Today’s piece is co-authored with Dean W. Ball.
For most of history, human labor has been performed with human muscle. We cut, hauled, sewed, dug, chopped, threshed, pushed, and pulled. For the simplest things we could use animals, but for any physical work that required higher intelligence, the human body was a necessity. Labor, while intrinsically physical, therefore did require intellect, and often quite a bit of cognitive skill. But outside the clergy, ‘knowledge work’ would have been an anachronistic concept.
Around the 17th century, beginning in England, that started to change. Mankind was awakening to a new kind of science that emphasized mechanistic inquiry into the workings of all things. People, particularly elites among the burgeoning capitalist and mercantile classes of Europe, began to see the world through the abstractions of Newton, Bacon, and Descartes. Matter—whether it be the heavenly bodies or leaves detaching from trees—was undergirded by forces that pushed and pulled. They quickly began to use this fundamental insight about nature to re-imagine their own activities of commerce and industry, building machines to push and pull, to take advantage of force, inertia, acceleration, and mass.
These machines came to predominate in industry with time, and multiplied tremendously in variety. Humans who had done things the manual way found themselves losing ground to the humans who learned to collaborate with machines. For the inventors, tinkerers, and conductors of the early Industrial Revolution, this period must have felt wildly dynamic and pregnant with opportunity. For the laborers who had to learn to accommodate the machines, these early decades of economic transformation were often unpleasant: wages remained stagnant, working conditions worsened, and hours labored increased.
It would take decades—more than a century, really—for things to sort out. Now, the great majority of people—whether they are “blue collar” or “white collar” laborers—spend their working hours orchestrating machines of various kinds: some to transform knowledge or bits, and others to transform atoms. Yet just a few decades ago, it would have been impossible to understand what it is that most people today call “work.”
Today, a relatively small group of technologists is starting to see the world through the lens of another fundamental discovery: deep learning, the approach to AI that has enabled machines to think and undergirded substantively all major advancements in AI over the past decade. And like their forebears at the beginning of the Industrial Revolution, these technologists are building new machines, uniquely enabled by the insights and abstractions furnished from the new science. Some believe new types of labor will emerge, concentrated on the orchestration of machines, or the tasks that remain best suited to the human touch. Others believe this time is different, and that human labor will soon be permanently obsolete.
We do not pretend to know the definitive answers. What we do know is that much of this future remains to be written, in no small part by the policy choices we make today. And what we hope to offer is a roadmap for how politicians and policymakers might bet on human agency under stark uncertainty.
Futures Not Yet Written
There are two fundamental stories one can tell about the impact of artificial intelligence on human labor. One is the pessimistic version: most of us are like the people in the early Industrial Revolution who could not learn to adapt or were stuck as mere cogs in factories. Very few of us, if any, will learn to orchestrate machines at a higher level of abstraction, and neither will we learn to invent new machines, since the artificial intelligence systems will soon exceed humans in their capacity for invention and discovery. That view is one of historical discontinuity: replacing knowledge work strikes deeper at the human uniqueness that has kept us employed than replacing various kinds of cognitive and manual labor has in the past.
The other story is optimistic: just like those early conductors and inventors of machines, we will continue our long human legacy of finding yet more to occupy our time, yet more activity that other humans find valuable. There is much more of this than we can possibly realize, because our collective imagination is bounded, yet our collective wants are limitless. How barren, in retrospect, do we find the mind of the man who thought the human touch was gone simply because we had invented machines stronger, more durable, and more reliable than us at physical labor?
Both stories will probably be true at the same time, but the unfortunate reality is that nobody knows in what proportion. More unfortunately still, it will be some time until we know: the temporary disruption that would portend broad displacement would look quite similar to the creative destruction that would come with just another industrial revolution. It’s easy for policymakers who first start to grapple with the notion of advanced artificial intelligence to reflexively adopt the pessimistic view: for so long, they’ve heard the idea that AI will be important and the idea that many jobs will be lost in the same breath that coming around on the scope of AI seems to imply believing that human labor is doomed. But that would be premature, and converts must resist becoming zealots.
