Pinned
A science of machine learning
Machine learning is becoming an empirical science of its own artifacts. I want its theoretical and empirical findings to be synthesised into powerful explanations, as they are in physics and biology, rather than remain a collection of disconnected results. The goal is to achieve the economy of thought characteristic of more mature sciences by identifying common principles that unify apparently different phenomena. Much of my research contributes to this effort by studying learning dynamics and the principles behind learning algorithms, with the aim of unifying existing results and methods. I also examine—and seek to improve—the methods by which machine learning generates scientific knowledge. Here, I have found the philosophy and history of science valuable not as commentary from outside the field, but as practical tools for doing better science.
Einstein captured the value of this perspective when he wrote:
“…independence created by philosophical insight is—in my opinion—the mark of distinction between a mere artisan or specialist and a real seeker after truth.”
Albert Einstein, letter to Robert Thornton, 1944
Philosophy’s most important contribution is conceptual clarification. This is not merely a matter of tidying up terminology. Clarifying a concept can improve the precision and usefulness of scientific language, but it can also open new directions for empirical research. The conceptual framework we adopt shapes the questions we ask, the variables we consider important, the experiments we design, and what we are prepared to accept as evidence.
In particular, experiments rarely stand alone as decisive tests of isolated hypotheses. They are conceived and interpreted within a wider network of theoretical assumptions. In this sense, observations are theory-laden: what researchers notice, measure, and report is partly shaped by their prior concepts and commitments. Even empiricism presupposes some theoretical framework, because an observation can become evidence only when we have an account of what is being observed, why it matters, and what conclusions it can support. Philosophy helps us identify these background assumptions, making them available for scrutiny and, where possible, empirical testing.
Machine learning remains a relatively young scientific field. The history of science therefore offers grounded case studies of how other fields matured: which practices allowed knowledge to accumulate, which conceptual confusions held progress back, and which methodological or institutional traps repeatedly appeared. Studying these cases can help us understand how machine learning might mature as a science and anticipate the obstacles it may encounter along the way. Philosophy can then sharpen the lessons we draw from history by clarifying the concepts and arguments on which those lessons depend.
Taken together, this is the kind of science of machine learning I hope to help build: one that searches for unifying principles while remaining conscious of the conceptual and methodological assumptions on which its results depend. The goal is to achieve the economy of thought characteristic of more mature sciences and translate that understanding into solutions to practical problems, much as physics provides the foundations for mechanical and electrical engineering.
Perhaps you were born for such a time as this
It has long seemed apparent to me that I am living through a consequential period of human history. Since 2025, I have become increasingly convinced that we are witnessing the unravelling of one international order and the uncertain emergence of another. The economic and political settlement associated with the Washington Consensus is weakening: many of its promises have gone unfulfilled, while even its principal backers—most notably the United States—appear less willing or able to bear the costs of sustaining global hegemony.
What does it mean to build durable institutions and govern them effectively in such an era? Since 2025, this has become one of my central concerns, particularly in relation to how a small state such as Singapore can navigate these unfamiliar waters. Through my past ventures and friendships, I have been fortunate to speak with people in and around Singapore’s government, as well as others across the world. These conversations have deepened my interest in institution-building, state capacity, and the relationship between economic policy and foreign affairs.
I continue to study these questions and to participate, where I can, in the conversations shaping our understanding of them. Over time, I hope to make this work a central focus of my life. These are, after all, the times we have been given to live through. I often return to Mordecai’s words to Queen Esther: “Perhaps you were born for such a time as this.” (Esther 4:14).
Essays
Towards a frontier-level scientific ecosystem in Singapore — Part I
On the character of scientific inquiry, and the institutional changes Singapore needs if its researchers are to do extraordinary work.
Read the essay →