Technology to solve global problems: an emerging consensus for green industrial policy?

Even as most mainstream policy analysts support the idea of active industrial policy to create new green industries and cut carbon pollution, important dissenting voices still question whether government intervention is possible without extreme waste. We suggest that many of today’s debates, which echo debates of the 1970s, need updating to reflect the reality that a lot has been learned about where and how government can pursue effective industrial policy. The more transformative the goals, the harder it is to know which policies, technologies and business models will work, and the greater the need for ‘experimental’ approaches to policy that put uncertainty as the centerpiece. Creating industrial transformation in the context of deep uncertainty is the central challenge for industrial policy. Solving this problem requires not just attention to policy design and industrial response but also possible reforms to the institutions that design and implement policies. Today’s policy institutions, like today’s firms, are mostly organized for the current industrial system—not necessarily the future.


Introduction
Scholars have long agreed that innovation is central to long-run economic growth and the expansion of human welfare (Schumpeter 1934, Solow 1956).Extending that logic, innovation is also central to shaping how problems of global change arise and are managed-a topic that Arnulf Grübler helped to pioneer (Grübler 1998).Indeed, social goals such as stopping the buildup of warming gases in the atmosphere will require technological change at an unprecedented pace in nearly every major sector of the global economy (Victor et al 2019, IEA 2020).
While there is widespread recognition that technological change must play a central role, the consensus frays when the subject turns to policy design and to choosing or creating institutions that can orchestrate the most effective industrial policy strategies.What should be the role for government in funding-perhaps even steering-technological change?This question is controversial in part because it arose most poignantly in the wake of the 1970s energy shocks-a founding time of IIASA, where Arnulf worked for most of his career and when IIASA's first most prominent contributions to public discourse happened (Häfele et al 1981).Just as governments and the public called for policy interventions to shore up the security of energy suppliesalong with cost and environmental footprints-many western governments were simultaneously trying to pare back 'the state' to make room for greater roles for 'the market' (Yergin and Stanislaw 1998).
In many ways, the academic and policy debates today rhyme with those of the 1970s with bold calls for government interventions as well as strident voices of dissent.There are differences in nuance, of course: today's calls are more about massive 'green new deals' (Mazzucato 2021) alongside traditional goals such as shoring up local energy security and independence (Siripurapu and Berman 2022).The Russian invasion of Ukraine has reminded voters and governments of the importance of energy security, a topic that had drifted from policy priorities for many.A loud chorus on the left sees opportunities to reinvent large segments of industry (Rodrik 2009, Mazzucato 2021).A smaller but vocal group of dissenters is warning that the state is not, in any form, equipped to intervene in the economy and create new futures (Neely 1993, Devarajan 2016, Lincicome 2021).Leaning toward dissent are policy experts who warn about the sheer difficulties in creating and implementing effective industrial policy-for doing so requires government institutions that are highly skilled at knowing when and where to intervene while also avoiding capture by interest groups that favor the status quo rather than disruptive novelty (Agarwal 2023).
Even as these debates have played out among academics and in politics for decades, they remain heavily ideological and risk lagging behind today's realities.In the 1970s and 1980s the debate focused almost existentially on whether 'the state' was much too large and whether government was even capable of intervening to steer the economy effectively (Cohen and Noll 1991).Since then, both sides of the debate have won important victories.Across much of the economy 'the state' imposes a lighter hand on the economy than before because markets play much bigger roles in determining the allocation of capital and effort (Yergin and Stanislaw 1998).And yet, at the same time, advocates for industrial policy have demonstrated that state intervention is possible not just theoretically but practically.Many policy programs have performed well (NRC 1999, DOE 2001).Most recently with the passage of the US Inflation Reduction Act and other national responses, governments have sent a clear signal: industrial policy is back, if it ever left.The question, increasingly, has turned not to 'whether' but 'how' (Rodrik 2009).
Whether government is good at steering the economy or not, current policy trajectories show that many governments are doing just that.We academics need to get ready to help and avoid getting bogged down in old debates about whether to intervene at all.