Here, then, is the first—and in some sense the most troubling—message for policymakers: nobody can know what is going to happen. Anyone speaking with confidence about predictions of this kind is either misunderstanding or misleading. It is not just that we do not know “the future,” in some broad sense. We also do not know the specific nature of any problems posed by AI to the labor market: we do not know what industries, age groups, levels of seniority, job types, and so on will be affected by AI automation in practice rather than in theory or in speculation. We do not know over what timeframe these still-hypothetical changes will occur.
And if AI really does profoundly upend the labor market, we still do not know what the resulting distribution of economic resources will look like. Will the AI labs profit immensely, absorbing huge swathes of economic value as many other institutions struggle to survive? Or will AI models and systems become commodified, with value accruing to the compute designers and manufacturers? Or is it some hybrid, with most firms in the economy seeing higher profits with fewer employees and, for whatever reason, not seeing a need to hire additional people to do anything? Will there be new, high-skilled jobs created that we need to retrain millions of people for? Or will there be no new jobs at all? We do not know, and we cannot know.
That is because we are still in the process of writing this future. The role of humans in future economies is not something we simply discover as it occurs. How we distribute tasks between humans and machines is largely downstream of a web of complicated economic incentives and technical features. Is the marginal unit of computing power better spent on smoothing over the jagged frontier so no role remains for humans, or for even further improving the spikes of AI capability? Does the tax system favor firms who spend the marginal payroll dollar on hiring a worker to oversee the machines or an agent to do the same? Is there a safety net to catch those hit by local disruptions to give them the room to reorient themselves, come back five years later, and fight for their place in a new economy—or do we mollify their drive with ill-placed subsidies long enough for them to grow docile and for the structures around them to calcify? All this is contingent, and when policymakers ask ‘what will happen’, they fail to see that they’re among the central live players in this question.
How should our leaders grapple with this double uncertainty of what they should want and what will happen? First, while any honest policymaker must take seriously the notion that AI really is different, and really will herald “the end of work,” a student of history and admirer of humanity must have some skepticism: we have found some awfully creative uses of our time. The fact that we can only dimly imagine what is next for the human touch is not some aberration; it is in fact the typical predicament. We have never been able to imagine our futures well. To suppose that the end of work is imminent, or that the only way to preserve work is to freeze the status quo in amber by either guaranteeing today’s jobs forever, or banning AI progress, is fundamentally to make a bet against human resilience, ingenuity, and creativity. It is, indeed, a bet against humanity writ large. Maybe, deep down, you are convinced that AI will be so powerful that we should bet against humanity this time. Even still, you might consider how such a pessimistic and dour message—which amounts to a preemptive declaration of defeat on the part of the human species—will resonate with the public of any democratic nation.
We would especially caution against state interventions that freeze the current labor market in place. Such interventions would include automation bans in existing occupations and industries, government-mandated procedural review requirements before layoffs, jobs guarantees and similar protections. Though these are tempting remedies for policymakers—after all, they address the object level concern of protecting a person’s job today—they are among the most damaging. Europe has long had similar labor protections; it costs many times more to fire a worker in Europe than it does in the United States. A long line of work in the creative destruction tradition, most notably by Philippe Aghion, has shown that stringent employment protections dampen innovation, firm entry, and reallocation across sectors—arguably more so than the technology regulations for which Europe is better known. The mechanism might not merely be that firms hire fewer workers at the margin when firing is costly; it is that rigid labor rules lock in the status-quo structure of corporations themselves, making the kind of restructuring that accompanies major technological shifts prohibitively expensive.
The result is that European firms are well-suited to incremental refinement of established industries but ill-equipped for the moments of radical reorganization that new general-purpose technologies demand. If you believe that AI will further accelerate the pace of economic competition, it follows that frictions like these will be a progressively greater burden on economies.
