The past couple of months have seen a flurry of activity related to the regulation of artificial intelligence. The US White House issued an executive order for the regulation of AI, the European Parliament and the European Council have agreed on a legislation to regulate AI, the UK hosted a summit on AI safety that resulted in the Bletchley Declaration, and at the GPAI summit in Delhi this week, participating countries agreed to create a global framework on AI trust and safety, among other things.
This wave of high level discussions will soon begin to highlight the downstream challenge — the urgent need for regulatory skill-building in the digital age. This is the problem of being able to develop capabilities at roughly the same pace as the pace of new risks emanating from new technology. While governments, as a whole, have a role in confronting the nascent challenges being thrown up, the specific arms of governments, regulatory agencies, will be at the forefront dealing with this issue.
This piece discusses the need to rethink regulatory capabilities given the potentially transformative impact of AI in sectors like banking, telecommunications, and insurance. Regardless of whether AI has the potential to be self-sustaining in the future, two consequences of the deployment of generative AI products have already become apparent. The first is the vast scope of use for the technology. The second is the rapid improvement in the quality of services on offer. ChatGPT is a whole lot better than it was on the day it was launched. This in turn means that the potential for adoption across the economy becomes even stronger.
The Economist Intelligence Unit reports that banks and credit card companies have started using AI for fraud detection, risk assessment and digital marketing. E-commerce companies are using AI to predict credit risk and personalise services. The Indian insurance industry now uses AI for risk-management. Similar uses are being adopted in other parts of the economy. While these are examples of limited use, AI usage may become much more prevalent soon as the technology improves and people start understanding how to use the technology better.
The Reserve Bank of India and the Securities and Exchange Board of India have both recently begun to develop AI tools for regulatory supervision. Since 2019, SEBI has required mutual funds to disclose the use of AI in their product offerings and product managements. However, they and other regulators will need to do much more to prepare for potentially transformative changes.
AI may also alter professional practices and norms. The use of AI for book-keeping and accounting practices will potentially change the way a chartered accountant and an auditor works. Similarly, AI-drafted contracts will reshape the work of corporate lawyers. These changes may be slow at first, but will soon reach an inflection point. With the change in the way the professions work, the role of professional bodies in charge of maintaining the integrity of professional norms and practices is also likely to
undergo a change.
Governments and regulators in some countries have so far implemented nascent regulatory frameworks for the use of AI. While these and other frameworks focus on the substance of AI regulation, regulatory agencies that have to implement these frameworks have to build the capabilities to implement them.
These capabilities are not easy to build in-house. Regulatory agencies will have to be nimble and proactive in order to acquire the necessary skills. The RBI for example, reportedly, has entered into a contract with McKinsey and Accenture to provide advanced analytics to help it discharge its supervisory functions. Even if regulators are able to hire external firms and experts with the necessary skills, they will have to develop the capabilities to evaluate the inputs from these external resources.
For example, some regulators globally have begun exploring the concept of algorithmic auditing. Algorithmic auditing is the audit of each part of a model’s lifecycle to gain a better understanding of how the model works, and whether its use leads to potentially problematic outcomes. However, in order to make use of this practice, regulators will have to develop the capability to understand algorithmic auditing. Similarly, disclosure related requirements will only be useful if regulators know how to evaluate the information being reported.
While market dynamics and consumer preferences may naturally mitigate AI’s rapid adoption, this underscores rather than diminishes the necessity for regulation. Effective regulation can facilitate market acceptance of AI products and services. Relying solely on private sector incentives for regulation, particularly in specific parts of critical sectors like banking and insurance,
will be inadequate.
The important question is not whether this capability should be developed, but how it can be developed at the scale of the Indian state and at speed. In the absence of a central coordinating function, the development of such capabilities will be uneven and ad-hoc. Enterprising regulators will devise mechanisms to rely on the expertise available in the private sector. This is, however, not a sustainable strategy in the long-term. Relying on outside expertise for the discharge of core regulatory functions can become problematic over a period of time.
Therefore, building these capabilities on a systemic scale will require deep thinking. The Indian state has managed to transition from an analog state to a significantly digital one over the last two decades. However, there is no significant body of knowledge that explains how this transition was achieved, that can in turn be replicated to build a new set of capabilities. In the absence of this body of knowledge, the central government must take on the mantle of understanding how to build this capability across government. The main challenge in AI regulation therefore seems to be to develop the capability to develop capabilities.
The writer is associate research director at Carnegie India. Views are personal
No comments:
Post a Comment