Citation:
Alan Wolfe, Algorithmic Justice, 11 Cardozo L. Rev.
1409 (1990)
Content downloaded/printed from HeinOnline
Tue Nov 27 03:17:16 2018
-- Your use of this HeinOnline PDF indicates your
acceptance of HeinOnline's Terms and Conditions
of the license agreement available at
https://heinonline.org/HOL/License
-- The search text of this PDF is generated from
uncorrected OCR text.
-- To obtain permission to use this article beyond the scope
of your HeinOnline license, please use:
Copyright Information
Use QR Code reader to send PDF
to your smartphone or tablet device
ALGORITHMIC JUSTICE
Alan Wolfe*
Julius stopped in front of his friend. "Listen, Rupert. If there
were a perfectly just judge I would kiss his feet and accept his punishments
upon my knees. But these are merely words and feelings.
There is no such being and even the concept of one is empty and
senseless. I tell you, Rupert, it's an illusion, an illusion."
"I don't believe in a judge," said Rupert, "but I believe in
justice. And I suspect you do too, or you wouldn't be getting so
excited."
"No, no, if there is no judge there is no justice, and there is no
one, I tell you, no one."1
I.
If there are no metanarratives, is justice possible? With the exception
of religious belief-to which it is often compared2 -the quest
for justice invariably has involved grand stories that take people beyond
the concerns of the material world into considerations of the
transcendental. The just act, the just person, 'and the just'society have
been viewed as possessing an other-worldly nature, as if only heroic
action on the part of heroic actors could achieve, or even approximate,
them. Plato's stories may be, in Geertzian language 3
"thick,"
while Rawls' are "thin," but neither points toward standards directly
observable in the everyday course of social practice, lying, as they are,
either hidden in shadows or behind a veil of ignorance.
Those philosophical dispositions known as postmodernism, poststructuralism,
and deconstruction 4-in questioning whether there exist
any standards of meaning, evaluation, taste, truth, or morality
outside of the specific ways we make contingent rhetorical arguments
about such contested terrains-lead inexorably to the conclusion that
* Presented to the Seminar on Law and the Humanities, Cardozo Law School, Yeshiva
University, and the Graduate Faculty, New School for Social Research, March 1990. Alan
Wolfe is the Michael E. Gellert Professor of Sociology and Political Science and Dean of the
Graduate Faculty, New School for Social Research.
I 1.Murdoch, A Fairly Honorable Defeat 218 (1970).
2 Levinson, Constitutional Faith 9-53 (1988).
3 C. Geertz, Thick Description: Toward an Interpretive Theory of Culture, in Interpretation
of Cultures 3-30 (1973).
4 There are obvious differences between these terms, as the polemics between Derrida and
Foucault indicate. Nonetheless, for purposes of the argument presented here, all these approaches
are suspicious of transcendental standards of justice that can be embodied in texts
like constitutions and thereby can be linked in this presentation--contingently, of course.
1409
CARDOZO LAW REVIEW
no transcendental metanarratives of justice can exist. To be sure,
skeptics such as Stanley Fish would claim, this does not mean that no
standards ofjustice are possible, just as Barbara Herrnstein Smith argues
that the absence of any uncontested aesthetic standard does not
mean that objects of art cannot be assigned value.' Such arguments
instead try to show that the standards we develop for such matters as
justice and truth are the products of specific language games, conventions,
shared normative understandings or community practices, due
to change when new contingencies arise from whatever source, including
pure happenstance.
There seems little question that the air admitted to discussions of
law through the windows opened by postmodernism has been refreshing.
Perhaps interpreters of texts will never again be quite so certain
in insisting that certain conclusions-including ones dealing with the
lives and liberties of real people-follow directly and automatically
from materials written down generations ago. But the epistemological
skepticism introduced by the overlap between law and literary
criticism does not resolve fundamental issues involved in the quest for
justice so much as it alters their focus. If meaning does not exist in
texts but instead in the interpretations brought to those texts by readers,
what we require, instead of a theory of the text, is a theory of the
reader.
.People read the texts that other people write. (Although the
structure of DNA has been compared to a text, to date I have not
seen any efforts to deconstruct the texts of living species besides our
own; similarly, non-living species, such as computers, can generate
texts which humans, if they wish, can deconstruct, but machines do
not seem capable of putting into any interpretative context the instructions
given to them). Moreover, not all people read and write.
Infants do not, and neither do the illiterate nor the brain dead. The
capacity to read and write is a potential, something that can only be
undertaken by a self: a mature, socialized human individual who has
grown up in a society and possesses the tools of culture given to her
by that society.' No adequate theory of readers is possible without a
sociological theory of the self, without some notions, coming perhaps,
5 S. Fish, Doing What Comes Naturally: Change, Rhetoric, and the Practice of Theory, in
Literary and Legal Studies (1989); B. Herrnstein Smith, Contingencies of Value: Alternative
Perspectives for Critical Theory (1988).
6 For a study of the impact that the ability to write (and, by implication, the ability to
read) has on the structure of society, see J. Goody, The Logic of Writing and the Organization
of Society (1986). The capacity of our reading material to link us together in a society sharing
a moral framework is emphasized in W. Booth, The Company We Keep: An Ethics of Fiction
(1988).
1410 [Vol. 11:1409
ALGORITHMIC JUSTICE
from writers like Mead, Schutz, Garfinkel, or Goffman-which seek
to define the self, not as found in nature (for in nature there are no
selves) but only as the product of society and its dynamics.7
Sociological theory since the nineteenth century has been premised
upon one or another form of philosophical anthropology. Theorists
may differ in how they claim humans to be a special and unique
species, but it is common to all the great thinkers in the sociological
tradition that humans do have special capacities and that these capacities
are a product of the way they organize their artificial environments.
Since interpretation, at the very least, assumes that human
beings can be reflective agents who can assign meaning to texts-including
those texts by which their affairs are regulated-and adjust
the meanings they find in those texts to meet the contexts and contingencies
within which they find themselves, contemporary philosophical
skepticism ought to find itself in historic continuity with the
philosophical anthropology of sociological theory. And yet it does
not. "Essentialism" runs against the grain of the contingency and relativism
so characteristic of these philosophical tendencies, since, in
assigning fundamental qualities to the human species, it assumes that
at least some things are certain and exist in spite of the interpretations
we give them. Indeed the distinction between nature and*culture
which lies at the heart of sociological theory, according to theorists
like Herrnstein Smith, needs to be disarmed "of its ideological
power.... With respect to human preferences," she writes, "nothing
is uniform, universal, natural, fixed, or determined in advance, either
for the species generally, or for any specific individual, or for any portion
or fraction of the species, by whatever principle, sociological or
other, it is segmented and classified ".... 8
In short, postmodern philosophical perspectives are not only not
neutral toward the way sociologists have defined the self, but actively
hostile. Foucault's description of man as a historically contingent creation
about to be washed away from the sand by the next epistemological
wave hovers over nearly all such contemporary approaches to
knowledge.' In the heady Nietzschean atmosphere of contemporary
thought, talk of the self verges close to humanism-only humans, remember,
have selves-and that particular combination of naivete and
arrogance alleged to be characteristic of Enlightenment thought.
From a postmodernist perspective, one is led to address such appar-
7 For a conception of the self close to the one I am discussing here, see C. Taylor, Sources
of the Self (1989).