The risks of setting up a new division to try a bold new innovation are high in any case; but they are much higher if there is an elevated cost of firing employees if the experiment fails. The AI era seems sure to require both risk-taking and significant corporate restructurings across the economy. Thus, protections of this kind—tempting though they may be—are among the worst suited to eras that require substantial corporate dynamism.
But deciding to reject bleak visions of stagnation or widespread displacement and instead bet on humanity needn’t mean blind optimism and faith in the future to go well. A person who is optimistic about a technological revolution in the long term does not have to believe that there will be no turbulence in the short term. A person who wants to make a bet on the human future must also remember that a bet requires collateral to be at stake. It is not time to play defense or to shrink away from challenges. It is time to put our chips forward.
Political Realities
It would be premature, and even irresponsible, for would-be political leaders to make promises to the public about the ability of the state to solve a problem that nobody understands, especially when that problem is so intimate as the future of human labor. The declining faith of the public in the federal government may well prove terminal if a future politician rises to power on some grand promise of a ‘new social contract,’ only to flail about when given the opportunity to bring that contract into being.
Yet the public, when faced with uncertainty, often desires confidence from their political leaders. What, then, is the honest policymaker to do? How does one govern over such a sensitive issue with such a high degree of uncertainty?
By starting with what we do know, and working from there. Polling indicates that the public are concerned about job automation, both affecting them and the country more broadly. It also indicates that people are uninterested in a universal basic income or similar redistribution of resources: they want to work for their living. These don’t seem to be transient attitudes on policy questions that politicians can fix by gaming the focus group or fine-tuning the proposal; they’re observable over years and decades, and we should assume that they will remain the backdrop of the conversation on AI and labor in the years to come. That fundamentally constrains our action space: there are many ‘no-regrets’ solutions that will fundamentally not satisfy the political will of an electorate scared of AI disruptions. Increasing state capacity on the margins, offering slightly-improved versions of policies that workers perceived as failures when they were first deployed against globalisation-driven job losses, or gesturing at extant-but-distrusted welfare systems simply will not do. If that is all that thoughtful policymakers have to offer, voters will rather pick solutions from the populist fringes, no matter their merit, because they’re at least taking the problem seriously.
So a leader hoping to prevail against the populist attractors needs to project a more decisive attitude. But in doing so, we should not constrain future action space too much. In AI labor policy, it is very easy to narrow the corridor of possible outcomes: if we subsidise a type of labor, we might never discover a fundamentally new arrangement for humans to work; if we tax one part of machine labor, we might push revenue to accrue elsewhere on the supply chain. We firmly believe that the best way to find out what gainful role humans can play in the economy of the future is not best dreamt up at campaign headquarters for a ‘28 hopeful, but through millions of distributed small experiments.
The result of these experiments might be that we need to keep today’s junior white collar roles alive artificially to retain a talent pipeline; or it might be that we’ll have to do away with the concept of the junior employee altogether. It might be that we should extend the centaur era of human-machine collaboration in software engineering, or that we should surrender coding to AIs altogether. Either of those could turn out to be the optimal arrangement, so whatever audacious policies we pick to correspond to political momentum, they cannot be based on the promise of keeping things one specific way or changing them to another specific new arrangement.
The optimal policy platform, then, needs to give us some of both: no-regret interventions that simply provide us with the capacity to act in the future; and more audacious bets on human agency that enable experimentation but do not foreclose paths to future paradigms.
