Kok Yam Tan, Deputy Secretary (Smart Nation and Digital Government), Singapore Prime Minister Office
The views expressed in this article are those of the authors. Their statements are not necessarily endorsed by the affiliated organisations or the Global Challenges Foundation.
Artificial intelligence based on machine learning could revolutionize the very core of the governance process. It could overcome the biases and limitations of the human mind, and offer a radically new way to address highly complex problems such as climate change. How can governments and global institutions learn to deploy such capabilities? Cities offer a good testing ground, and Singapore has made some early inroads. However, to harness the full possibility of machine learning, we must invest in the know-how to deal confidently with questions of ethics and risk management.
At the Montreal international airport, near the immigration counter, a sign proudly declares, “the border goes digital”. In that context, the message is clear: border controls are being automated and shall be a lesser pain to all who pass. Strip away the airport context, and the statement is deliciously ironic. Modern digital technologies, by their nature, invalidate physical borders, they do not augment them. Their reach and operating span transcend distances and boundaries. In this digitally connected world, what new opportunities and risks come? And how can and should nations and governments now work together?
There are many emerging areas at the intersection of governance and technological development where greater international exchange of knowledge is needed. These include particularly new models of city management powered by sensor networks and big data, as well as new forms of monetary transactions through blockchain technology. There is one area, however, that may revolutionize the very core of the decision making process itself: it is the increasing potential for machine learning to support administration and governance.
Automation by machine learning is quite different from automation by rule-based algorithms. The machine is not controlled by well-defined protocols or parameters; instead, it autonomously develops its own problem-solving rules, in a way that is often opaque to its creators themselves. No human directly established the rules that determined Alphago’s decision; it “figured out” what to do from a huge set of game data, and thus it could develop moves and responses that surprised every human player of the game.
With increasing advances in artificial intelligence, there is no doubt that governments must learn to deploy such capabilities to serve their constituents better. For Singapore, a key thrust of our Smart Nation initiative is to develop capabilities in machine learning at the national level. Among other endeavours, we have invested in a significant R&D programme called “AI Singapore”. Within the government, we are also beginning to scratch the surface in the use of machine learning on a daily practical basis. Early use cases tend to involve needle-in-a-haystack problems where “judgment” needs to be scaled, because rules are not easy to codify, and there are not enough humans to go around and give human evaluation. One example is the task of picking out anomalous transactions from voluminous sets of data points for audit purposes.
Inevitably, though, governments will come to the point where it becomes technically feasible to use machine learning more broadly, either to replace human judgment (for example, determining who is at fault in minor traffic accidents) or even to substitute codified rules themselves (for example, determining who requires more insurance coverage). In the long term, harnessing this technology for governance purposes can help us overcome the biases and limitations of the human mind. Our cognitive abilities may not allow us to easily grasp some of the greatest challenges, particularly where it is a complex problem that involves systems of systems, where the direction of causality is not clear, and where — perhaps most confoundingly — we ourselves are part of the system. At the global level, climate change is one such challenge. Machine learning can be the basis for a more effective decision making process yielding significant progress towards sustainability.
However, there is a difficult problem to overcome, because effectiveness of decision making is not the only consideration. Human evaluators, as well as written codes and rules, exist also to explain and communicate decisions to those who are subject to those rules. Society and organizations can scrutinize the decisions, by scrutinizing the rulebook and/or the people who made the decisions. We can debate their validity, and alter them over time.
How is that possible when decisions are made in the black box of a learning machine? How do we maintain, influence and alter such a machine to ensure the validity of its decision making over time with changing needs and societal values? These are in many ways technical questions, but they have societal, policy and organizational implications.
As the field of artificial intelligence advances to be more powerful and pervasive, it becomes increasingly necessary for governments to invest now in the know-how to deal confidently with such questions; the other option would be to do ourselves a disservice by avoiding the technology altogether. Moreover, governments’ role does not merely stop at using the technology for itself, but in regulating the broader use of machine learning in banking, healthcare, insurance and other industries. Regulation must be informed by good science; hence the importance of research in AI governance, risk management and safety, particularly to address the challenge of aligning AI with societal values that will change over time. These issues tend not to attract as much commercial interest, because the benefits are diffused and the outcomes are not directly monetizable. But if not addressed, the amazing potential of machine learning may bring about serious threats that we do not fully comprehend.
Early movers in this field — industries and governments — should come together with researchers and technicians, and develop common language and frameworks to address the issue of managing learning machines. Such work should be grounded in, and be relevant to, the practical applications of machine learning for now and the near future. For this, cities offer a testing ground for models that, in the longer term, could revolutionize not only local, but national and global governance institutions.
The impact of the rapid advances in digital and smart technologies can be transformational to people and businesses. As Singapore pursues its Smart Nation initiative, it becomes increasingly clear that cities and governments will require deep know-how to deploy, maintain and govern these emerging technologies. This work is too important, and too multi-faceted and complex, for any party to pursue alone. It would be ironic if we do not connect our thinking in this, when the very nature of the technology itself is inter-connectivity. We would be putting new wine in old wineskins. We can still make digital borders, but we will have to cross them to truly advance digitally.