Citation: 110 Nw. U. L. Rev. 1169 2015-2016
Content downloaded/printed from HeinOnline
Fri Jul 21 11:53:31 2017
-- Your use of this HeinOnline PDF indicates your acceptance
of HeinOnline's Terms and Conditions of the license
agreement available at http://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
Copyright 2016 by Toni M. Massaro and Helen Norton Printed in U.S.A.
Vol. 110, No. 5
SIRI-OUSLY? FREE SPEECH RIGHTS AND
ARTIFICIAL INTELLIGENCE
Toni M Massaro& HelenNorton
ABSTRACT--Computers with communicative artificial intelligence (Al)
are pushing First Amendment theory and doctrine in profound and novel
ways. They are becoming increasingly self-directed and corporal in ways
that may one day make it difficult to call the communication ours versus
theirs. This, in turn, invites questions about whether the First Amendment
ever will (or ever should) cover Al speech or speakers even absent a
locatable and accountable human creator. In this Article, we explain why
current free speech theory and doctrine pose surprisingly few barriers to
this counterintuitive result; their elasticity suggests that speaker humanness
no longer may be a logically essential part of the First Amendment
calculus. We further observe, however, that free speech theory and doctrine
provide a basis for regulating, as well as protecting, the speech of
nonhuman speakers to serve the interests of their human listeners should
strong Al ever evolve to this point. Finally, we note that the futurist
implications we describe are possible, but not inevitable. Moreover,
contemplating these outcomes for Al speech may inspire rethinking of the
free speech theory and doctrine that make them plausible.
AUTHORS-Toni M. Massaro, Regent's Professor, Milton 0. Riepe Chair
in Constitutional Law, and Dean Emerita, University of Arizona James E.
Rogers College of Law. Helen Norton, Professor and Ira C. Rothgerber, Jr.
Chair in Constitutional Law, University of Colorado School of Law. Our
thanks to the Northwestern University Law Review and Alexander Tsesis
for their excellent leadership of this symposium, and to Barbara Babcock,
Jack Balkin, Derek Bambauer, Jane Bambauer, Ashutosh Bhagwat,
Andrew Coan, Danielle Keats Citron, Stephen Golden, Margot Kaminski,
Andrew Koppelman, Genevieve Leavitt, Elizabeth Massaro, Julia Massaro,
David McCraw, Paul Ohm, Mary-Rose Papandrea, Scott Peppet, Blake
Reid, Nancy Sharkey, Harry Surden, Susan Tarrence, and Alexander Tsesis
for their very thoughtful comments. Genevieve Leavitt inspired the title.
1169
NORTHWESTERN UNIVERSITY LAW REVIEW
INTRODUCTION ........................................................................................................... 1170
I. FREE SPEECH THEORY AND STRONG AI SPEAKERS ................................................ 1175
A. Arguments Basedon DemocracyandSelf-Governance.............................. 1176
B. Arguments Basedon the MarketplaceofIdeas............................................ 1178
C. Arguments Basedon Autonomy ................................................................... 1178
II. FREE SPEECH DOCTRINE AND STRONG Al SPEAKERS ............................................. 1183
A. The ProtectionofNontraditionalSpeakers ................................................. 1183
B. The ProtectionofNontraditionalContent................................................... 1186
C. PotentialLimits andDistinctions:Speech or Conduct?.............................. 1186
III. WHAT NEXT? IMPLICATIONS AND LIMITATIONS ..................................................... 1189
CONCLUSION .............................................................................................................. 1192
INTRODUCTION
We live in an age where vast amounts of information and
communication are produced, gathered, synthesized, and disseminated
through increasingly sophisticated expressive technologies. Humans have
achieved unprecedented mastery over computerized services and products
that "speak" in different ways, even as new forms of communicative
technology seem to have gained considerable dominion over us.
We welcome their movie, restaurant, and book selections, not to
mention their ability to guide airplanes and surgeons, keep us safer from
domestic and foreign perils, help us avoid bad financial and health
decisions, and foil sneaky consumer scams. We find eerie comfort in Siri's
liquid voice as it (she?) answers our random questions and wake-up call
requests, and we gratefully obey the stem Garminadmonition to make a Uturn
as soon as safely possible when we have lost our way in a foreign
place.