Easy Wins
Even Footing
First, we should simply cease government activity that puts human labor at an outright disadvantage compared to AI systems. Overhead costs and taxes on employing humans are high; capital expended on AI infrastructure and systems that run on it is favoured and incentivised in many ways. It would be deeply regrettable if it turned out humans were competitive with AI systems on equal footing, but arcane tax incentives from a pre-AI era tipped the scales to favour putting human workers out of business. In that context, moving some of the tax burden from payroll tax to capital investments in AI specifically seems prudent. That policy, calibrated for the correct balance, should be net neutral for employers that employ many workers and boost them through AI, thereby avoiding the failing incentives of approaches like token taxes. This change should be narrow—this is not the place to increase net tax revenue, and any attempt to use it as such could pull employers away from important adoption. Done this way, it means the government no longer has a finger on the scale when a business decides between a junior worker and an artificial agent.
Retraining
Second, we support efforts today to bolster workforce training and development. While we are skeptical of most “retraining” programs, other workforce development programs have proven more promising, especially when implemented at the state and local level in collaboration with private-sector employers who have an immediate need for the skills being taught. More empirical work into this domain—especially empirical work that creates natural experiments in institutional and policy design—would be helpful. We also commend federal agencies, as well as exemplary state and local governments across the country, for their efforts to both improve workforce development programs and bolster upskilling programs for high-demand areas of the labor market, as highlighted by the Trump Administration’s AI Action Plan.
Measurement
Third and most importantly, we should actually want to know what is happening in as close to real time as we possibly can. The institutions America uses to measure its economy—federal agencies like the Bureau of Labor Statistics (BLS) and the Bureau of Economic Analysis (BEA)—came together during the Progressive and New Deal eras. Though these institutions have evolved since then, these agencies were designed with the measurement technologies and methodologies of their era: large-scale surveys, field interviews, and the like. What would a version of those institutions designed with the full suite of technologies devised since then—the internet and web-scale data, cloud computing, mobile telephony, advanced terrestrial and space-based sensing, and, yes, AI, among numerous others—look like? What could it do that would not be possible with the technologies of the 20th century? What would the BLS look like if, say, Google designed it? We support ambitious efforts—including from the frontier labs and related philanthropic ventures—to experiment along these lines. Governments, too, can and should run these experiments, and where they cannot, they should pay even more attention to encouraging other institutions to step in and fill the urgent gap in expertise.
Regardless of institutional evolution, private-sector involvement seems essential to measuring the effect of AI on the labor market. Today, data and qualitative information on labor market effects is scattered and still locked away in silos, many of them privately held. Frontier labs hold unique data about AI usage, as both OpenAI and Anthropic have demonstrated in recent studies. Deeper collaboration between government agencies and frontier labs in particular would therefore be very useful; at present, the federal government’s understaffed and underfunded statistical agencies are on the bad end of an information asymmetry. Policymaking cannot be performed well by stabbing blindly in the dark; information is the boring, but essential, prerequisite to sound AI labor policy.
From inside an AI developer with a view on capability progress and usage patterns, you might famously see one particular dramatic trajectory of near-term displacement; from the economist’s perch with a view on historical trends, you might be very skeptical of that. But if it’s the government that has to guide and support this transition, it should not be at the most unfavourable end of the information asymmetry, equipped only with an unimaginative and understaffed Bureau of Labor Statistics. Much more so than with cyber or bio risks, AI developers alone are untrustworthy narrators of the labor impact story since so much of their revenue prospects hinge on white-collar uptake specifically, and yet they alone have access to the privileged usage pattern and token spend data that could tell us so much about what is happening to the labor market. If we believe that the magnitude of cyber risks requires a well-equipped national cybersecurity apparatus with privileged information-sharing with the labs—and we do!—we should expect at least the same level of privileged information-sharing of economic usage data with well-equipped government institutes staffed with top-tier economic experts.