8 B. Herrnstein Smith, supra note' 5, at 78.
9 M. Foucault, The Order of Things: An Archaeology of the Human Sciences 387 (1970).
1990] 1411
CARDOZO LAW REVIEW
ently human matters as desire and need without positing the existence
of autonomous human agents, as, for example, again in the case of
Herrnstein Smith-who coins the phrase "desired/able" in order to
indicate
that the valued effect in question need not have been specifically
desired (sought, wanted, imagined, or intended) as such by any
subject. In other words, its value for certain subjects may have
emerged independent of any specific human intention or agency
and, indeed, may have been altogether a product of the chances of
history or, as we say, a matter of luck."°
Smith's concern is with the process of evaluation, with the way in
which we establish standards of aesthetic judgement. But the question
raised by her denial of human agency can be raised for theories of
justice as well, since conceptions ofjustice always involve questions of
evaluation. Indeed the stakes involved in developing a theory of
human agency are higher when we discuss justice than when we discuss
taste, for one can imagine people living in the absence of any
transcendental standard of the latter-although I am not sure I would
want to-but it is almost impossible to imagine them, at least in
human form, living in the absence of the former. Yet legal theorists
attracted to postmodernism are as reluctant as literary theorists to
acknowledge the existence of an autonomous self; Thomas Heller, for
example, discussing the indeterminacy of the law, points out that it
"does not arise because the standpoint of the human individual is in
some way privileged or central. Rather, indeterminacy is an element
of the grammar of complex systems or a feature of the observation/
system relationship.""
The minimal condition for a theory of justice is that we find a
justification or legitimation for constraint. Other participants in the
intellectual division of labor, especially economists, may argue for
freedom, although it is not too difficult to perceive that the market is a
prison as well as an opportunity.' 2 Law talk, by contrast, is always
explicitly about regulation, the intellectual problem at hand being one
of understanding-and in some cases trying to change-rules that
make possible life in groups. Note that even those most committed to
a skeptical epistemological stance in no way deny the constraints in-
10 B. Herrnstein Smith, supra note 5, at 193.
11 Heller, Accounting for Law, in Autopoietic Law: A New Approach to Law and Society
307 (G.Teubner ed. 1988).
12 For relevant demonstrations of this point see, e.g., Lindblom, The Market as Prison, 44
J. of Pol. 324 (1982); Preston, Freedom, Markets, and Voluntary Exchange, 78 Am. Pol. Sci.
Rev. 959 (1984); West, Authority, Autonomy, and Choice: The Role of Consent in the Moral
and Political Visions of Franz Kafka and Richard Posner, 99 Harv. L. Rev. 384 (1985).
1412 [Vol. 11:1409
ALGORITHMIC JUSTICE
volved in thinking about law; their point is simply that those constraints
have no ultimate justifications, only local, contingent, and
socially constructed ones. Given that there will be constraints, we can
judge a theory ofjudgment by the legitimacy of the standards it establishes
for restraining our actions. The postmodernist reply that there
are no non-contingent standards is useless; for even the most radical
versions of postmodern theory, as we have seen, still presuppose a
human ability to interpret the contexts in which people find themselves.
If we accept that minimal philosophical anthropology as our
non-contingent standard, the question we can ask is: how, in the absence
of both a theory of texts and a theory of people, can postmodern
theories of justice legitimize obedience to rules in such a way as to
make those who are subject to such rules better interpreters of the
rules that rule over them?
Epistemological skeptics imagine two ways by which human affairs
will be regulated if we deny the possibility of any standards of
justice outside the purely contingent and local. One was suggested by
Thrasymachus-the first postmodernist-and is repeated, in more elegant
form, by Foucault and those inspired by him. Everything being
power, the only antidote to oppression is a transformation in the relations
of power. Appeals to justice, from such a perspective, are naive
and self-defeating, a lingering symptom of wooly-headed humanism.
One might just as well ask an earthquake to stop rumbling as ask
holders of power to bind their actions in accord with some preexisting
standard of justice. Replace all justice discourse by power discourse
and then we can begin to talk about who makes the rules and how.
"Does might make right?" Stanley Fish asks. "In a sense the answer I
must give is yes, since in the absence of a perspective independent of
interpretation some interpretive perspective will always rule by having
won out over its competitors."' 3 Or, more epigrammatically, "the
gun is always at your head."' 4
It seems doubtful that an approach emphasizing the ubiquity of
power and force in human affairs could generate an account of justice
that takes cognizance of human interpretative capacities, although
someone may come up with an argument to that effect. The conventional
response, in this case, seems like the correct one: if all knowledge
reflects the power of contending forces, then the way to
constrain individuals is not to rely on persuasion but coercion. Fish,
who believes that persuasion--e.g., rhetoric-is coercion, consequently
holds that human agents have strikingly little freedom in
13 S. Fish, supra note 5, at 10.
14 Id. at 520.
1990] 1413
CARDOZO LAW REVIEW
these matters: "In the end we are always self-compelled, coerced by
forces-beliefs, convictions, reasons, desires-from which we cannot
move one inch away."15 The theory of the self associated with any
such answer to the quest for justice is a theory asserting that there can
be no self, or at least not a very autonomous one.
As an alternative to the justice-lies-in-the-interests-of-thestronger
kind of argument, there is another way to think about constraint
contained within postmodern approaches to legal regulation,
and it is the one on which I want to focus in this paper. There being
no truths or standards outside the operation of a system, this way of
thinking argues, then the rules that structure the system lie within the
system. Each system is governed by its own laws, and such laws have
as their goal the reproduction of whatever system in which they are
found. The inspiration for such ways of thinking about rules comes,
not from the grand tradition of Western metanarratives about justice,
but instead from cybernetics, information theory, economics, population
ecology, quantum physics, cellular autonoma, linguistics, sociobiology,
artificial intelligence, DNA, and chaos theory. I will call such
conceptions of justice algorithmic. They offer a different solution to
the nihilism that seems to lie within deconstruction. We do not, if we
follow such an approach, haveto conclude that because there are no
metanarratives there are no rules. Rather we can govern our affairs.
and at the same time avoid privileging any one version of the good by
imagining our rules to be self-referential to that activity, whatever it
is, in which we find ourselves engaged.
Although algorithmic notions of the good avoid the stark view of
coercion inherent in arguments that equate knowledge and power.
they are even less charitable toward the possibility of an interpretative
and autonomous self. Algorithms are rules designed to be followed
with as little interpretive variation as possible. They may help explain
how computers function and how species other than our own regulate
their affairs, although, as I will try to show, there is a strong case that
even in those cases non-algorithmic rule following is more important
than researchers, at first, realized. They can, however, only be applied
to human affairs if we accept the notion that humans are
precoded rule followers. Yet if humans are following instructions algorithmically,
then they will have no interpretive capacities, will be
unable to read texts, will not be able to supply meaning to documents
that can inherently have no meaning, and, as a result, will be subject
to a fate offollowing rules without any input into how those rules are
15 Id.
1414 [Vol. 11: 1409
ALGORITHMIC JUSTICE
formulated and applied. Surely that is a conclusion that
postmodernists would wish to avoid at all costs.
