Artificial Intelligence has the potential to significantly enhance our digital public infrastructure by bringing variety to a DPI approach that has so far been optimised for standardisation.

This is a link-enhanced version of an article that first appeared in The Mint. You can read the original at this link.

Last month, at an event in Bangalore, a select audience was given a glimpse of what the future of artificial intelligence (AI) might look like. In the room were companies from across the AI ecosystem, presenting what they were building and getting to know what else was going on in the space. There were product demonstrations, ecosystem presentations and workshops, all of which were rolled up into a grand vision of what it would take to make India the AI-use-case capital of the world.️️️

I came away with a number of new insights from the event. For instance, I realised that compute does not have to be centralised in a massive data centre, but could instead be distributed across a number of micro data centres accessible over an interoperable protocol. Or that we do not need to wait for a new AI law to come into force if every Indian AI company can pledge to abide by a code-of-conduct that will guide their innovation in this space. But, of all the ideas that were presented that evening, the one that really gave me pause was how AI could radically expand the reach of our digital public infrastructure (DPI).

Standardisation

An essential feature of the DPI playbook is the unbundling of traditional processes so that they can be rebuilt as DPI solutions. This is very often the only way to transform existing services so that they can reach population scale. Standardization, therefore, lies at the heart of effective DPI design. It ensures that the largest cross-section of society can avail these services, and that, regardless of which entity is providing the solution, the underlying service is consistent and reliable. DPI solutions that have been built using standardised protocols are capable of supporting cross-sectoral applications.

The trouble with standardised solutions is that they are unable to account for diversity beyond a point. By optimising for scale, the DPI approach has sacrificed variability. To compensate for that, we let the private sector access and leverage DPI, trusting that market forces would encourage them to innovate in order to meet the diverse requirements of their customers. While there are numerous examples of how private innovation has enhanced DPI solutions, AI can exponentially accelerate this.

Unlike traditional systems that do not easily adapt to the needs of a richly varied population, AI can dynamically adjust to individual preferences and contexts, progressively making the delivery of public services more responsive and user-centric. Rather than forcing users to learn how to use new technology systems, AI can learn people’s preferred modes of engagement and adapt. It can address linguistic and cultural diversity by adapting to suit local contexts, rather than forcing users to upgrade their digital skills.

DPI and AI

To illustrate the potential of how AI can enhance existing DPI, Setu, a financial services company, demonstrated an AI-enhanced personal financial management (PFM) app. The app it showcased was built entirely inside a WhatsApp bot, presenting a familiar environment within which users could interact through a conversational interface. The way it works is that it first uses the Account Aggregator system to pull whatever information it needs to analyse the user’s financial behaviour. Having identified sources of income, current lending history and individual spending patterns, it can assess the user’s discretionary spending capacity. Having calibrated this baseline, the chatbot is ready to offer the user advice on purchase decisions.

For instance, if the user wants to buy a refrigerator, the app can integrate with the APIs of various popular e-commerce sites to identify the best available deal. It can then trawl through the full range of financing options available to the user and identify the most favourable. Finally, it can crunch all this data and present the user with 4-5 options that are best suited to her individual financial circumstances.

As impressive as this might seem, none of it is particularly novel. There are already a number of services that provide users with similar financial management services which, once they have been integrated into the Account Aggregator ecosystem, will be able to achieve similar outcomes. What differentiated this from anything I had seen so far was the intuitive and conversational manner in which users are able to interact with it. Even for the most financially savvy, PFM is a daunting subject. What the demo showed was that it is possible to navigate complex issues like this relatively easily if people are engaged using a friendly chat interface within a familiar online environment.

Diversity Through AI

This is what has been missing from our DPI playbook so far. While we have successfully scaled our DPI solutions to reach the entire population, we have struggled to do so in a way that adequately addresses the diversity of our population. As useful as it has been to allow private parties to innovate on the edge, we have found that they still struggle to serve users with low resources and low capacity to deal with digital services.

AI can bring variety to a DPI approach that has so far been optimised for standardisation. It can do so by dynamically adjusting to individual contexts while still remaining scalable. As more and more of our companies start to integrate AI into the DPI solutions they provide, we will finally be able to bring diversity to the solutions we scaled through standardisation.

It is only once we do this that our digital public infrastructure approach will truly be of benefit to every citizen.