Difficult Bets
The Junior Job Subsidy
The above three items are the policy remedies we believe merit public, private, and philanthropic funding, legislative effort, and similar today. But we readily acknowledge that these policies alone may not be sufficient for the task that will soon be at hand. As we discuss above, it is difficult to design policy responses to a problem that has not (yet) manifested itself. Nonetheless, there are some emerging contours of a problem beginning to present themselves: the displacement of junior employment in knowledge-work and some service professions. Put simply, this is the notion that firms in industries ranging from customer service to software engineering to accounting to the law will have less need for younger, less experienced workers because AI can directly replace most of their labor—a notion that seems increasingly supported by an admittedly still inconclusive set of incoming data.
On the one hand, an organic reallocation of young workers away from knowledge work and toward labor-starved fields (especially physical-world jobs in manufacturing, skilled trades, and the like) would be entirely consistent with a positive “reindustrialization” story. In some important sense, if one desires reindustrialization of the U.S. economy, one should not just expect, but want, to see this trend.
On the other hand, many of the junior workers who are now struggling to find entry-level jobs did not train for these new kinds of jobs, may not be geographically proximate to them, and, particularly for higher-end career paths, may have taken on substantial college debt on the assumption of a large income that the newly available jobs simply will not support. Moreover, the overproduction of elites, which some have argued is already a problem, is often correlated with political instability. Put in crass terms: young people were raised to believe they could join the knowledge-work elite, only to discover that they are now functionally barred from that opportunity. They may feel as their society has failed them, and may therefore be prone to rebellion of various forms.
The problem of junior employment thus satisfies three separate tests: (1) it makes logical sense, from first principles, that large language models and related AI technologies would cause particular problems for young workers; (2) there is emerging data that precisely this may in fact be happening; and (3) there is reason to believe that such a phenomenon, were it to grow significantly, could pose both economic and political problems. We therefore believe that preparing policy responses to this problem, if not yet implementing them, is prudent at this time. In this instance, “preparing” might mean ideating policy designs, building theoretical and empirical evidence bases for different policies, and, most broadly, fostering public debate and scrutiny over such ideas. If we do not do so now, we might have to do so under worse conditions once the wave breaks—or not get around to it at all, instead watching powerlessly while momentary disruptions turn into long-term path dependencies.
We put to you that the solution to deal with junior job losses might be to keep these jobs around by brute force for a while, so that the critically important economic incentive to explore how to use junior workers does not cease. More specifically, we might do so by restructuring the tax code to subsidize junior employment. In its simplest and most naïve form, this subsidy could operate by making a substantial portion of a young person’s wage tax-deductible for employers for a certain period of time, declining each year (say, 50% of the wage of a person entering a new industry or starting a first job for five years, declining by 10 percentage points for each year the person received the subsidy). The subsidy could be gated to particular industries, occupations, or job levels based on objective criteria set by federal statistical agencies. That places it hand in hand with and downstream of the measurement improvements we suggested above: only a government that had a very good understanding of ongoing labor market trends could ever hope to introduce this kind of subsidy without letting it be gamed by employers in search of free subsidies. For similar reasons, this measure would have to have a global sunset provision: after years, perhaps a decade, from inception, the subsidy could expire altogether.
Such a subsidy would encourage employers to hire young people, almost functioning in practice like an apprenticeship program. Assuming that human-overseen (even if heavily automated) firms will want to maintain that human oversight in the long run, older managers will presumably want to maintain a steady flow of young workers into the firm, even if the number of new workers hired is lower than today. The elder partner at a law firm—whose value rests upon relationships, tacit knowledge, and other hard-to-automate skills—if he cares about the long-run existence of his firm, will have an incentive to cultivate the next generation of partners. And the broader economy and its ability to conduct those many small experiments, as we’ve argued above, would greatly benefit from a class of junior workers that has grown up amid AI disruption instead of watching it from the sidelines, unemployed.
