Hi everyone! I’m Shakked, a PhD student in economics at MIT, and this comment summarizes The Short-Termism of ‘Hard’ Economics, a chapter in Essays on Longtermism that I coauthored with my dad Ilan, a Professor of Economics at Victoria University of Wellington in New Zealand.
The chapter is about the academic economics profession. We think the profession has not yet, and will not in the foreseeable future, produce much useful longtermist research as part of the mainstream research it systematically produces, publishes in top journals, and rewards professionally. (Individual economists might still produce useful longtermist research as voluntary side-projects, as we note below.) In the chapter, we argue that this is because the profession is subject to a constricting set of methodological norms that preclude the kinds of research that might be useful from a longtermist perspective.
Specifically, we think that useful longtermist research—by virtue of its speculative nature and the thin historical evidence base it has to rely on—will tend to have a few attributes. It will draw on a variety of sources of empirical evidence, including expert forecasts, narrative historical commentaries, quantitative projections, interviews and focus groups, and so on. It will make theoretical arguments that are often institutionally-specific or draw on a diversity of concepts. As a consequence of these diversities of evidence and argument, it will tend to be multi-disciplinary.
Almost all of the above are ruled out by the current methodological norms of the academic economics profession. An obsession with methodological “hardness”—a concept which George Akerlof says is used to “classify sciences according to a hard–soft hierarchy, with physics at the top and sociology, cultural anthropology, and history at the bottom”—results in the imposition of severe restrictions on the kinds of work economists accept. On the empirical side, this means that only carefully identified causal estimates are permissible forms of empirical evidence. This rules out descriptive, explanatory, or predictive work and narrows attention to topics where sufficiently large micro datasets are available, implicitly narrowing the geographic and temporal scope of research. On the theoretical side, this involves a focus on mathematical generality and technical sophistication and difficulty, which precludes both arguments that are impossible to formalize mathematically and arguments that are trivial to formalize.
The chapter goes into a lot of detail about the exact shape of the norms and applies them to three areas of longtermist interest: long-term decision-making, climate change, and AI.
The chapter was written in mid-2022, before the release of ChatGPT, so the past 3 years have constituted an interesting out-of-sample test of the arguments and predictions we make. We think the chapter has held up pretty well. There’s been an enormous growth in research on AI in economics; the majority of this new research has taken the form of the kind of short-termist empirical or backwards-looking theoretical work we describe in the chapter. There’s also been an increase in interest in longtermist perspectives on AI, including an upcoming NBER conference on the Economics of Transformative AI, as well as someexamples of genuinely useful research. But so far our sense is these developments show no sign of penetrating the top journals or professional reward processes.
Hi everyone! I’m Shakked, a PhD student in economics at MIT, and this comment summarizes The Short-Termism of ‘Hard’ Economics, a chapter in Essays on Longtermism that I coauthored with my dad Ilan, a Professor of Economics at Victoria University of Wellington in New Zealand.
The chapter is about the academic economics profession. We think the profession has not yet, and will not in the foreseeable future, produce much useful longtermist research as part of the mainstream research it systematically produces, publishes in top journals, and rewards professionally. (Individual economists might still produce useful longtermist research as voluntary side-projects, as we note below.) In the chapter, we argue that this is because the profession is subject to a constricting set of methodological norms that preclude the kinds of research that might be useful from a longtermist perspective.
Specifically, we think that useful longtermist research—by virtue of its speculative nature and the thin historical evidence base it has to rely on—will tend to have a few attributes. It will draw on a variety of sources of empirical evidence, including expert forecasts, narrative historical commentaries, quantitative projections, interviews and focus groups, and so on. It will make theoretical arguments that are often institutionally-specific or draw on a diversity of concepts. As a consequence of these diversities of evidence and argument, it will tend to be multi-disciplinary.
Almost all of the above are ruled out by the current methodological norms of the academic economics profession. An obsession with methodological “hardness”—a concept which George Akerlof says is used to “classify sciences according to a hard–soft hierarchy, with physics at the top and sociology, cultural anthropology, and history at the bottom”—results in the imposition of severe restrictions on the kinds of work economists accept. On the empirical side, this means that only carefully identified causal estimates are permissible forms of empirical evidence. This rules out descriptive, explanatory, or predictive work and narrows attention to topics where sufficiently large micro datasets are available, implicitly narrowing the geographic and temporal scope of research. On the theoretical side, this involves a focus on mathematical generality and technical sophistication and difficulty, which precludes both arguments that are impossible to formalize mathematically and arguments that are trivial to formalize.
The chapter goes into a lot of detail about the exact shape of the norms and applies them to three areas of longtermist interest: long-term decision-making, climate change, and AI.
The chapter was written in mid-2022, before the release of ChatGPT, so the past 3 years have constituted an interesting out-of-sample test of the arguments and predictions we make. We think the chapter has held up pretty well. There’s been an enormous growth in research on AI in economics; the majority of this new research has taken the form of the kind of short-termist empirical or backwards-looking theoretical work we describe in the chapter. There’s also been an increase in interest in longtermist perspectives on AI, including an upcoming NBER conference on the Economics of Transformative AI, as well as some examples of genuinely useful research. But so far our sense is these developments show no sign of penetrating the top journals or professional reward processes.