But we are ambivalent about this heightened computerized attention to
and fulfillment of our basic needs. The Spike Jonze movie Her captures
this eloquently. We are guided through a fantasy about technology
designed with sufficient human-like qualities to relieve loneliness-one of
the most powerful sources of human suffering. Then we confront the
dystopian downside of love for an operating system that has no
programmed capacity for genuine regret or fidelity. It does not care for us,
in the end, any more than the most uncaring humans do.
We also worry that these highly anthropomorphized but not human
creations are controlling us more than we intended. Barbie dolls,
automobiles, and other consumer products now can communicate about us
1170
110:1169 (2016)
via the Internet with other computers and devices., Algorithms drive the
stock market, credit decisions, and social reputation in ways few of us
understand.' Some computer companions "die" if left unattended and thus
have the power to pull on our emotions.'
The list of potentially life-altering communicative technologies grows
daily. These developments provoke in us a mixture of apprehension and
excitement. We are ambivalent in part because technology is like money: it
is neither inherently good nor bad. Thus to declare it uniformly good or
bad, useful or disruptive, presumptively protected from government
regulation or presumptively subject to regulation, would be foolish. It
should and will depend on context, and on what the new technology does to
us and for us.
Our ambivalence also springs from the many unknowns. Much will
depend on how artificial intelligence (AI) and other new technologies
evolve. Like money--especially large concentrations of it-new
technologies likely will change how we look at many settled conventions,
including legal conventions. As Jack Balkin noted over a decade ago, new
technologies can make salient what went unnoticed before. 4 They also may
evade conventional legal categories in ways that will push courts to
redefine the older categories, with effects we find difficult to fully imagine
in advance. Much of this will occur contextually, as the ways in which
humans actually use new technologies shape the legal doctrine designed to
govern them. But significant legal change, including constitutional change,
seems inevitable.
To be sure, the prospect of change is old news. That new machines
can destabilize or compel adjustments to old legal orders is an enduring
problem. That machines can improve human lives but also cause human
suffering likewise is no surprise. We have a long history of managing the
challenges that machines can pose to our legal and constitutional certitudes.
We now face such challenges in First Amendment law as computers
gain exceptional speech-relevant capacities. Modern computers can gather,
I See Irina D. Manta & David S. Olson, Hello Barbie:FirstThey Will MonitorYou, Then They Will
DiscriminateAgainstYou. Perfectly., 67 ALA. L. REV. 135, 136-37, 179 & n.232 (2015) (discussing the
built-in hardware and software that permits these devices to interact with other devices).
2 See FRANK PASQUALE, THE BLACK Box SOCIETY: THE SECRET ALGORITHMS THAT CONTROL
MONEY AND INFORMATION 29-35 (2015) (expressing concerns about data-driven management, given
the opacity of data's origins and destinations, the risk of bias and error within them, and possible
cascade effects if information in one piece of software is repeated in systems throughout the economy).
3 See ROSAIND W. PICARD, AFFECTIVE COMPUTING 109-11 (1997) (describing a virtual pet that
the owner must feed, groom, and play with to prevent it from "dying" of neglect).
4 Jack M. Balkin, Commentary, DigitalSpeech andDemocratic Culture:A Theory of Freedomof
Expressionfor the InformationSociety, 79 N.Y.U. L. REV. 1(2004).
1171
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
create, synthesize, and transmit vast seas of information even as they
become more "human-like": they are increasingly interactive, affective,
and corporal.5
Such computer speakers also are increasingly self-directed or
"autonomous"-which is to say, the computer generates content further
afield from human direction. Some forms of Al already are better speakers
than humans in certain respects, given their superior ability to evade some
of the distortions of bias and baser emotions, their immunity from fatigue
or boredom, and their capacity to manage complex ideas in ways mere
humans cannot. Scientists also are at work designing computers with
enhanced emotional intelligence and other features that may narrow the gap
between the capacities of human and computer speakers.6 At some point,
one might imagine such computer speakers may be disconnected enough
and smart enough to say that the speech they produce is theirs, not ours,
with no human creator or director in sight.
