfrog into the future.
A version of this article was published on Mar 31, 2024 by Business Today.