At one level, postmodernists certainly do wish to avoid such a
conclusion; Fish, for example, finds in Chomskian linguistics an almost
complete algorithmic system, a formalism of truly nightmarish
dimensions.16 One alternative to Chomsky, of course, would be to
develop a kind of sociolinguistics-such as that associated with
ethnomethodology in sociology-in which meaning would be understood
as what human speakers provide in the contexts within which
their conversations take place. 7 But a move in this direction is a
move toward the self, constituting a step back up the slippery slope of
essentialism that we have just, in turning to postmodernism, slid
down. Postmodern theories of justice, I will argue, faced with a
choice between making a commitment to a theory of the self demanded
of their interpretive face and relying on algorithmic conceptions
of rule following associated with their skeptical face, tend to
adopt the latter. Sometimes this is explicit, such as in the case of
Niklas Luhmann, Gunther Teubner, and others attracted by cybernetics
and information theory. 8 At other times the move toward algorithmic
justice is more reluctant, opting not for "hard" algorithms,
such as those associated with artificial intelligence and Chomskian
linguistics, but instead for "soft" algorithms associated with the automatic
and "natural" following of the rules of a practice. Still, hard or
soft, what characterizes algorithmic justice is a lack of appreciation
for the rule-making, rule-applying, rule-interpreting capacities of
human beings and an emphasis instead on the rule-following character.
The price postmodernism pays for its flirtation with algorithmic
conceptions of justice is a very high one: the denial of liberation, play,
and spontaneity that inspired radical epistemologies in the first place.
II.
To provide legitimacy for the enormously difficult task of coordinating
our actions toward common goals without relying on force, a
conception ofjustice must mean something to those who will be governed
by its imperatives. Yet meaning is precisely what texts cannot
possess according to much of the philosophical inclination under dis-
16 Id. at 315-20.
17 See, e.g., Schegloff & Sacks, Opening Up Closings, in Ethnomethodology: Selected
Readings 233-64 (R. Turner ed. 1979).
18 See the bulk of the essays in Autopoietic Law: A New Approach to Law and Society,
supra note 11. One of the exceptions, the arguments of which overlap to some degree with
those here, is Rottleuthner, Biological Metaphors in Legal Thought, in Autopoietic Law: A
New Approach to Law and Society, supra note 11, at 97.
14151990]
CARDOZO LAW REVIEW
cussion here. Texts nonetheless contain words. Do those words convey
anything if they do not convey meaning?
At least for some thinkers working within postmodern philosophical
assumptions, texts, if not capable of conveying meaning, are
capable of conveying information. It ought to be immediately clear
that information and meaning are not only not the same thing, but
that they can work at cross purposes. (American voters, for example,
have more information than ever before about the candidates for
whom they vote, yet seem to cast votes that are less meaningful than
ever before, understood in the sense of making sense of how they behave.)
Meaning is a macro phenomenon that involves making larger
sense out of smaller bits, while information, especially in the computer
age, reduces larger complexity into smaller, and presumably
more manageable, units. If we accept one distinction between symbols
and signs-that the former work top down and the latter bottom
up'9-then symbols can have meaning, while signs convey
information.
From the standpoint of a theory of communication, information
has remarkable properties, ones that have been seized upon by theorists
to develop information processing machines of great potential.
When it was discovered that certain phenomena found in nature, such
as the structure of DNA, were also understandable as an information
processing mechanism, the possibility of a unified theory of cognition
began to seem possible. Surely a number of human activities, such as
language, could be understood as the reduction of complexity through
information processing, and, since thinking was believed to take place
in its own language, 20
it was a short step to the conclusion that the
human brain was also an information processor. Once that insight
was accepted, then all human activities-including not only how we
speak, but how we write poetry, compose music, make our laws, conduct
our economic activities, and everything else--could be understood
to be governed by similar dynamics. The unified theory of
cognition promised by the information processing model, in other
words, offered to unify, not only what we understand to be the sciences,
but to link the sciences together with both the social sciences
and the humanities.
What is often called postmodernism is fascinated by the potential
of information processing. This is certainly true of the inventor of the
phrase, Jean Francois Lyotard, who has a tendency to take extreme,
and rather dubious, positions vis-A-vis the capabilities of information
19 See, for a typical account, H. Pagels, The Dreams of Reason 192-94 (1988).
20 J. Fodor, The Language of Thought (1975).
1416 [Vol. 11: 1409
ALGORITHMIC JUSTICE
processing, such as suggesting that artificial intelligence will be capable
of translating from one "natural" language to another, that computers
could "aid groups discussing metaprescriptives by supplying
them with the information they usually lack for making knowledgeable
decisions," and that data banks will serve as "nature" for postmodern
individuals.21 But the fascination with information processing
is not just a Lyotardian quirk. The writings of Deleuze and Guattari,
for example, are filled with images of machines that program
other machines in ever-recurring fashion, down to the notion that the
structure of desire takes the form of a binary system.22 It may well be
the case that the tendency to attribute to information all the capacities
that one has stripped from meaning characterizes many thinkers who
believe that knowledge is defined by relationships among signs, rather
than by reference to any "reality," including symbols containing
meaning, standing behind the signs themselves.
Information theory is usually thought of in connection with developments
in cognitive science, mathematics, linguistics, decision
theory, and rational choice theory-all of them. closer in spirit to the
epistemological certainty and rationalistic clarity that postmodernism
rejects. Yet the matter is clearly more complicated than that. The
two intellectual giants who created the framework for postmodernism
and deconstruction-Nietzsche and Saussaure-were both attracted
to cybernetic notions of self-regulating systems because the rules governing
such systems made it possible for the relationships between
things to keep them suspended in air without being either ground
down to a reality beneath them or tied to a reality above them.
The case of Nietzsche is particularly instructive in this regard
given the importance that he has assumed in the law and literature
debates.23 But it is not the Nietzsche whose perspectivism is so attractive
to critical legal scholars that is important here, but instead the
Nietzsche of Zarathustra. In speaking of the metamorphosis of the
lion into the child in one of his early speeches, Zarathustra introduces
the image of a "self-propelled wheel," preparing the way for his later
discussion of the eternal recurrence-images and concepts quite similar
to the ideas of Goedel, Escher, and Bach which have been found to be
21 J.Lyotard, The Post-Modem Condition 4, 51, 67 (1984).
22 See, e.g., G. Deleuze & F. Guattari, Anti-Oedipus: Capitalism and Schizophrenia 14
(1977).
23 Levinson, Law as Literature, in Interpreting Law and Literature: A Hermeneutic
Reader 155-73 (1988); for an argument that Levinson has misapplied Nietzsche's ideas about
how texts can be interpreted, see Weisberg, On the Use and Abuse of Nietzsche for Modern
Constitutional Theory, in id. at 181-92.
1990] 1417
CARDOZO LAW REVIEW
compatible with the age of information machines.24 Moreover, even if
we do not accept the notion of the eternal recurrence as a cosmology-which
Nehamas, in defending Nietzsche asks that we do notwe
can still accept it not as a "theory of the world but a view of the
self."' 2- Nietzsche's somewhat mysterious references to the notion
that if we could live our lives over again we would live them in exactly
the same form as we have can, therefore, be read as a kind of
gedankenexperiment designed to show that the world is still possible
without selves that can be defined by essential, non-contested,
features.