Yet the value of young workers to a firm may be limited, particularly absent organizational and institutional evolution downstream of AI. During the transitional period, when advanced AI enters firms whose structures were largely conceived in the 20th century, there simply may not be much use for those young employees in practice, no matter the theoretical benefit of cultivating a new generation of managers. In raw economic terms, these young workers may simply cost more than they earn. Simple competitive logic would suggest that this means their jobs may be eliminated. Predicting this does not even require a particular view on the actual labor-market relevant capabilities of AI systems: firms could cease hiring in anticipation of capability trends that never come, of a compute supply that isn’t there, or simply in risk aversion that never pays out. From the individual firm level, such choices are prudent and would be recoverable through re-hiring later on. But for workers, it could be catastrophic: if the market takes a few years to re-rationalise, a generation of young workers might have already lost career momentum and motivation by then.
But if one believes firms may yet again find a use for young workers after the institutional transformations wrought by advanced AI—if one takes the bet on a human future—it may make sense to attenuate the raw market logic that would obtain during the transitional period. Otherwise, we risk stymieing not only the careers of junior workers who would bear the initial disruption, but the broader economy, which would lose its incentive to learn how to employ them productively.
And last, if we’re mistaken, and we are in fact headed for the end of work as we know it, the junior job subsidy can serve as the gateway to broader and more expansive measures of redistribution centered around purposeful human work. If the bet on human agency fails, we might have to pivot hard, into policy terrain that would stymie creative destruction if we went there too early. In that spirit, the junior job subsidy might be the first step toward familiarising the electorate as well as the state with more expansive measures like job guarantees. Either way, whatever lies on the other side of the river, this instrument is the first stone we can feel in our attempt to cross it. That is the fundamental case for a junior-employment subsidy.
Our case makes a dangerous trade-off: it decisively biases the labor market away from how it would naturally deal with AI in an attempt to combat an even greater bias—the bias that would be caused by disruptive shocks disincentivising junior workers and firms from exploring new arrangements together. We readily acknowledge flaws in policy design, political economy, and even fundamental conception. Perhaps our assumption that firms will want to—or be able to—retain human oversight is incorrect. Perhaps we will not be able to ramp up measurement ability in time to prevent large-scale grift and exploitation of the subsidy. In this case, a junior-employment subsidy would probably be neither desirable nor effective.
But in this specific domain, effects could be so imminent as to require intervention based only on circumstantial evidence, or we risk locking in path dependency toward a future where we have not figured out in time how to bring together the demands of our economy and the skills and qualifications of the junior worker. This is what seems to us the least bad option.
Taxes
Last, preparing for AI-focused labor policy also means preparing for the fiscal burden it will imply. No current budget could handle the costs of even narrow job subsidies, and definitely not of broader social policy—especially not once tax revenue wilts away as economic activity moves from personal income to corporate income and capital. Any serious platform needs an answer to that trend. But it’s very easy to let contentious premises bleed into that answer: whoever you think is the culprit, you might be inclined to focus the burden on them. Is it the billionaires, and so you want a wealth tax? The incessant adopters, and so you want a token tax? The AI labs, and so you want to take a stake in them? If you’re prepared to admit that you do not yet know, you’d rather choose an agnostic instrument.
We believe the most appropriate agnostic instrument is to raise and homogenise the corporate income tax. Wherever the value created by AI accrues, it seems like a safe bet that it will be with corporations—and so if we want to get the fiscal structures in place to be able to pivot fast, we should set up a way to do so through corporate taxes. Insulating this established tax structure against attempts to circumvent or undermine it—from accounting trickery to outright capital flight—is not trivial, but at least well-explored in decades of research and policy experimentation on the issue. Ensuring its legality, constitutionality, and conformity with existing institutions is similarly much easier than it would be for any attempt to invent a new and narrow tool.
It’s tempting, but we think mistaken, to pursue more specific modes of taxation. A ‘token tax’, for instance, seems appealing at face value: it places the burdens with those that seemingly adopt AI the fastest. But adopting AI the fastest is not the same as either capturing its value or as being at fault for displacement: you could easily imagine large legacy firms being among the highest token spenders in their attempt to equip their workers with sufficient AI assistance to weather the storm and fend off lean competitors. Do we really want to risk disincentivising them from rapid adoption and further increase their exposure to offshored AI firms or international rivals? No: if we believe that rapid AI adoption might be an effective tool in fending off outright AI displacement, our taxes cannot asymmetrically hit that adoption.
