Governments worldwide, experience the need to transform from being data-rich to being insights-enriched
The rapid advent and growth of machine learning (ML) and artificial intelligence (AI) promises to enable this transformation. As ML & AI capabilities evolve, there emerges a stronger case than ever, for maximising utilisation of all kinds of data captured by Government bodies be it census, programme performance, weather, cropping. This pro-data wave is evident in various studies. For instance, IDC estimates that the US Government’s investment in cognitive and artificial intelligence technologies will grow at a CAGR of 54.3% between 2018–2021.
However, is Government data primed for Data Analytics/ ML/ AI?
By its very nature, Government data is:
- Even a newbie at ML/ AI will tell you that in case of data, the more the better. Even if we were to look at only district-level data, populous nations will always offer huge repositories of data to mine. Larger volume datasets could often mean higher accuracy outputs.
- Significant portions of Government data is information collected in a structured format. Data is collected in various forms — a subsidy deposited into the citizen’s bank account, an approved loan application, a medical prescription issued by a Government clinic. Handling structured data is relatively easier for machines to understand and is easily searchable by basic algorithms.
- Whether in digitised or digitised form, Government records are maintained over the years; hence an available source of timeline data. This enables data scientists to work with linear data (data regarding the same citizen, collected over several years), which can utilised for citizen-level analysis.
Such characteristics make for an ideal match between Government data and Analytics.
Applying analytics/ ML/ AI to Government data — 4 intervention areas
Like Corporates offer Customers a product/ service, Governments offer citizens certain ‘services’. In large democracies like India (over 40 ministries), tracking service-delivery can be a humongous challenge. Imagine a scenario where citizens could evaluate elected representatives basis a report card, indicating to what extent promised ‘services’ were indeed delivered.
A single, unified tracking system can offer slicing-and-dicing at various levels: by town-size/ electoral wards within a city/ representative candidates within a particular party etc. This would, in turn, deliver 3 key benefits:
- Enhanced constituency knowledge: Data pertaining to performance on all programs/ schemes — be it healthcare/ education/ infrastructure — can reveal rich insight into each constituency’s needs, challenges, available resources. Such a refined, updated understanding can improve resource allocation, program management as well as spends optimization.
- Increased accountability: State-level and Ministry-level performance data helps voters evaluate to what extent, each elected representative delivered to pre-election promises. Furthermore, Government ministry officials could review performance against pre-set goals and course-correct, to ensure that goals are met. In both cases, elected representatives become more aware of their accountability to their managers as well as the public at large.
- Cluster-focused governance and development: Patterns, trends and insights uncovered about impact of Government spends and actions, per constituency can be leveraged to redesign development plans and redirect spends to constituencies in need of support. Citizens’ feedback about completed projects could be reviewed, to enlist their support in future projects’ implementation.
The success of using ML/ AI for performance management depends on several factors eg.
- Whether a data-positive mindset existed, among those in charge of tracking performance as well as those executing programs on the ground
- Whether specific, measurable and time-bound goals were preset, per program/ project
- Whether periodic updates of each deliverable/ metric were made available to those tracking performance, to enable course-correction
This application of AI involves use of cognitive computing and ML to improve management of public bodies and process design. Tracking processes across Healthcare, Education, Subsidies & Benefits can improve governance manifold. Public bodies can be held accountable by citizen groups, if information about their efficiency is made public.
Regulators and Non-profit organizations alike, can adopt a data-first approach to address critical issues through AI & ML:
- Identify blind spots: Officials from ministry offices can speedily spot factors slowing down progress of a particular program and redesign programs to meet goals
- Improve emergency response: Hospitals checking home locations of incoming patients can help understand disease origin and transition. This, in turn, can alert Public Health bodies to redirect preventive measures to a particular location, depending on patient-inflow patterns identified.
- Address resource misallocation: Regulators can use AI to better manage distribution of medical supplies by identifying leakages, fake products and fraudulent claims of medical benefit. AI can also help predict communities and geographies with medicine shortages by studying population, location, disease data and health records.