A fascination with eternal recurrence, with the notion that automatic
processes can generate exactly similar responses over and over
again, would seem to characterize all those thinkers who are skeptical
of the possibility of autonomous, choosing, selves. Considering the
importance attached to notions about the death of the author associated
with Barthes-let alone the Derridian suspicion of there being
anything outside the text-self-recurrence takes on a special fascination
in the literary culture inspired by postmodernism, recognizing in
Borges, for example, the postmodernist par excellence.20 One ought
not to be surprised, consequently, that postmodernist thought, which
is so inspired by Nietzsche, can also overlap so significantly with the
"self-propelled wheels" now known -as Turing machines or cornputers.
What they all have in common is a distrust of the active self
and, as a result, an attraction to algorithmic imagery. Writing about
artificial intelligence, for example, Sherry Turkle points out that:
If mind is a program, where is the self? [AI] puts into question not
only whether the self is free, but whether there is one at all.... In
its challenge to the humanistic subject, Al is subversive in a way
that takes it out of the company of rationalism and puts it into the
company of psychoanalysis and radical philosophical schools such
as deconstructionism .... Artificial intelligence is to be feared as
are Freud and Derrida, not as are Skinner and Carnap.2 7
From this perspective, the contrast between the rationalism of cognitive
science and the irrationalism of postmodernism and deconstruc-
24 D. Hofstader, Godel, Escher, and Bach: An Eternal Golden Braid (1979); F. Nietzsche,
Thus Spoke Zarathustra, in The Portable Nietzsche 139 (1954).
25 A. Nehamas, Nietzsche: Life as Literature 150 (1985).
26 Foucault's The Orderof Things, which ends with the image of man being washed away
in the sand, begins with a discussion of a passage from Borges. M. Foucault, supra note 9, at
xv. Poe and Roussel, two other novelists fascinated with the entrapping imagery of machines,
are also important literary sources for postmodernist speculations.
27 Turkle, Artificial Intelligence and Psychoanalysis: A New Alliance, 117 Daedalus 241,
245 (1988).
1418 [Vol. 11: 1409
ALGORITHMIC JUSTICE
tion takes a backseat to their common attitude toward the nonautonomy
of the self.
The extreme representative of the common ground shared by information
theory and literary postmodernism is Michel Serres, who
has incorporated all the reference points for information theory--entropy,
Maxwell's demon, the second law of thermodynamics, Claude
Shannon, and Boltzmannian quantum physics-into a theory of the
origins of language. Information theory allows Serres to develop a
theory of communication without there necessarily being any communicators.
In contrast, for example, to Habermas, who specifies two
parties to a communication (and who, in so doing, inspires heavyhanded
critique from Lyotard),2 Serres shows how language may be
possible without knowing anything about its origins:
I know who the final observer is, the receiver at the chain's end:
precisely he who utters language. But I do not know who the initial
dispatcher is at the other end. I am confronted indefinitely
with a black box, a box of boxes, and so forth. In this way, I may
proceed as far as I wish, all the ways to cells and molecules, as
long, of course, as I change the object under observation.29
As might be expected, Serres theory about the origins of language has
little to do with the notion of an autonomous self.
There is only one type of knowledge and it is always linked to an
observer, an observer submerged in a system or in its proximity.
And this observer is structured exactly like what he observes....
There is no more separation between the subject, on the one hand,
and the object, on the other ....
As Serres's remarks would seem to indicate, communication is
possible within the terms of information theory, but interpretation is
not. Information can only be transmitted, not read. The act of reading,
by bringing an interpreting self in confrontation with a text, can
only be viewed, from the perspective of information theory, as noise.
Although one may argue that Serres's approach provides "a unique
example of the possibilities opened up by bringing literary culture and
scientific thought into play with one another,"'a3 it is hard to see how.
Hence Paulson, who wants literary critics to take information theory
28 J. Lyotard, supra note 21, at 60-67. Lyotard also has strongly critical views toward
Luhmann, yet, at least with respect to their joint fascination with cybernetics and the reduction
of complexity, they share more than he is prepared to admit. See N. Luhmann, The
Differentiation of Society (1982).
29 Serres, The Origin of Language: Biology, Information Theory, and Thermodynamics, in
Hermes: Literature, Science, Philosophy 82 (1982).
30 Id. at 83.
31 W. Paulson, The Noise of Culture: Literary Texts in a World of Information 31 (1988).
1990] 1419
CARDOZO LAW REVIEW
seriously, winds up concluding that even though literature, as an artifact
of culture, may only be a form of noise, still "[w]hat literature
solicits of the reader is not simply reception but the active, independent,
autonomous construction of meaning."32 Without ever explicitly
suggesting so, Paulson's study suggests that although there are strong
similarities between deconstruction and information theory, the former
at least allows for readers, even if it does not theorize much about
them, while the latter does, and can, not. In pushing information theory
to its logical conclusion, these efforts make clear why a purely
algorithmic approach to communication is inappropriate to the texts
that human beings write and read: meaning exists when human selves
attribute characteristics to the symbols around them, while information
requires only relationships between signs irrespective of whether
there exist selves reading into those signs anything whatsoever.
Surely the attraction of information theory is its promise that it can
bypass the problem of meaning in a philosophical culture where
meaning has become so problematic. In doing so, however, it renders
the readers of texts into passive receivers of information as if they
were computer programs or DNA molecules. Recent developments
in both artificial intelligence and biology suggest, more than ironically,
that the notion of an algorithmic transmittal of information is
not only of little relevance to humans, but also not completely characteristic
of what takes place either in machines or in other living
species.
The recent history of artificial intelligence ("Al"), in fact, constitutes
a major attack on the notion of algorithmic rule following.
Although a number of starts were made in AI research that were nonalgorithmic
in nature-including Frank Rosenblatt's notions of
perceptrons and the expert systems approach adopted by Newell and
Simon 33-the early decades of work in AI were inspired by efforts to
represent the real world in machines through the device of giving machines
programs written as precisely as possible.34 If the software instructions
were precise enough, the argument ran, then whether the
von Neumann architecture of a central processing unit actually modelled
the way human brains worked was irrelevant. It turned out,
however, that the scripts and frames proposed by researchers such as
32 Id. at 139.
33 A. Newall & H. Simon, Human Problem Solving (1972); F. Rosenblatt, Principles of
Neurodynamics (1962).
34 Even the simplest instructions imagined in the Turing machine, however, can, from
Wittgensteinian premises, be understood as something more than the simple following of a
rule. See on this point Shanker, Wittgenstein Versus Turing on the Nature of Church's Thesis,
24 Notre Dame J. of Formal Logic 615 (1987).
1420 [Vol. 11:1409
ALGORITHMIC JUSTICE
Minsky and Schank to represent the real world were so brittle in nature
that the limits of a purely algorithmic approach to artificial intelligence
were quickly reached." As one critic pointed out, the
problem with such an approach was that for the machine to know
anything, it first had to know everything.16
Machines, in short, could
clearly be programmed to follow rules, but whether such rule following
constituted intelligence in anything like the way that quality is
generally understood was another matter. Intelligence, at least in
human form, is, according to two neurobiologists and one mathematician,
non-algorithmic in nature; human brains work, not by following
rules, but by recognizing realities in the larger world and thereby incorporating
experience and context into the thinking process.37
The failure of "software" approaches to Al were hailed in some
quarters of the artificial intelligence community, especially among
those who believed that the proper way to design machines was not by
creating software programmed with precise instructions, but literally
by designing machines to resemble the presumed architecture of
human brains. Connectionist, neural net, or parallel data processing
models-as they came to be called-did not begin with the assumption
that memory could be stored in a CPU, to be accessed through
instructions in the form of rules. Instead networks of electrical
charges were constructed in such a manner that machines could
"learn" by using the strengths between connections to narrow down a
problem until a solution was found that was correct, or at least, less
incorrect than a series of possible solutions that were rejected.3" Precision-what
I have been calling, following Nietzsche, eternal recurrence-was
sacrificed in such approaches for the flexibility introduced
by allowing machines to "settle in" to solutions.39 Algorithms, in
35 R. Schank & R. Abelson, Scripts, Plans, Goals, and Understanding (1977); Minsky, A
Framework for Representing Knowledge, in Mind Design 95 (J. Haugeland ed. 1981).