Taking a stake in frontier developers, whether through expropriation or fair purchase, again commits the mistake of overindexing on a specific trajectory—one where much of the revenue accrues with frontier labs. But that trajectory is far from certain: many other market participants, from infrastructure providers to downstream users enabled to reach much greater productivity with the use of AI systems, are capturing much of the value of advanced AI. It is understandable to want to tax the AI developers in particular because they seem in some sense more directly responsible for what is happening. But that is confusing the purpose of a tax in this context—which is to dispassionately generate revenue for the interventions we will need—with the purpose of other policy interventions that govern the influence and power of these frontier labs. We’ve both endorsed many such other policy interventions, but we should not use taxation as their vehicle.
But we also can’t be naive about the deeply political character of tax policy changes. There might be comparable, perhaps even more sophisticated methods to capture AI revenue—thoughtful researchers have suggested broad consumption taxes, for instance. But the changes we make to taxation are not only in service of satisfying a demand for greater fiscal resources, but also in service of signaling to the electorate that everyone is paying their fair share. If we are headed for a time where many workers feel disoriented and at times displaced, it will be politically prudent to clearly communicate to them that the beneficiaries are paying a ‘fair share’. Raising corporate taxes fits that brief well; increasing consumption taxes that would directly and visibly show up as a distinct item on every bill they pay would have the opposite effect, even if the technical revenue flows shake out the same. In that context, we favour the imperfect solution for its ability to guide instead of resist the political current.
A Path Ahead
There may well come a time when the labor market is, or is perceived to be, in a state of crisis because of AI. At that point, policy solutions will become politically unavoidable. Just as with catastrophic risks, it is wise for those who wish to preserve economic and technological dynamism to own these issues, rhetorically and in substance, so that policies amenable to future dynamism stand a chance at winning. By putting such ideas on the shelf, it is likelier they are the ones policymakers will reach for when and if the time comes. Our version of the junior subsidy and the tax code adjustment is only one such idea—our expertise is not in finding the most technically prudent version, but in identifying the requirements between technology and politics, in hopes that others can develop the solutions that fit them best. This is a challenge all techno-optimists will need to grapple with, sooner or later. Realizing time is the scarcest resource, we opt for sooner and encourage others to do the same.
The attitude we suggest you take on this issue is uncomfortable, because it lacks certainty in convictions. It asks you to bet on humans to figure out what to do, but not to idly sit back and watch it play out. Instead, we ask governments to take their thumb off the scale wherever they currently hinder human experimentation, and build the capacity to remain watchful enough to steer the trajectory back on track if we must. But there are risks that could irreversibly change the future economy for the worse, and on these you should act swiftly and decisively, in ways we’d usually never suggest a government should act.
If you strike that balance, you might end up with both a meritorious policy agenda and an actual political message: you can be an audacious protector, not of your own view of what should be done, but of your people’s ability to chart their course in the new economy on their own terms.
I find the tax policy recommendations to be very confusing. Why would optimal taxation be a high corporate tax, a tax whose incidence is highly industry dependent and also highly contested? Similarly, because capital and labor are complements, increasing taxation on capital will tend to harm labor as well unless there is a sufficiently strong substitution effect from capital to labor; how do you know that will be the case?
Going back to first principles, why would the recommendations not be to refocus the tax base further towards taxing land and consumption? Land is almost perfectly inelastic and land rents tend to appropriate most surrounding economic activity. Consumption taxes work because the income has to go somewhere, if they are strongly progressive then it doesn't matter if that income is concentrated, and because they don't impact investment they can bear much higher marginal rates than income taxes.