This Article considers the potential First Amendment consequences of
such an evolution. More specifically, it engages in a thought experiment as
to whether computer speakers with strong Al might ever be treated as
speakers covered by the First Amendment.7
We conclude that this is entirely plausible. We do not mean to
overclaim: we are not insisting that free speech law requires First
Amendment coverage for computer speakers, nor are we insisting that
strong Al inevitably will happen.' To be clear, we understand that as
5 Illah Reza Nourbakhsh describes robots as "a new form ofliving glue between our physical world
and the digital universe we have created ....They will be embedded in our physical spaces-our
sidewalks, bedrooms, and parks-and they will have minds of their own thanks to artificial intelligence
(A)." ILLAH REZA NOURBAKHSH, ROBOT FUTURES, at xiv-xv (2013).
6 See generally RAFAEL A. CALVO & DORIAN PETERS, POsrrIvE COMPUTING: TECHNOLOGY FOR
WELLBEING AND HUMAN POTENTIAL (2014); BLUEPRINT FOR AFFECTIVE COMPUTING: A
SOURCEBOOK (Klaus R. Scherer, Tanja Banziger & Etienne B. Roesch eds., 2010); THE OXFORD
HANDBOOK OF AFFECTIVE COMPUTING (Rafael A. Calvo, Sidney D'Mello, Jonathan Gratch & Arvid
Kappas eds., 2015); PICARD, supranote 3.
7 We refer to these as-yet-hypothetical machines that actually think as "strong AIs," as opposed to
"weak AI" machines that "act as ifthey were intelligent." STUART J. RUSSELL & PETER NORVIG,
ARTIFICIAL INTELLIGENCE: A MODERN APPROACH 1020 (3d ed. 2010); see alsoHarry Surden, Machine
Learning andLaw, 89 WASH. L. REV. 87, 97 (2014) (describing ways in which machines can "learn"
through employing heuristics and proxies "that ultimately arrive at the same or similar results as would
have been produced by a similarly situated intelligent person employing higher order cognitive
processes and training").
8 See, e.g., Ryan Calo,Robotics andthe Lessons of Cyberlaw, 103 CALIF. L. REV. 513, 528 (2015)
("Little in the literature gives me confidence that artificial intelligence will approximate human
intelligence in the foreseeable future. There are analytic and technical reasons to believe robots will
never think like people.").
1172
110:1169 (2016)
communicative technologies continue to develop, their trajectory towards
the strong Al of our imagination remains very uncertain. We point out
simply that very little in foundational free speech theory and doctrine rules
out coverage for these as-yet-hypothetical speakers.
Imagine, for example, a Supreme Court case called Robots Unitedthat
involves a First Amendment challenge to the government's regulation of
speech by a robot with strong Al, where the regulation restricts speech on a
matter of public concern based on the expression's content and its
speaker's identity. Judy Jetson-a former Solicitor General of the United
States now representing private parties-is the robot's lawyer and argues
that the robot itself has free speech rights. Nobody in the courtroom laughs
out loud, though some remain skeptical. The Justices realize that precedent
already affords free speech coverage to the speech of artificial entities as
well as to those who lack fully mature autonomous capacity. Indeed, the
case law hinges more on pragmatism and on expression's informational
value than on any philosophical purity about speaker personhood or rights.
This audience-focused perspective may lead the Court to a remarkable
conclusion: speaker humanness may be a sufficient but not logically
essential part of its theoretical and doctrinal constitutional calculus.' Judy
Jetson and her robot client may prevail.
Many no doubt will resist this conclusion. Their first response may be:
"No speech rights for computers!" Constitutional rights, they will insist,
should attach to the humans who create and use the technologies, but not to
the tools themselves (and to date we can still identify a human speaker as
the "rights holder" in the variety of technologies that currently exist). Many
will likely feel that computers with strong Al always will be "missing
something" that humans possess and that seems indispensable to free
speech rights. 1°
9 See Toni M. Massaro, Some Realism About ConstitutionalLiberalism,28 CONST. COMMENT. 383
(2013) (discussing the tension between liberal rights and communitarian ends posed by constitutional
liberalism and free speech libertarianism).