To intervene or not to intervene, that is not the question
In many ways, today's academic debates are lagging reality because academics have failed to make headway on key intellectual questions, such as (a) how variations in government institutions affect the design of new policy, (b) how policy interventions actually affect the direction of technological change, and (c) how political forces shape what's 'politically feasible' in ways that make industrial policy even more indispensable for addressing problems like climate change.These have long been important questions, but answers have remained elusive in part because theories of cause-and-effect have been hard to pin down; measurement, too, has been challenge.Despite the challenges, offering crisper guidance on 'how' to implement industrial policy will require crisper insights on these three fronts.
First, scholars must continue to learn how national contexts come to bear on making good industrial policy.Ever since the 1980s it has been clear that different governments intervene in different ways, with widely varying impacts on economic growth.In the study of innovation this was best embodied in the 'national systems of innovation' literature (Lundvall 1992, Nelson 1993).In comparative politics this was reflected in the idea that there were 'national varieties' of capitalismwhere individual nations put their 'stamp' on the allocation of effort and resources between state and market and civil society (Hall and Soskice 2001).These were intriguing lines of scholarship, but they did not sustain much interest among scholars of innovation and technology, in part because of the challenge of measuring country-wide patterns and in part because of the recognition that innovation happens in an increasingly global context.Instead, both lines of scholarship shifted toward more quantitative research, which necessarily focused on more easily quantifiable micro-factors within countries and industries to describe macro patterns.As more governments intervene more aggressively in their economies it is time to resurrect these questions of national patterns (Meckling et al 2022) in how different countries respond to the challenges of global change.It is undeniable that national capabilities and institutions are major factors in shaping how governments are choosing to intervene.Some of these national factors affect the resources in play-such as the sheer size of the US Inflation Reduction Actalong with age old questions of balance between state and market.
Second, scholars have struggled to understand and model the processes of innovation-the links between the emergence of new ideas, the funding of experiments, the creation and testing of novel technological systems, and the role of policy in those innovation processes.For decades, the standard model was a 'pipeline' that started with novel ideas and eventually, after crossing numerous funding and organizational valleys of death, viable new products and services.The government's role is early in that processas a slug of funding and research to ensure that science creates an 'endless frontier.'(Bush 1945) In the language of economists, scientific knowledge is a public good that the market, left to itself, will not supply adequately; government incentives such as grant subsidies are essential (Jaffe et al 2005).Other approaches have pointed to a variety of different innovation models, some that rely less on a pipeline and more on ideas that emerge directly from the innovator to a viable products (often mediated by a regulator) (Stokes 1997).Everyone knows that the innovation process is steeped in uncertainty and is loopy and circularfar from a pipeline.But absent a few valiant efforts to model these processes better-including one led by Arnulf (Grübler and Gritsevskii 1997)-we have not much bridged the intuition that innovation is complex with the analytical tools used in most of the models that help guide energy policy.Today where technological innovation is considered at all it tends to be with learning curves and based on the idea that we know, ex ante, the slope and shape of those curves.
Getting this right matters because a lot of what's happening in the 'energy transition' today is not just about the emergence of new technologies that are cleaner but also a shift toward technologies that have different innovation models-often models where ideas move much more quickly from Eureka moments to viable commercial products and with bigger spillovers from outside 'energy.'Much of what has happened with distributed energy resources, grid modeling, improved batteries and the like reflect this shift in innovation model-away from one with long pipelines and toward models that benefit more from spillovers and from 'direct to market' innovations (National Academies of Science, Engineering, and Medicine 2021).
Third, scholars have struggled to update policy models to fit modern understanding of what innovation policy will look like, given what is politically feasible.It has now become widely recognized that the 'first best' policy strategies that have been advocated for a long time-such as carbon taxes and marketbased cap-and-trade systems-will not do most of the hard work to push the economy toward much cleaner futures (Victor and Cullenward 2020).This recognition arises for many reasons, such as the realty that climate change is the result of multiple market failures not merely the externality of unpriced pollution.But the most important failure is political-carbon taxes and other market-based policies are highly transparent in their operation, and that transparency is toxic to many important political constituencies.