36 I.Rosenfield, The Invention of Memory 112 (1988).
37 The neurobiologists are I. Rosenfield, id. at 144-45, and G. Edelman, Neural Darwinism
44 (1987). The mathematician is R. Penrose, The Emp-ror's New Mind 405-18 (1989).
38 For various approaches, see P. Churchland, Neurophilosophy: Toward a Unified Science
of the Mind-Brain (1986); Foundations (Parallel Distributed Processing: Explorations in
the Microstructure of Cognition No. 1, 1986); S.Grossberg, Neural Networks and Natural
Intelligence (1988); C. Mead, Analog VLSI and Neural Systems (1989); Churchland &
Sejnowski, Perspectives on Cognitive Neuroscience, 242 Science 741 (1988).
39 Although PDP approaches are less attracted to the kinds of eternal braids discussed by
Hofstadter, they resonate with postmodern themes in another way. Like deconstruction, connectionist
approaches to Al believe that meaning does not lie behind any text but only
emerges, if it exists at all, as a relationship between signs. As D. A. Norman puts it, "I believe
the point is that PDP mechanisms can set up almost any arbitrary relationship. Hence, to the
expert, once a skill has been acquired, meaningfulness of the relationships is irrelevant." Norman,
Reflections on Cognition and Parallel Distributive Processing, in Psychological and Bio-
1990] 1421
CARDOZO LAW REVIEW
short, already found to be inappropriate to humans, were similarly
found to be inappropriate to machines.
Purely algorithmic understandings of information transmittal
have received a blow from another quarter: that of the process by
which DNA sends instructions through its replication and thereby
makes possible species development. The roughly parallel discoveries
of realizing Turing machines of great power and the uncovering of the
structure of DNA presented an irresistible challenge to sociobiologists
in particular: genes present information in the form of instructions
which determine the trajectory by which a species evolves. As Richard
Dawkins put the matter, "We are survival machines-robot vehicles
blindly programmed to preserve the selfish molecules known as
genes."' So convinced was Dawkins of the appropriateness of the
computer metaphor to the evolutionary process that he developed a
software program that would enable the user to trace the patterns of
many different evolutionary possibilities by specifying relevant features
at the beginning of the process in order to understand how even
very slight flaws in the transmission of messages (such as those contained
in DNA) create fantastic variation over enormous periods of
time.4'
One should, therefore, note that both the two leading representatives
of sociobiological thinking-Dawkins on the one hand and Edward
Wilson on the other-found that a purely algorithmic
understanding of genetic transmission ultimately could not explain
the speed of human evolutionary changes. Dawkins, for example, after
spending an entire book discussing selfish genes, concluded that
the day of the gene was passed; in the future, cultural transmissionrepresented
in what he called memes (from mimesis)-would take
over and, being superior, drive out genetic transmission entirely.42
Meanwhile Wilson, together with Charles Lumdsen, rejected the analogy
with computers completely, on the grounds that the memory capacity
of human brains would have to be larger than we can imagine
to contain all precoded instructions sufficient to account for human
evolution.4 3 Instead Lumsden and Wilson argued for what they
logical Models 531, 544 (Parallel Distributed Processing: Explorations in the Microstructure
of Cognition No. 2, 1986).
40 R. Dawkins, The Selfish Gene ix (1976).
41 R. Dawkins, The Blind Watchmaker (1986).
42 R. Dawkins, supra note 40, at 205-214. Dawkins does not address the issue of whether
memes reproduce themselves less algorithmically than genes. He does write, however, that
"[t]he computers in which memes live are human brains." Id. at 211. Since, as we have seen,
human brains are believed to function non-algorithmically, the shift from genetic to cultural
transmission would seem to indicate a less algorithmic process.
43 C. Lumsden & E. Wilson, Genes, Mind, and Culture 332 (1981).
1422 [Vol. 11: 1409
ALGORITHMIC JUSTICE
called "gene-culture coevolution,"' a process by which the biological
and the social share in determining the course of human. evolution.
They introduced a distinction, between primary and, secondary, epigenetic
rules as a way of recognizing that the latter allowed for the possibility
of autonomous. minds affecting the course of evolutionary
development,45 an important concession, but one that: still left open
the possibility of third-order epigenetic rules (and others' beyond that)
in which mind was understood to play even a greater role than they
were prepared to admit. Their concessions to their critics, in short,
were probably not enough to explain how evolution took over and
produced such enormous variation in the development of our species
in such a remarkably short (by evolutionary standards) amount of
time. Algorithmic understandings of evolutionary dynamics, in any
case, have been found as problematic in biology as-they have been in
computer science.
Perhaps the information processing model associated with these
sciences is flawed, even in areas where, unlike with humans, it had
been expected to work. If so, then there may be reason to question
some of the assumptions of information theory. Qyama, for example,
has argued that the notion of self-reproducing feedback loops so essential
to information theory is a, metaphor developed- because the
existence of computers provided the relevant imagery. But one would
be incorrect, in her view, to adopt a preformatist attitude toward' information,
that is, to conclude that information always exists before
the means developed for its transmission are imagined. "The developmental
system," she writes; ". . . does not have a final form, encoded
before its starting point and realized at maturity."' 6
We need to conceptualize
the transmission of information, rather, as developmental,
as a process that adds something. to the process in the course of its
evolution rather than spinning around within. the same already existing
information. Oyama's arguments, of course, are-not-a refutation
of information theory, and one may still argue, as-some biologists
do, that life itself is "autopoietic" in the sense that "living beings are
characterized in that, literally, they are continually self[re]producing."47
Still the possibility that self-recurrence. may not be
a characteristic found in nature does raise the question; of whether it is
a helpful way to think about society, since society is generally held.to
44 Id. at 19-34.
45 Id- at 53-58.
46 S. Oyama, The Ontogeny of Information 23 (1985).
47 H. Maturana & F. Varela, The Tree of Knowledge. 43 (1987). As might be expected
Maturana: and Varela are fascinated by ideas of self-recurrence, using a drawing from. Escher
to make their point. Id. at 25.
1990] 1423
CARDOZO LAW REVIEW
be populated by human agents whose actions can alter purely algorithmic
codings.
Two conclusions, then, can be reached about the search for the
perfect algorithm. One is that if there are perfect algorithms, they are
incompatible with the notion of freely choosing autonomous selves.
