Supercharging Statistics with AI

ETHOS Issue 27, July 2024

With its ability to process data, recognise patterns, and enhance operations, artificial intelligence (AI) has revolutionised many domains, including statistics. Organisations worldwide are integrating AI to streamline processes, gain insights, and tailor services, with public agencies leveraging AI for efficiency and service delivery.

In Singapore, the Department of Statistics (DOS) has adopted an approach towards AI that focuses on robust data governance, engages the workforce to upskill and harness AI effectively, and transforms statistical processes digitally to better address evolving needs for data.

AI and Data: Twin Engines of Growth

In recent years, traditional structure data sources have been supplemented with large and unstructured new data sources, such as social media data, online data, and data from smart devices. The proliferation of these data has enhanced the capabilities of AI and other technologies that thrive on information. In turn, these technologies have amplified the volume and variety of data—both structured and unstructured—by enabling the mass collection and analysis of information from diverse sources, further fuelling developments in AI. This has resulted in a self-reinforcing cycle of technological advancement and data expansion.

AI—a key component of this cycle—is adept at processing vast amounts of information, learning from data patterns, and performing complex tasks with increasing proficiency. This offers PublicService agencies numerous opportunities to enhance their operations and service delivery.

For instance, to enhance the efficiency of fieldwork operations, the Central Provident Fund Board’s Retirement Health Study Office developed a machine learning model to identify optimal time slots for interviewers to call on respondents without prior appointments.¹ In another example, the Housing Development Board collaborated with the Building and Construction Authority, Enterprise Singapore, and National Research Foundation to develop capabilities in drone façade inspection and visual analytics technologies, enabling safer and more efficient building inspections.²

Data Stewardship and Governance: A Cornerstone for Successful AI Adoption

Successful adoption of AI involves more than just technological upgrades. Data quality, quantity and privacy are critical components of successful AI deployment. Each comes with its own challenges. Poor data quality can lead to subpar model performance, while insufficient data may lead to an under-representation of real world scenarios and skewed prediction outcomes. Inadequate governance and lack of privacy assurance undermine trust in data management, making people feel less confident in entrusting their data to organisations for AI learning. It is imperative to have a robust framework to address these challenges.

As the National Statistical Office (NSO),DOS has extensive experience in the statistical domain, allowing it to establish a robust data governance framework³ that comprehensively covers the entire data lifecycle (i.e., from planning, survey design, data collection, data processing, data analysis, and dissemination). This framework complies with Government-set guidelines, benchmarked against standards set by internationally recognised organisations and advanced NSOs, and is regularly updated to ensure its relevance. The framework thus serves as a good knowledge base for adopting key principles of responsible and ethicaln AI use, such as data confidentiality, model transparency and interpretability, and accountability.

Advocacy of statistical best practices and a track record of good data governance has ensured a high level of data quality produced within DOS and across the whole of government. This contributes to the building of a digital government through the Government Data Architecture initiative. In 2019, DOS was designated by the Smart Nation and Digital Government Office as the Trusted Centre for Individual and Business Data to support the implementation of Singapore’s Government Data Strategy. DOS shares core individual and business administrative data within the government under the Public Sector (Governance) Act,⁴ facilitating the use of data for policy analysis and service delivery.

Through engagements with stakeholders such as the local Research and Statistics Units (RSUs), Single Source of Truths⁵ which are custodians of core data, as well as other data producers in public agencies, a collaborative environment is fostered to uplift data stewardship practices and ensure that these are not just understood but actively applied. As data and AI practices are intertwined, the same engagement channels are used to identify and address common challenges. For instance, data privacy is a critical concern for government agencies due to the handling of vast amounts of sensitive information; DOS has thus engaged with agencies such asthe Ministry of Education and Ministry of Health to harness privacy-preserving technologies and other measures to safeguard data privacy.

Continuous Upskilling: From Passive Learning to Active Practice

To effectively leverage AI in the work of DOS officers, continuous upskilling is imperative, given the rapidly evolving technological landscape. Fortunately, statisticians’ traditional expertise in statistical theory and computer programming has already placed them in a good position to pick up advanced AI and Machine Learning (ML) skillsets.

Tailored training workshops and an AI playbook help DOS officers stay up to date with the latest statistical and technological advancements as well as practical AI applications in the statistical and data analytical fields. Additionally, an in-house ML Toolkit supports DOS officers in their AI journey by making it easier for those with limited ML knowledge to build ML models from their data, as well as to appreciate key ethical considerations.