For all the awareness that market-based strategies for addressing global environmental ills will not do most of the hard work of creating new industries, the academic debates remain stylized.Most models still mimic 'policy' in the form of a price even though, in reality, 'second best' strategies such as heavy subsidy and industrial policy are the actual means of inducing major changes.(These subsidy-oriented strategies came to the fore at a time when interest rates were very low and huge COVID-related stimulus spending weakened political opposition to excess government spending; whether these programs can be sustained as those contextual factors shift remains to be seen.)Bringing these real-world political, organizational, and human factors into models will help make them better represent reality (Peng et al 2021).And abandoning the pretense that stylized policy instruments work in reality will help force the modeling community and the real policy community to align better and collaborate.
The concept of 'political feasibility' is not static, of course.What's infeasible today might become more feasible tomorrow.When it comes to carbon taxes, for example, feasibility may rise as more governments grapple with the fiscal consequences of subsidy-heavy industrial policies-creating the need to raise revenue along with the value of taxing pollution externalities.And perhaps most important is that feasibility depends on political organization.As more industries emerge that have a stake in cleaner futures what was previously infeasible becomes a new political norm-with huge levels of uncertainty around those processes.

Designing for uncertainty
Our research suggests that an essential starting point would be an effort to grapple much more fully with uncertainty.Even as we make progress in all the factors identified in the previous section, the interactions between unknowns means that highly transformative industrial policy-which is the kind of industrial policy that is most needed and most difficult to design-must grapple with deep uncertainty.This is not the familiar problem of risk (Knight 1921)-where outcomes are not known but the ranges are-but rather a problem of deep uncertainty where choices must seek new information along with adaptation to that information (Morgan and Henrion 1990).
One place to see this uncertainty in action is to look at real world changes in technological performance as a function of experience-what many people call 'learning.'With a wry sense of humor, Arnulf has reminded us analysts that experience does not automatically generate learning nor even improvementwith his most striking reminder from his study, shown in figure 1, of the performance of French nuclear plants (Grübler 2010).Also shown on figure 1 is the learning curve for solar.Today it is widely assumed that new renewable technologies will follow the solar path-and that the attributes of new technologies such as modularity and high replicability (Wilson et al 2020) will unleash learning (Grübler et al 1999).But the nuclear case is a reminder that technological, political and regulatory factors can all conspire to drive up costs and narrow technological options.
How can we square the reality that technological change is uncertain-and the more transformative the changes the greater the uncertainty-with the need to make decisions to back particular technologies?Industrial policy, after all, is about allocating resources-making decisions to back some horses and not others.
The key to answering that question lies with institutions.Effective industrial policy requires building (and maintaining) institutions that are good at learning how to make decisions in the context of high levels Industrial policy when uncertainty is high requires building and nurturing special kinds of institutions that are highly skilled at decentralizing the processes of experimentation and searching for new knowledge and then centralizing the process of assessing what works.
Over the last few years, working with Chuck Sabel, it has been possible to identify the patterns of institutional design that are good at performing these functions-what Chuck and David call 'experimentalist' institutions (Sabel and Victor 2022).Key actors are motivated to invest in experiments that are likely to destabilize the status quo because they see no alternative but to shift.They are under pressure, such as from the public or governments, to change old methods of doing business such as those that pollute.But they do not know what to do, so they run experiments and learn.A key role for government in this process is to help lower the risk of experimentation by, for example, paying some of the cost and helping firms collaborate-so that lessons learned can be communicated across an industry.(Another key role for government is to help firms that are motivated to lead find an incentive to go first-to destabilize their industry and gain a first mover advantage).