- Reduce agriculture-related inefficiencies: ML can address challenges such as inadequate demand prediction, lack of assured irrigation, and overuse / misuse of pesticides and fertilizers. Some use cases include improvement in crop yield through real-time advisory, advanced detection of pest attacks, and prediction of crop prices to inform sowing practices.
Evolved data analytics economies have been observed to apply AI/ ML/ Analytics to a wide variety of civic issues:
- Public safety: Predictive Analytics to identify buildings more vulnerable to fire incidents
- Traffic management: AI can be used to develop traffic optimization and control software, that manages traffic flows at several intersections, optimizes traffic systems, reduces travel time/ number of traffic stops/ wait time.
- Backlog clearance: Backlogs of various types — cases, requests, applications etc. can be managed better, using well-trained bots that apply ML to prioritize items basis urgency/ time of raising request and other predefined parameters
Post the 2016 U.S. presidential elections, this is possibly the most commonly talked-about application of Analytics/ ML in the Public Sector. Mining digital platforms like social media, online news, blogs etc. provides an opportunity to understand and/ or influence:
- Public mood
- Public expectations from public servants
- Perceptions about a party/ candidate
- Reaction to manifestos published by a party/ candidate etc.
Such deep understanding can help political parties frame their messaging and public campaigns. In the long run, tracking of public sentiment also helps reframe public policy and schemes. Speech Analytics using voice recognition can also be utilized to gauge shifts in opinion (pre & post speeches).
AI is a recent addition, to the tech stack adopted by parties. AI-proponents claim that predictions arrived at using AI, harness data way more efficiently than survey results do. Moreover, AI can be used to program political bots to step in when people share articles containing misinformation. Or, bots could issue a warning that the information is suspect and explain why. Lastly, it can also be used to deploy micro-targeting campaigns that educate voters on variety of political issues, thus enabling them make electoral choices. For instance, if one is interested in Environment Policy, an AI-based targeting tool could be used to help them identify each party’s stance on Environment.
These applications of AI, however, tend to be controversial, generating much debate. Take for example, the recent debate around political parties using citizens’ digital imprints, to directly influence their perceptions and choices.
ThinkBumblebee researched use-cases in the India scenario, to understand to what extent Governments were leveraging the power of ML & AI
Application of both domains in the public sector is nascent today; possibly constrained by limitations like:
- A majority of Public Sector institutions (other than Banking, perhaps) might be stuck in legacy IT systems, which do not support adoption of ML/ AI.
- Governments’ organizational models might have to be reimagined, to reshape how work is conducted; so as to adopt ML/ AI initiatives.
- Data privacy perceptions, myths and concerns among Government institutions could be a key barrier. Limited understanding of how data should be collected/ stored for ML/ AI would limit acceptance. Similarly, resistance to embrace open source could also slow down, adoption of Analytics.
- Lack of talent pool. Analytics is a ‘sunrise’ industry and Government institutions possibly find it hard to attract/ retain ML/ AI talent.
- Much of data collected is in local languages. Translation technologies become a critical component in application of ML to such data.
Yet, aware of the potential that AI offers, Governments are in early stages of adoption. For instance, the Telangana government entered into an agreement with I.T. industry body NASSCOM to set up a data science and artificial intelligence center with an initial investment of INR 40 crores. In addition, a handful of other states (Karnataka, AP, Maharashtra) have initiated adoption of AI/ ML. The Government thinktank Niti Aayog, entrusted with the responsibility to develop India’s AI programme, released the national strategy for AI in June 2018. Having pushed aggressively for AI adoption, Niti Aayog is now planning to develop an ‘AI readiness index’ that will rank states on their capacity to adopt AI for public service delivery and exploit its potential. This traction is expected to speed up, as more and more State & Union ministries see the uplift that AI/ ML/ Analytics can bring in.