Systems, not their components, have autonomy in a purely algorithmic
world, just as, in some of the more mechanistic views of
sociologists like Durkheim and Parsons, social structures, not individuals,
determine consequences. 48 At the same time, however, we have
also seen that even if imaginable in theory, a perfectly algorithmic
system is in practice far more difficult to realize than at first understood.
Algorithmic machines are too brittle to resemble human intelligence.
Genetic transmission of information, especially in the case of
humans, takes place over time periods far too short for algorithmic
principles to be able to explain them. Information theory, far from
providing the basis for a unified theory of cognition, may be highly
limited in its applications to relatively contrived situations. The
search for the perfect algorithm is both futile and self-defeating.
III.
If algorithmic notions are as problematic as they seem, even in
areas such as artificial intelligence and genetic transmission, they
seem even less likely to be of use in such human and social activities
as reflecting on justice (or knowledge, morality, and taste). They do,
nonetheless, possess one feature which makes them attractive to
postmodernist thinkers: their denial of the possibility of autonomous
human agency overlaps with the suspicion of humanism and the negation
of the self so characteristic of Derrida, Lacan, Barthes, Foucault
and other influential thinkers. Hence although algorithmic thinking
is highly formalistic and anti-interpretative, many contemporary theorists
cannot avoid the temptation to introduce algorithmic conceptions
into their arguments.
Perhaps the most interesting example of the power of algorithmic
imagery is the effort by Barbara Herrnstein Smith to make a case
against any non-contingent standards of evaluative judgement. For
Herrnstein Smith, the Western humanistic tradition has sought stan-
48 "The division of labor does not present individuals to one another, but social functions."
E. Durkheim, The Division of Labor in Society 407 (1964). Interesting enough, theorists in
artificial intelligence, stumped by how essentially dumb bits of information can be linked together
into something smart called intelligence, utilize images strikingly similar to Durkheim's:
"Each mental agent by itself can only do some simple thing that needs no mind or
thought at all. Yet when we join these agents in societies-in certain very special ways-this
leads to true intelligence." M. Minsky, The Society of Mind 17 (1986).
1424 [Vol. 11:1409
ALGORITHMIC JUSTICE
dards of "transcendence, endurance, and universality ' 49 in its evaluation
of literary works, but her own personal -relationship to
Shakespeare's sonnets convinces her instead that "everything is always
in motion with respect to everything else."5 If value is therefore
never a fixed attribute of any particular product under
evaluation, how do certain cultural products come to be seen as worthy,
while others are assigned to the dust bin of culture? Hermstein
Smith relies on economics for an answer: each of us has a personal
economy of needs and resources. "Like any other economy, moreover,
this too is a continuously fluctuating or shifting system, for our
individual needs, interests, and resources are themselves functions of
our continuously changing states in relation to an environment that
may be relatively stable but is never absolutely fixed."'"
Markets, then, play a role in the creation of literary standards,
and not only in the narrow economic sense of money. But the important
question is what kind of market this is: are we talking of the
kinds of rigged and fixed markets which radical critics since Marx
believe drive capitalist societies or instead the purely automatic, homeostatic
markets envisioned by eighteenth century liberals? For
Herrnstein Smith, it is clearly the latter: markets are interesting to her
because they work independently of the desires of the agents in the
market.52 Like contemporary rational choice theorists, Herrnstein
Smith argues that self-interest drives everything we do: "We are always,
so to speak, calculating how things 'figure' for us-always pricing
them, so to speak, in relation to the total economy of our personal
universe."53 But unlike rational choice theorists, Herrnstein Smith
does not believe that "we" do this calculating consciously and as autonomous
choosers:
Most of these "calculations," however, are performed intuitively
and inarticulately, and many of them are so recurrent that the habitual
arithmetic becomes part of our personality and comprises
the very style of our being and behavior, forming what we may call
our principles or tastes-and what others may call our biases and
prejudices.
54
Since we find in Herrnstein Smith's account a picture of the mar-
49 B. Herrnstein Smith, supra note 5, at 28.
50 Id. at 15.
51 Id. at 31.
52 There are occasions in her text, however, where Smith takes the opposite tack and argues
that markets are rigged: "The linguistic market can no more be a 'free' one than any other
market, for verbal agents do not characteristically enter it from positions of equal advantage or
conduct their transactions on an equal footing." Id. at 111.
53 Id. at 42.
54 Id. at 43.
14251990]
CARDOZO LAW REVIEW
ket which is far more invasive than anything found in writers like
Gary Becker and Richard Posner who see markets everywhere, we do
not wonder that algorithmic conceptions of self-regulating systems
come to dominate her account of how standards of taste become established.
So much is in motion at such speeds that the only possible
regulation of the whole process is automatic regulation, or what
Herrnstein Smith calls an 'evaluative feedback loop':
Every literary work-and, more generally, artwork-is thus the
product of a complex evaluative feedback loop that embraces not
only the ever-shifting economy of the artist's own interests and resources
as they evolve during and in reaction to the process of
composition, but also all the shifting economies of her assumed
and imagined audiences, including those who do not yet exist but
whose emergent interests, variable conditions of encounter, and rival
sources of gratification she will attempt to predict-or will intuitively
surmise-and to which, among other things, her own
sense of the fittingness of each decision will be responsive."
Not surprisingly, therefore, Herrnstein Smith finds attractive any
way of thinking that emphasizes algorithmic processes. In the course
of her discussion, she touches on the possibility that human brains
may be cognitively hard-wired in predetermined ways;56 criticizes the
Habermasian notion of rational communicative standards on the
ground that we speak, as we spend, only out of self-interest, so thathonesty
in speech, if it ever exists, is the product of a Mandevillian
lack of intention;57 adopts information theory as the model for an
epistemology in which "what is traditionally referred to as 'perception',
'knowledge,' 'belief,'... would be an account of how the structures,
mechanisms, and behaviors through which subjects interact
with-and, accordingly, constitute-their environments are modified
by those very interactions;" uses information theory to explain that
evaluative classifications exist so that "energy need not . . . be expended
on the process of classification and evaluation each time a
similar array is produced;"59 argues that often such classifications are
"fixed in the DNA;"' and relies on Brownian motion and Nietzsche's
"play of forces" to criticize any who suggest an "overall, underlying,
or ultimate governing outcome toward which each instance of human
productive-acquisitive or consummatory-expenditure activity (all
55 Id. at 45.
56 Id. at 101.
57 Id. at 108-09.
58 Id. at 95.
59 Id. at 122.
60 Id.
1426 [Vol. 11:1409
ALGORITHMIC JUSTICE
making, getting, and spending, we might say) is directed .... "61
Most illustrative of all, however, is Herrnstein Smith's account of
why certain products of culture have entered our canon. In answering
this question, she is not only more economistic than the most
committed rational choice economist, she is also more taken with genetic
theories of evolution than most sociobiologists. Artistic texts
survive the way species do: "These interactions are, in certain respects,
analogous to those by virtue of which biological species evolve
and survive and also analogous to those through which artistic
choices evolve and are found 'fit' or fitting by the individual artist. 62
Evolutionary feedback loops allow Herrnstein Smith to resolve the
question in aesthetic theory proposed by Hume: why do we consider
Homer great? The answer is not that Homer survived because he was
great but that because he survived he is considered great. "Nothing
endures like endurance. ' 63 Images of eternal recurrence combine
with Durkheimian functionalism to explain the secret of Homer's
success:
Repeatedly cited and recited, translated, taught and imitated, and
thoroughly enmeshed in the network of intertexuality that continuously
constitutes the high culture of the orthodoxly educated population
of the West (and the Western-educated population of the
rest of the world), that highly variable entity we refer to as "Homer"
recurrently enters our experience in relation to a large
number and variety of our interests and thus can perform a large
number of various functions for us. . ...