10See Lawrence B. Solum, Legal PersonhoodforArtificial Intelligences,70 N.C. L. REv. 1231,
1262-76 (1992). The arrow, though, points in both directions, as demonstrated by emerging debates
about the enhancement of humans and moral rights. See generally ALLEN BUCHANAN, BEYOND
HUMANITY?: THE ETHICS OF BIOMEDICAL ENHANCEMENT 209-36, 214 (2011) (discussing the moral
rights implications of biomedically enhanced humans, and how humanness may be a sufficient but not
necessary predicate to rights if "post-humans" were to have capabilities or interests that ground rights).
The ultimate questions about whether Al resembles human intelligence enough for legal purposes are
inherently interdiaciplinary. Sec Stuart Watt, Can Peoplo Think? Or Machines?A Unified Protocolfor
Turing Testing, in PARSING THE TURING TEST: PHILOSOPHICAL AND METHODOLOGICAL ISSUES IN THE
QUEST FOR THE THINKING COMPUTER 301, 301-18 (Robert Epstein, Gary Roberts & Grace Beber eds.,
2009); cf Jonathan Zittrain, UbiquitousHuman Computing, 366 PHIL. TRANSACTIONS ROYAL SOC'Y A
3813, 3813, 3817 (2008) (discussing development of "cheap and plentiful sensors, fast processors and
1173
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
We understand these intuitions, but we explain that surprisingly little
in contemporary First Amendment theory or doctrine blocks the path
towards strong Al speakers' First Amendment protection. Moreover,
because speech generated by computers can offer enormous value to
human listeners (and other human users), the risk that government
suppression of that speech will compromise important free speech interests
further points towards coverage of such speakers. Imagine, for example, an
Orwellian government that restricts Al speech that does not sing the
government's tune. If interpreting the First Amendment to protect strong
AIs as speakers is necessary to prevent this, courts may well do so.
At the same time, the choice to cover such speakers carries its own
sobering risks. Like corporations, smart machines and their outputs already
wield great social and economic power. They already have the capacity to
inflict grave harms to human autonomy, dignity, equality, and property.
With fortified constitutional armor, new technologies may deflect worthy
forms of government regulation and thus alter the relationship between
humans and machines in profound and unfortunate ways.
On the other hand, these problems may be judicially manageable, as
free speech theory and doctrine also contain pockets of resistance to
unbounded speech protection on which courts can draw to regulate
computer speakers. This is especially the case where the communicative
relationship calls for us to privilege listeners' First Amendment interests
and thus permit content-based regulation designed to address the harms of
certain expression. In short, interpreting the First Amendment to cover
strong Al speakers would not necessarily mean that human needs no longer
play the primary role in the First Amendment analysis.
We also anticipate concerns about the practical and normative
difficulties in giving free speech rights to strong Al speakers. We see these
difficulties too. But law may adapt to these difficulties to address harmful
consequences while preserving valuable speech, as the normative and
practical concerns might well inspire closer examination and even revision
of the free speech doctrine and theory we now have.
For example, because the prospect of conferring free speech rights on
nonhuman speakers may be counterintuitive and perhaps deeply troubling
to many, it may inspire a shift in doctrine. In other words, even though
Citizens United and a range of other developments have pushed us in the
high-speed saturating wireless networks" that may "allow much human thinkingto be as far as we like
from the people initiating it and using itsresults" and noting that "[i]n discovering how to farm out
through the Net the tasks that computers cannot do, we may find ourselves treating-and makingpeople
more like computers").
1174
110:1169 (2016)
"speech, not speakers" direction," the logical extension of continuing in
that direction may be so uncomfortable that it inspires a rethinking of
current theory and doctrine. As just one possibility, recall that although
corporations have been held to be First Amendment rights holders, they
have long been considered to hold "derivative" First Amendment rights to
speak in ways that inform natural persons, rather than holding rights for
their own sake or "in their own right. ' 12 Courts may similarly treat strong
Al speakers only as derivative rights holders, with rights that may differ
from those held by natural persons. In short, nothing we say here about the
implications of existing law and theory means these foundations could not
be, or should not be, revised.