Apart from comprehensive training and tools, it is equally vital for staff to apply their knowledge to real-world scenarios. DOS is nurturing a community of practice with an inclusive culture that empowers practitioners to align AI/ML initiatives with their personal interests, thereby boosting their motivation and commitment to achieving successful outcomes. This strategy is similar to Public Service Division’s Gig Work Opportunities, where officers can apply to participate in posted projects that allow them to contribute their expertise and experience. For instance, the DOS Intelligent Classification Engine (DICE) project team was augmented with officers who were keen to leverage ML to streamline manual classification tasks. This strategy also ensures a variety of perspectives and experiences from diverse backgrounds, fostering innovation, creativity, and peer-to-peer learning.

DOS seconds selected officers to private companies with strong data science and AI/ML capabilities, allowing DOS to glean best practices and innovation from the private sector and benchmark itself against industry standards. Thematic sharing sessions on data science and analytics with other RSUs further enhance efficiency and encourage knowledge sharing across the statistical community. Such approaches can benefit all employees, not only those with technical expertise. Every officer has the opportunity to contribute valuable insights to enhance operations and productivity. Their comprehension of business processes and operational challenges aids in identifying areas where AI can make a substantial impact. Dedicated team channels facilitate these exchanges and encourage all to contribute to the conversations. This promotes a more inclusive learning environment where no officer feels left out from the rapid development of AI.

Digital Transformation of Statistical Processes

Due to budget and manpower constraints, coupled with the need to continue delivering core statistics, DOS has adopted a focused and agile approach to embedding AI/ML and automation technologies into business processes. DOS selects use cases strategically, based on key metrics such as the potential scale of impact, projected benefit, required effort, and the availability of related data.

In 2022, DOS established a Digital Transformation Unit (DTU) to work closely with various Divisions in DOS, serving as a coordinator to identify synergies between AI/ML projects. Operating under a hub-and-spoke model, DTU acts as the central hub, responsible for gathering, refining, and sharing AI use cases and knowhows, to minimise duplicative efforts. Meanwhile, the Divisions, or spokes, leverage their specialised expertise to apply AI/ML in ways that are most relevant to their specific areas of statistical work. The DTU also helps channel relevant talents to ensure that collaborative endeavours have the necessary expertise to kick-start the development of AI solutions and accelerate the time-to-market.

Over time, numerous internal solutions have been deployed to enhance statistical operations in DOS, covering tasks such as data collection, compilation, analysis, and distribution. For example, DOS is going beyond conventional business surveys for data collection by incorporating AI-extracted data from unstructured financial accounts. This reduces respondent burden and increases the scalability of data collection for more comprehensive data of businesses in the economy.

DOS has also used AI to find a novel way of updating the Statistical Business Register,⁶ to improve the quality and efficiency of classification coding through the development of DICE, and to uncover statistical insights with Big Data.⁷

In addition, DOS has participated in the Singapore Government’s AI Trailblazers programme,⁸ which aims to accelerate the development of generative AI solutions that address real-world challenges for organisations. The programme enabled DOS to build and test their own generative AI solutions in a controlled and dedicated cloud-based environment, while identifying potential use cases.

None of these projects used AI for its own sake; the methods were employed only where they adhered to statistical concepts and were deemed most suitable in meeting user needs. Only after the value of an AI/ML method has been established through proof-of-concept experiments, and the model rigorously tested, will the solution be implemented organisation-wide at DOS or offered to the RSUs and the rest of the government.

Conclusion

Successful AI adoption and digital transformation demand more than just technological advancements. DOS has taken a multi-pronged approach to harnessing the full potential of AI and digital technologies. This includes instituting rigorous data management protocols, comprehensive security measures, and privacy safeguards to ensure the integrity and security of DOS’s data. In addition, DOS focuses on enhancing officers’ capabilities through curated training, as they are best positioned to grasp the business challenges and identify impactful ways for AI implementation. By strengthening internal and external engagements, as well as collaborating on digitalisation projects, DOS has enabled knowledge sharing on AI developments and fostered a culture of continuous improvement. These strategies have allowed DOS to harness AI productively, which in turn enhances DOS’s data products and statistical services.

DOS is currently exploring new AI initiatives that focus on data collection, such as expanding our in-house web-scraping capabilities and utilising optical character recognition technology to extract information and code directly from files and records. Additionally, efforts are being made to improve the user experience in searching for relevant information and data as part of data dissemination. These initiatives are aimed at continually enhancing the collection, production, and consumption of statistical data in Singapore. As AI evolves with better capabilities and greater proficiency, DOS expects to discover more statistical use cases for this promising technology in future.