Although Barbara Herrnstein Smith is not writing about justice
but about evaluation, her analysis demonstrates the linkage between
the position that there are no non-contingent standards in the world
and the need, consequently, for automatically functioning regulatory
mechanisms. When we turn to writers who are directly concerned
with justice, we find exactly the same linkage. The clearest example is
Luhmann, who finds in cybernetics and information theory an answer
to the question of what makes society possible. Luhmann wants to
understand how societies-which are not only enormously complex,
but are also, in their modern form, more complex than ever beforereproduce
themselves through time. Arguing, in a tradition that goes
back to Mandeville and Adam Smith, that no conscious direction can
ever guide a system so complex, Luhmann turns to methods by which
systems reduce their complexity. Computers, of course, reduce com-
61 Id. at 144.
62 Id. at 47.
63 Id. at 50.
64 Id. at 53.
1990] 1427
CARDOZO LAW REVIEW
plexity by dividing all information into bits that can be expressed as
zeros and ones. So, argues Luhmann, do legal systems. A legal system
can exhaust the entire realm of the possible through the legal/
illegal dichotomy. That distinction, in a sense, constitutes the "hardware"
of a legal system. In order for the system to take a decision in a
specific case, "software" programs access the system, feeding back
into the "memory" and thereby creating new rules that anticipate future
programs.65
For Luhmann, the dynamics of specific cases introduced into the
system continuously redefine the binary codes, interacting again with
new programs in ways that resemble eternal recurrence. The whole
system, he argues,
is a matter of a specific technique for dealing with highly structured
complexity. In practice this technique requires an endless,
circular re-editing of the law: the assumption is that something will
happen, but how it will happen and what its consequences will be
has to be awaited. When these consequences begin to reveal themselves
they can be perceived as problems and provide an occasion
for new regulations in law itself as well as in politics. Unforeseeable
consequences will also occur and it will be impossible to determine
if and to what extent they apply to that regulation. Again,
this means an occasion for new regulation, waiting, new consequences,
new problems, new regulation and so on.66
"Autopoietic law" thereby, according to Luhmann, avoids many of
the problems faced by other philosophies of law. It explains how a
legal system can change, for example, as well as provides a way of
thinking about the law that guarantees its autonomy from other
systems.
Luhmann's theories about the law overlap with postmodernism
because autopoietic systems are non-hierarchical. Being circular in
their dynamics, they avoid privileging any one set of legal norms over
any other; as Luhmann expresses it, "There can therefore be no norm
hierarchies," or, somewhat more self-reflectively, "legal forms are
valid because they are valid."' 67 But if legal scholars turn to autopoietic
theories of justice out of a generalized commitment to principles
of equality, they may find the equality not worth having. Since
notions of self-regulation come primarily from biology on the one
hand and artificial intelligence on the other, any legal system designed
by such principles cannot incorporate specifically human capabilities,
65 N. Luhmann, Ecological Communication 64-66 (1989).
66 Id. at 66.
67 Luhmann, The Unity of the Legal System, in Autopoietic Law: A New Approach to
Law and Society, supra note 11, at 21, 23.
1428 [Vol. 11:1409
ALGORITHMIC JUSTICE
such as the possibility that autonomous human subjects can interpret
the instructions given to them out of their history and contexts. For if
the agents ruled by laws can interpret laws, then automatic self-regulation
no longer exists. Any conception of justice that might emerge
from such a system would have to be just in the way ant colonies are
just or the evolution of different species or worms is just or computer
programs are just. Justice would thus be defined as having reached
some kind of stable equilibrium that makes possible the continued reproduction
of the system. What justice could not be, under such a
conception of law, is a quality that enhances a specific human capacity
to bring meaning to situations and contexts in order to guide them
toward any purpose defined by a community of autonomous actors.
Stanley Fish's reflections on the law illustrate the dynamics of
algorithmic justice in a slightly different way than Luhmann's. Fish is
well known for his insistence that standards do not transcend the particular
points of view of the communities that interpret them.
Whether he is correct or not is not the point on which I want to focus.
I want to argue instead that by introducing the term "community, ' 68
Fish is under a certain obligation to talk sociology: to discuss what a
community is, how its members act, what relationship exists between
individual needs and community concerns, and other typical concerns
of sociological theory. After all, an enormous emphasis is being
placed by Fish's approach on the practices carried out by human
agents, including not only judges, but professors of law and literature,
readers of texts, and, presumably, all those affected by the legal decisions
which in turn are affected by how judges and legal intellectuals
make their arguments.
Surprisingly, however, questions involving sociological practice
play relatively little role in Fish's writings. Consider his answer to the
question of why we ought to be concerned about the interpretation of
legal texts in the first place. For Fish the overlap between law and
literary criticism is the result of a glitch in democratic theory: the
existence of judicial review, which in enabling judges to overrule democratic
decisions in the name of fidelity to an earlier text, builds
counter-majoritarian tendencies into our political system.6 9 Yet why
do we have constitutional texts, and procedures for reviewing them, at
all? Surely both the Constitution and the practice of judicial review
illustrate a larger sociological problem: one identified quite clearly by
the rational choice philosopher Jon Elster in his discussion of Ulysses
and the Sirens. Constitutions, as Elster argues, deal with the problem
68 S. Fish, Is There a Text in This Class? (1980).
69 S. Fish, supra note 5, at 338-39.
1990] 1429
CARDOZO LAW REVIEW
of binding, the ways in which one generation attempts, like Ulysses
tying himself to the mast, to make it possible for the next generation
not to be seduced by the temptations of immediate gratification and
self-interest.'0 Judicial review, by contrast, grows out of the recognition
that the bonds, if tied too tightly, result in bondage. Far from
being a quirk in the system, judicial review exists as part of a dynamic
process by which societies continuously reform themselves and their
institutions over time to insure a balance between the contradictory
goals of adhering to foundation norms and allowing for change. As is
the case with so many other aspects of legal practice, judicial review
makes it necessary to pose a host of questions about the agents doing
the review: Who reviews the founding document? What standards of
practice ought to guide them? Are they freely choosing agents or part
of a larger social structure? What qualities of mind do they have?
