Report to the Technical Commission of the ISSA on Emerging Technologies in Social Security.
In 2016, at the World Social Security Forum in Panama, the International Social Security Association (ISSA) delivered a landmark paper “Ten Global Challenges for Social Security”. This paper was the result of research and consultation with more than 280 governments and social security institutions worldwide. It recognized that governments and public administrations are faced with an increasingly volatile and complex environment, analyzed how social security systems are impacted by these global challenges, and outlined how they can mitigate risks and prepare for the future.
Among the identified global challenges was “The Technological Transition”, recognizing increasing citizen expectations for e-government and self-service, the widening application and penetration of mobile technologies, and the exponential growth in digital datastores. The challenge for social security administrations is to use Information and Communication Technology (ICT) not only for automation of service delivery processes, but as a strategic enabler of innovative solutions in response to these societal transformations.
We at the SAP Institute for Digital Government (SIDG) have been involved in the exploration and implementation of a multitude of emerging technologies, including social security applications of Blockchain and IoT technologies, but none more so or more successfully than that of Machine Learning. For us, this last triennium (2017-2019) has seen the rise of the machines.
Governments and public administrations have traditionally adopted ICT systems for their ability to collect, process and store data. But what has become increasingly apparent is that ICT systems also excel at generating data – “big data”. Many social security systems are now generating data faster than their human operators can consume it. Policymakers and service delivery officers are struggling to see through the fog of big data to derive meaningful information and actionable insights. Perhaps ironically, the solution is to implement another type of ICT system – one that is designed to efficiently consolidate, navigate and investigate big data – a Machine Learning system.
Machine Learning is a type of Artificial Intelligence (AI) that allows software applications to become progressively more accurate in predicting outcomes, without being explicitly programmed. For example, Machine Learning can be applied to predict the probability of certain outcomes for individuals, based on observed patterns in historical data of people with a similar profile, and to recommend interventions that have been successful under similar circumstances in the past. A key feature is the machine’s ability to autonomously learn from its experiences, considering not only historical data, but also data generated through applied interventions and the resulting outcomes.
Example applications of Machine Learning (ML) in social security administration include:
- An Australian government agency is using ML to predict that a customer might go into debt, with the potential to recommend early interventions to help avoid future debts;
- A Korean government agency is using ML to identify suspicious data when matching asset declarations across agencies, with the potential to improve data integrity and compliance; and
- A UK government agency is using ML to proactively identify households and individuals who might be eligible for additional social security benefits, with the potential to recommend targeted interventions for vulnerable people.
All these examples share a common design principle, which we at the SIDG call “human in control”. Due to a combination of factors – ML’s status as an emerging technology, inconsistent data quality inputs, and unresolved ethical questions around AI – governments and public administrations have so far stopped short of realizing the potential for fully autonomous Machine Learning systems. For most social security agencies, AI currently stands for “Augmented Intelligence”, in which machines uncover potentially useful insights buried in big data holdings and present these in a consumable way (often using a visualization) to their human controller. It’s ultimately up to the human to judge the quality of the machine’s recommendations (including the potential for data bias), and to decide on the appropriate course of action.
Requirements for algorithm transparency and decision traceability also represent significant hurdles in the widespread adoption of advanced Machine Learning techniques (such as Deep Neural Networks) by governments and public administrations. Deep Learning algorithms, by their nature, operate as a “black box”, autonomously learning from sample data and presenting findings as probabilities – not via the decision trees and tables that users of rules-based systems are accustomed to. This makes it difficult to trace the rationale for ML recommendations, which can be problematic for caseworkers who need to explain decisions to citizens, auditors who need to ensure that decisions are consistent with social security legislation, and executives who might be asked to justify agency actions to government Ministers. In response to this challenge, the SIDG is partnering with leading Universities to undertake research into “AI Interpretation and Consumption in Public Services”.
This triennium, we at the SIDG have been encouraged by the eager exploration and adoption of emerging technologies by social security agencies. Through the application of Machine Learning in particular, governments and public administrations are taking proactive steps to address the ISSA’s identified global challenge of “The Technological Transition”. We look forward to continuing to work with the ISSA, academia partners and our social security customers on the challenges of the next triennium.
This report was cited by the ISSA’s Technical Commission on Information and Communication Technology in their triennium report “Applying Emerging Technologies in Social Security”, which can be downloaded at: