For the Balkinization Symposium on the Global Political Economy of Artificial Intelligence.
Ava Liu
There is a problem with the concept of “predistribution”, which is that the idea is ill-defined. At the broadest level, predistribution is defined as what is not tax-and-transfer redistribution. Predistribution is associated with what "can shape the distribution of income and opportunity in a society that are distinct from simply taxing and providing benefits" or that “instead of equalizing unfair market outcomes through tax-and-spend or tax-and-transfer (redistribution), we instead engineer markets to create fairer outcomes from the beginning”.
These definitions are helpful, but they are
residual. Predistribution is located as what remains outside of
tax-and-transfer redistribution. We know that predistribution advocates are
concerned about making markets more
fair and
equal. We know it is about structuring the market—"the means by which primary market
distributions of income and wealth are determined”—or that it is
what shapes primary inequality rather than after-tax income.
We know, too, that
predistribution is generally understood as operating before
redistribution; if redistribution happens with outcomes at a time (T2),
then there was a time (T1) before this point. But as legal
scholar Alex Raskolnikov recently summarized of the field, “the
line between redistribution and predistribution, it turns out, is not
well-defined and possibly nonexistent.” Philosopher Martin O’Neill has
argued that there is no
line between predistribution and redistribution in the substance of their policies.
This is a problem if we care about
predistribution as a domain of the economy we want to understand or a set of
policies related to inequality we aim to organize
around. The term “predistribution” has been around in the political
vocabulary for fifteen years yet the idea remains somewhat murky, and advocates
self-efface with how blurred it is as a concept. This is limiting if predistribution is to gain traction as a field rather than
remain a slogan for policies we intuitively care about but cannot precisely
identify.
Fleshing out a positive definition would go a long way. I offer one here.
I present a positive definition of predistribution
based on power over production. This I locate through a critique of tax-and-transfer
redistribution and what it cannot do, which points to what
predistribution does. I also start from the premise that there is a T2
of realized outcomes which is the point we are redistributing pre-tax income
from, and before that therefore some T1 in which those outcomes were structured
and produced.
In my own work on artificial
intelligence and technology governance, I first encountered this problem
through the labor automation debate, which was originally
dominated by
calls for universal basic income (UBI). UBI has
been heralded as the answer to technology-driven displacement: if workers lose
their jobs to automation, the state can tax and transfer the proceeds to provide
workers an income floor. In a welfare capitalist market economy, UBI is a
classic tax-and-transfer policy that aims to redress income inequality.
What
does UBI not allow us to do, though? It does not allow us to govern over
technology. It
distributes post-facto, after initial primary market distributions of income
and wealth are determined. This is not a contingent
failure of UBI as policy design. Rather, it’s a structural feature of tax-and-transfer
redistribution as a category and welfare transfer payments as a class. Tax-and-transfer operates
at T2 after outcomes have been produced in the market. An income tax
redistributes purchasing power in the realm of consumption. UBI cannot address
the problem of what to do about labor automation because it is not about what
to do about technology, a problem of production. In this sense, UBI’s position
in the labor automation discourse is a misattribution of a policy about income inequality
to a different problem of how to govern over technology. Its predominance in
the AI discourse is a symptom of the lack of collective imagination around
questions of economic production: what gets made, how, and under what
conditions. Nonetheless, examining that misattribution allows us to see more
clearly how redistribution is limited and what “predistribution” is.
We know that power to set the agenda over economic
outcomes is important. This could achieve fairer
outcomes than just post-facto redistribution as it concerns the realm of production.
In the labor automation debate, I’ve identified this as a technology innovation
and governance question: what we should do about AI is a set of decisions that
could be broadened beyond a small subset of management and shareholders who are
legally delegated to hold that power. Distributing agenda-setting power and
distributing voice democratize the process. So what are the levers that distribute
that power?
The argument defining the contents of
predistribution runs as follows. I then identify four categories of
predistribution.
Redistribution operates at T2; tax-and-transfer operates at
a point at which the market has results to redistribute from. Its defining
features are that it takes market outcomes as given and adjusts them. It is
post-facto and downstream.
We
know that prior to those outcomes, there exists upstream governance. This
carries agenda-setting power and the capacity to shape the conditions under
which outcomes are produced. This is precisely what redistribution cannot do. These
outcomes were structured at T1, which was the realm of production.
Predistribution
and redistribution are related and defined against one another, with
redistribution happening at T2 and upstream governance occurring before this at
T1. Redistribution is structurally incapable of setting the agenda or
structuring outcomes. What redistribution cannot do, let us name as what
predistribution does. Therefore, predistribution is the distribution
of power over outcomes.
Predistributive institutions distribute power
over outcomes. Predistribution advocates concerned with equality aim at moving
the power up to exercise control over mechanisms that can shape outcomes before
the redistribution of market outcomes.
Much of the work on predistribution has reflected
this logic. Public investment and government expenditure, such as through K-12
education and infrastructure, are cited as examples
of predistribution. The state can directly produce and provision goods
and services. Power has been exercised and the outcome set. The direct
provision of goods and services, depending on execution and the extent to which
it removes market dependence, can have a decommodifying effect: decommodification
occurs when access to a good is no longer conditioned on market participation.
These outcomes are directly provided without one needing to enter market
exchange to receive them. The distance of the outcome as mediated through the
market has collapsed.
Another set of policies grouped under predistribution
involve redistributing bargaining power. One can change the rules of a market, or
what Steven Vogel calls marketcraft.
Antitrust rules are the paradigmatic example. Changing the legal rules that
underlie a market shapes the engine of the market. Now a market is structured
differently with different rules and incentives, therefore producing different outcomes.
Even a redistributive program can have a predistributive effect that
changes bargaining power and the incentive structure of the market: consider
UBI's possible effect on raising workers’ exit options through increasing their
reservation wages, thereby raising the bargaining
power workers bring to market exchange.
Another type of
predistribution is distributing ownership and entitlements to assets. Land reform is the paradigm case: in U.S. occupied Japan and revolutionary
Communist China, land was compulsorily transferred from landlords to tenant
farmers and peasant populations. In the domestic
U.S. context, a similar mechanism appears as eminent domain, the
state's authority to compulsorily acquire private property. Public ownership is
another instance of this power. An interesting case is the social dividend: a
return distributed to citizens as co-owners of productive assets, as in Alaska's Permanent Fund or a state sovereign
wealth fund. This is a “predistributive” form of UBI, distinct from a
market-liberal basic income funded through tax-and-transfer; the entitlement
derives from ownership, not from downstream redistribution of market outcomes. In
the United States or other Western industrial democracies, where we are steeped
in the tradition of welfare capitalism, our understanding of redistribution is
tax-and-transfer, rather than directly changing entitlements as in radical
forms of land redistribution. However, this general class is also quite simply
redistribution. This discussion connects to Rawls’ idea of property-owning
democracy (POD). I elaborate on this form of predistribution—or redistribution
of ownership—and its relationship to Rawlsian POD in my forthcoming work on
“predistributive UBI”. Determining who owns productive assets establishes who is
entitled to property and returns, and allocates residual control, at least
nominally, although this is distinct from the next class of redistributing
control and governance rights.
Another type of
predistribution—more fundamental to the element of control and more
undertheorized as a field—is the distribution of governance power over
outcomes. Change who governs and in whose interests, and the outcomes
are structured differently. This distributes the power to determine the
agenda before redistribution of market outcomes. These means include collective
bargaining mechanisms, labor unionization, and industrial
policy, which have all been identified with predistribution. These
mechanisms change who decides and thus what is decided. Governance power can be
exercised through directly distributing voice, such as through union
representation in the firm. Notably, ownership and governance are often
bundled: worker cooperatives combine both, as does land reform to smallholders,
but they are analytically separable forms.
Scholars have observed there are limits to how much one can reduce inequality
with the help of tax mechanisms. I am highlighting those limits as structural
rather than practical features. What is “predistribution” is actually
the primary determinant of economic life, particularly the realm of production,
whereas post-facto tax-and-transfer in the market is actually the residual
ambit of welfare capitalism (the historical intellectual moment we are steeped
in), and its understanding of redistribution. Tax-and-transfer
redistribution does not allow for governance over market outcomes, and
yet it is commonly identified as the imaginary limit of what we think of as policy
solutions, whether it be climate change (through carbon tax) or labor
automation (through early calls for UBI). Tax-and-transfer redistribution cannot
change who the decision-maker is. That is predistribution, which occurs at T1
and directly sets the agenda. In forthcoming work on technology governance and
artificial intelligence, I elaborate on these themes and present a framework
for predistribution and governance over the economy, focusing specifically on
the ownership and governance categories of predistribution.
Near the end of his life, Rawls wrote
that “welfare-state capitalism permits very large inequalities in the ownership
of real property (productive assets and natural resources) so that the control
of the economy and much of political life rests in a few hands.” Rawls, the
defining theorist of liberalism, writing about welfare-state capitalism which
relies on redistributive tax-and-transfer, points out there is no control in
it. This points to where control over the means of production lies: in
predistribution.
Ava Liu is Academic Fellow in Law, Technology and Labor, Harvard Law School. You can reach her by e-mail at avliu@law.harvard.edu.