In Parts I and II of this Article, we explain why very little in current
free speech theory or doctrine makes First Amendment coverage contingent
upon a human speaker. In Part III, we briefly outline some of the
implications-both positive and negative-of this development and discuss
ways in which courts might manage those implications going forward.
I. FREE SPEECH THEORY AND STRONG Al SPEAKERS
This Part explains how the elasticity of modem theoretical free speech
justifications makes it difficult to place nonhuman speakers wholly outside
their embrace. Many similarities exist between much computer speech and
human speech that we already protect, especially if we focus primarily on
expression's value to listeners, rather than on its source.
As many have observed, there is no unifying theory of the First
Amendment. 13 The most influential theories have been clustered into
arguments based on democracy and self-governance, 4 a marketplace of
ideas model, 5 and autonomy. 6 Each of the theories identifies freedom of
11See Citizens United v. FEC, 558 U.S. 310, 392 (2010) (Scalia, J., concurring) ("The [First]
Amendment is written in terms of'speech,' not speakers.").
12 The longstanding premise that corporate speakers are not dignitary speakers in their own right,
however, now faces growing pressure. See, e.g., Rodney A. Smolla, Freethe Fortune 500! The Debate
over CorporateSpeech and the First Amendment, 54 CASE W. REs. L. REv. 1277, 1295-96 (2004)
(noting a paradigm shift in which the focus of the Court's commercial speech doctrine "has moved from
consumer protection to speaker protection").
13 See, e.g., HARRY KALVEN, JR., A WORTHY TRADITION: FREEDOM OF SPEECH IN AMERICA 3
(Jamie Kalven ed., 1988) ("The Court has not fashioned a single, general theory which would explain
all of its decisions; rather, it has floated different principles for different problems.").
14 See, e.g., ALEXANDER MEIKLEJOHN, FREE SPEECH AND ITS RELATION TO SELF-GOVERNMENT
(1948).
15See Abrams v. United States, 250 U.S. 616, 630 (1919) (Holmes, J., dissenting) ("[T]he best test
of truth is the power ofthe thought to get itself accepted in the competition of the market, and that truth
is the only ground upon which their wishes safely can be carried out.").
1175
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
thought as a central First Amendment concern. Each defines its boundaries
mindful of ideals about a hoped-for greater good and assumes that it will
produce something beneficial. Although the variations among them are
relevant to how strong Al speakers fit in the free speech universe, none
rules out such speakers as part of that universe. A brief survey
demonstrates why.
A. Arguments Basedon DemocracyandSelf-Governance
Arguments rooted in democracy and self-governance link freedom of
speech to the political cornerstones of liberal democracy and to notions of
public discourse. For example, Alexander Meiklejohn famously noted that
what matters for freedom of speech is not that all speak, but that
"everything worth saying shall be said."'7 Taken literally, speaker identity
should be irrelevant to Meiklejohn's inquiry, and strong Al speech should
be protected no less than human speech provided that its speech contributes
to the democratic process-i.e., that it is "worth saying."
More recently, Robert Post draws his theory of freedom of expression
from principles of self-government under which there must be a "chain of
communication... 'sufficiently strong and discernible' to sustain the
popular conviction that representatives spoke for the people whom they
purported to represent."' 8 For Post, the First Amendment is "designed to
protect the processes of democratic legitimation."' 19 In his view, because
corporations do not themselves "experience the value of democratic
legitimation,"2 they do not themselves hold free speech rights equivalent to
individuals but instead hold derivative First Amendment rights to speak in
ways that "may be useful to natural persons who seek to participate in
public discourse."'" In other words, corporations "do not possess original
16 See, e.g., C. EDWIN BAKER, HUMAN LIBERTY AND FREEDOM OF SPEECH (1989) (emphasizing
individualistic concerns and speaker liberty); RONALD DWORKIN, TAKING RIGHTS SERIOUSLY 201-05
(1977) (focusing on speaker dignity and respect). We include Seana Valentine Shiffrin's work on a
thinker-based First Amendment under the autonomy theory umbrella. See Seana Valentine Shiffrin, A
Thinker-BasedApproach to FreedomofSpeech, 27 CONST. COMMENT. 283 (2011).