What qualities of mind ought they to have? To not address sociological
questions about the nature of real people in discussing the interpretation
of legal texts is like discussing the plot of Ulysses without
reference to the character of the man who tried to save his ship and
his men.' 1
Although the question of how human beings follow practices
thus assumes great importance for Fish, his analysis of what a practice
means is to quote a pitcher for the Baltimore Orioles. Like Dennis
Martinez, who just throws the ball to get the batter out without
thinking about ultimate goals, an agent "need not look to something
in order to determine where he is or where he now might go because
that determination is built into, comes along with, his already-in-place
sense of being a competent member of the enterprise.' 2 Agents, in
Fish's view, are part of a preformative chain, not one, to be sure, of
the automatic transmittal of information without consciousness, but
nonetheless one that works automatically and without requiring autonomy
and self-judgement. There is no autonomy in this view, or,
more precisely, all autonomy lies with the historical events that determined
the patterns of a practice. Individuals, being "deeply situated,"'73
just follow rules, rules which themselves are so deeply
situated that individuals may not. and probably are not, aware that
they are following them. "'Be the best you can be,'" Fish writes in
response to Dworkin, "finally means nothing more than 'act in the
70 J.Elster, Ulysses and the Sirens (rev. ed. 1984).
71 Although I am in strong disagreement with his moral philosophy, I am in agreement
with Maclntyre that it is impossible to understand Greek conceptions of virtue without discussing
the matter of character. A. Maclntyre, After Virtue (1981).
72 S. Fish, supra note 5, at 388.
73 Id. at 387.
1430 [Vol. 11: 1409
ALGORITHMIC JUSTICE
way your understanding of your role in the institution tells you to
act.' 9-74
Since agents naturally follow the rules determined by their roles
in the institutions that define their practices, little would be amiss if
Fish were to focus on those institutions themselves. After all, if the
members of a professional subcommunity developed through some
kind of democratic practice an agreed-upon set of norms, little would
be wrong in expecting that each of them would then bind themselves
to the rules that defined their practice. Yet, no subcommunity can
possibly develop such standards fairly without agreement upon larger
normative and procedural issues, such as that their decisions will be
made by majority rule, that the practices to which they adhere will
not violate Judeo-Christian beliefs, that their behavior will be guided
by law, etc. In other words, the development of a theory about how
agents act is intimately linked to a discussion of the larger, normative
standards of the society (not just the subcommunity) in which agents
practice. If there are norms viewed as just in the society and practices
viewed as procedurally correct in the professional subcommunity,
then agents can follow rules naturally, or even algorithmically, without
violating their autonomy.
The problem, of course, is that Fish believes that transcendental
standards ofjust or moral behavior can not exist. Within the confines
of that argument, the failure to look beyond the automatic and natural
following of the rules of a practice becomes a serious matter indeed.
As he often points out, Fish is neither an anarchist nor a
nihilist. Quite the contrary. Like many thinkers attracted to algorithmic
imagery, he imagines structures that are so tightly organized
as to make anarchism impossible. For this reason, Fish, for all
his distaste for formal algorithmic thinking, concludes with a position
not all that distinct from Chomsky's. To be sure, the rules of transformational
grammar work in such a way that the personal proclivities
of the speaker are irrelevant, whereas the rules of thumb of a
practice depend very much on the idiosyncracies of the person engaging
in the practice." Yet in both cases particulars are embedded in
generals, in the one case the rules of grammar, in the other "the individual
who is always constrained by the local or community standards
and criteria of which his judgement is an extension."76
74 Id. at 391.
75 Id. at 317.
76 Id. at 323.
14311990]
CA RDOZO LA W REVIE W
IV.
Those skeptical of the possibility of any transcendental standard
of justice posit that none of the theories we may have about justiceboth
of the transcendental and anti-transcendental sort-matter.
Even the most skeptical, such as Fish, do, however, believe that theory
talk matters, even if theory does not.77 But what if theory talk is
theory? It would be under the assumptions I have been making in
this paper. To be sure, there is no such thing as a view from nowhere,"8
universally valid, from which we can deduce standards of
justice that bind our actions for all time. Yet purely contingent understandings
of what binds us together, being contingent, cannot bind,
or at least bind what we need to be bound. Already existing interpretative
sub-communities, which by definition share normative standards,
are not the ones that need grand narratives about justice, but
instead communities seeking to answer questions such as these: How
do we resolve our disputes with one another and make those resolutions
binding? How do we aim to make our resolutions as fair as
possible to the parties to the conflict? How does a community exist in
time, passing on the rules by which it regulates its affairs to those who
did not participate in the original making of such rules? What do we
expect from newly admitted members of our community in return for
their membership in the community? Who makes the rules? Who
follows them? Who questions them? Who changes them? Social justice
exists when diverse communities can be knitted together because,
whatever their other differences, the one thing they require is some
normative consensus, however vague, about the purposes that define
their society.
Faced with the dilemma of how we can develop standards ofjustice
that are more than contingent and local but less than universal
and permanent, we rely on the minimal philosophical anthropology
that even postmodernism, indirectly, concedes. Since we are all members
of a larger community governed by a text and the interpretations
of that text we bring to it, we seek standards ofjustice that recognize
and allow us to develop our capacity as readers and interpreters.
Those who do what Fish calls theory ti k are engaged in the process
of sharpening and refining the standards by which real human beings
interpret texts. They are setting an example, using their powers as
thinkers and writers to create provisional standards of justice, recognizing
that these are socially created rather than found in nature or
77 Id. at 14.
78 T. Nagel, The View from Nowhere (1986).
1432 [Vol. 11:1409
ALGORITHMIC JUSTICE
theology, and that, because these standards are recognized as minimally
transcendental, we expect that they will, before changing, last
for a considerable period of time-say across two generations-and,
during that time, be accepted as generally binding.
If that is the case, then the most likely place to find a standard of
justice lies, after all, in the major texts that frame talk about theorythe
Constitution, decisions of the important courts, and articles in law
reviews and similar outlets that debate both specific laws as well as
standards of interpretation. From such a point of view, writers like
Fish, by engaging in theory talk, contribute to theory. What they say
matters a great deal and, moreover, matters in a transrhetorical way,
just as I believe that what I am saying, while rhetorically presented,
also involves more than rhetoric. One embarks, Fish claims, on a
slippery slope down the anti-formalist road: once you question formalism's
first assumption, you have no choice but to question them
all. Yet, as I have tried to show in this paper, there is a road back up
the slope again. Once we understand that whatever permanence and
universality we lose in any transcendental standard of justice is more
than compensated for by the recognition that the transcendence we
get, however temporary, is a product of our own efforts, we require a
sociological theory of the self to'put back into pe6ple and their efforts
and practices the meanings about justice that we have, rightly,
stripped away from texts.
The failure of postmodern theorists ofjustice to develop an adequate
philosophical anthropology dealing with the capacities of
human selves undermines much of the strength of its critique of intentionalism
and other problematic theories of interpretation. The notion
that truths are not embedded in texts in such a fashion that we
can divine what to do simply by reading the words is a profound idea.
In turning the question of justice back to us, to those who read and
interpret texts, postmodernism makes possible a self-governing political
community capable of interpreting its rules for the benefit of its
members. But that potential can only be realized, not by denying humanism,
but by welcoming it, by recognizing that what makes us
human is our ability to shape and interpret rules according to the
contexts in which we find ourselves. If that means that we have to
accept at least some minimal transcendental standards and distinctions-that,
for example, there is a difference between nature and culture,
that humans do have special abilities, that the socially
constructed can be transcendental without necessarily being permanent-then
this is a small price to pay for gaining control over the
rules that we simultaneously make and follow. Why bother to argue
1990] 1433
1434 CARDOZO LAW REVIEW [Vol. 11:1409
that the rules are not made by God or nature only to argue instead
that, however they are made, our only choice is to follow them rather
than remake them through all the practices in which we engage.