It is hard to measure the level of uncertainty that a government decision is designed to address, but one way is to look at the policy instruments selected.In the US, for example, when tax credits are offered-where a known technology or outcome earns a known credit-then uncertainty generally is low because government offers the credit to all valid applicants.('Second best' climate policy is one reason government does this-if government is unable to tax externalities then it might do the opposite by subsidizing low-externality technologies.)But when uncertainty is high then it is not clear which technologies and processes to tax or subsidize.In those settings policy must fund well-designed experimentsallocated across a diverse portfolio of options, with higher uncertainties leading to a more diverse portfolio.Figure 2 represents efforts to apply these categories to how policy is designed.'Tax preferences' are subsidies that flow to any purveyor of qualified technology-a good way to subsidize deployment of technologies with known performance.By contrast, direct 'innovation support' is essential for more experimentalist approaches.(Even then, just supplying money for experiments is no guarantee the money will be spent well-which is why institutional design is so important.)Most US government spending goes to tax preferences, suggesting  (JCT 2020(JCT , 2022)), compiled using methods as similar as possible to the original CBO report, which relied on the same data sources.US DOE Innovation Support includes funding for basic energy science, energy efficiency, and R&D for fossil fuels, nuclear energy, renewable energy, and electricity delivery and reliability-the main programs, according to the CBO, that 'promote the development of specific fuels or energy technologies or further the scientific knowledge on which those new technologies rely' .The calculation excludes many other programs within the DOE, including the DOE Office of Science (for example, funding for fusion energy sciences and high energy physics), as well as any R&D programs outside of the DOE, and thus may be an underestimate of total direct federal support for energy innovation.US Tax Preferences includes the estimated cost to the government of energy-related tax preferences supporting the production and use of fossil fuels, nuclear power, renewable energy, and energy efficiency.Since the 1970s the allocation of these costs across each of the four categories has shifted dramatically: Between 1985 and 1990, 69% went toward fossil fuels, and only 24% to renewables.Between 2010 and 2015 that ratio flipped, with 21% for nuclear and 58% toward renewables.
that government probably needs to take bigger risks and spend more in high uncertainty modes.
If we in the analyst community want to become more relevant to the industrial policies that governments are already advancing, we should do more at the intersection of technological change and uncertainty.We should pioneer better methods for assessing the relationships between policy design and uncertainty, such as implied by figure 2, and we should convene scholars who are measuring the effectiveness of instruments for industrial policy on a regular basis so that different methods can be combined.
Looking beyond measurement, we must also learn more about institutional design and how it affects the abilities of governments and firms to create and steer new technological outcomes.Foundational work by Arnulf helped reveal the true depth of the technological uncertainties involved.Those uncertainties create huge challenge for policy-where choices must be made about which ideas to back, and where organized interest groups typically favor the familiar even as industrial policy, by design, is about backing the unfamiliar.Now that industrial policy is back in the foreground, it is encouraging to see so much scholarly attention.To be most effective, we should focus our spotlight on how policy institutions confront uncertainty and how, with reforms, they might do better.

Figure 1 .
Figure 1.Global learning for two clean energy technologies: nuclear and solar.Global solar cost and capacity data are from IRENA (IRENA 2021, 2022, 2023), compiled by Greg Nemet and colleagues, and updated by these authors.French nuclear cost data are from Grübler (2010), converted to 2020 USD.French nuclear capacity data are from the World Nuclear Association (WNA 2023).We label each extreme with the year corresponding to the cost and capacity data point.

Figure 2 .
Figure 2. Two types of US government support for energy technology: tax preferences for technologies with known performance and direct innovation support, ideally for technologies steeped in greater uncertainty.Data for 1985-2016 are from the Congressional Budget Office (CBO 2017).Data for 2017 on are directly from DOE annual budgets and the Joint Committee on Taxation(JCT 2020(JCT  , 2022)), compiled using methods as similar as possible to the original CBO report, which relied on the same data sources.US DOE Innovation Support includes funding for basic energy science, energy efficiency, and R&D for fossil fuels, nuclear energy, renewable energy, and electricity delivery and reliability-the main programs, according to the CBO, that 'promote the development of specific fuels or energy technologies or further the scientific knowledge on which those new technologies rely' .The calculation excludes many other programs within the DOE, including the DOE Office of Science (for example, funding for fusion energy sciences and high energy physics), as well as any R&D programs outside of the DOE, and thus may be an underestimate of total direct federal support for energy innovation.US Tax Preferences includes the estimated cost to the government of energy-related tax preferences supporting the production and use of fossil fuels, nuclear power, renewable energy, and energy efficiency.Since the 1970s the allocation of these costs across each of the four categories has shifted dramatically: Between 1985 and 1990, 69% went toward fossil fuels, and only 24% to renewables.Between 2010 and 2015 that ratio flipped, with 21% for nuclear and 58% toward renewables.