17 ALEXANDER MEIKLEJOHN, POLITICAL FREEDOM: THE CONSTITUTIONAL POWERS OF THE
PEOPLE 26 (1965).
18 ROBERT C. POST, CITIZENS DIVIDED: CAMPAIGN FINANCE REFORM AND THE CONSTITUTION 8
(2014) (quoting JAMES WILSON & THOMAS MCKEAN, COMMENTARIES ON THE CONSTITUTION OF THE
UNITED STATES OF AMERICA 30-31 (1792)).
'9 Id. at 41.
20 Id.at 69.
21 Id. at 73-74.
1176
110:1169 (2016)
First Amendment rights to participate in public discourse as speakers," but
they can be rights holders in ways that differ from natural persons.22
The logical extension of Post's theory to strong Al speakers is that
such speakers should also be protected if and when they produce
information useful to natural persons who seek to participate in public
discourse.23 That a computer, not a human, produces the useful information
should not matter. To be sure, under this view, limits can and should be
imposed where the speech does not serve this audience-sensitive value, and
strong AIs as derivative rights holders may hold rights that differ from
those held by natural persons.
Other democratic speech theorists, such as Jack Balkin, argue that
emerging communicative technologies require a refocus of free speech
theory to protect democratic culture.24 Balkin defines democratic culture as
"a culture in which individuals have a fair opportunity to participate in the
forms of meaning making that constitute them as individuals. '25 That is, he
goes beyond representative democracy justifications for free speech. His
primary anxiety is that technologies promise wider participation but also
carry the means of controlling democratic participation in new ways, and
he argues for attention to the latter in theorizing about First Amendment
constraints on regulation of digital networks.26 Balkin's account focuses
directly on humanness when he notes that: "Human beings are made out of
culture. A democratic culture is valuable because it gives ordinary people a
fair opportunity to participate in the creation and evolution of the processes
of meaning-making that shape them and become part of them." 7 But he
22 Id.at 71-74 ("[O]rdinary commercial corporations have neither the right nor the responsibility to
contribute their views to public opinion. Instead, ordinary commercial corporations have the right only
to publish such information as may be useful to natural persons who seek to participate in public
discourse."). Post characterizes robots as similarly unable themselves to participate in democratic
legitimation. Id.at 68 ("The value of democratic legitimation applies to persons, not to things. If there
were a self-perpetuating viral communication on the Intemet, it would not possess First Amendment
rights. This is because computer programs cannot experience the value of democratic legitimation. That
is why the speech ofrobots does not form part of public discourse.").
23 Note that we are not attributing this view to Post; we argue instead it flows from his theory of
democratic participation as the rationale for protecting speech.
24 Balkin, supra note 4; see also Jack M. Balkin, Cultural Democracyand the First Amendment,
110 Nw. U. L. REv. 1053, 1060 (offering a liberty-democratic theory of free speech that considers the
vast range of influences that shape human identity, and that uses the word "democratic" to mean
"cultural participation-the freedom and the ability of individuals to participate in culture, and
especially a digital culture").
25 Balkin, supra note 4, at 3.
21 Id.at 2-3.
27 Id.at 33.
1177
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
then adds that the "processes of meaning-making include both the ability to
distribute those meanings and the ability to receive them."2
Human creativity is sparked by an endless array of cultural stimuli,
and Al speech can contribute to receivers' meaning-making too. Balkin's
democratic culture perspective thus would not rule out cases in which
strong Al speakers contribute to the democratic disco. Indeed, Balkin's
more explicitly ecumenical account of how humans make meaning-from
a wide variety of idiosyncratically relevant sources-renders such
computer speech more obviously important than do more traditional, public
discourse models.
B. Arguments Basedon the MarketplaceofIdeas
Like democracy-based theories, the marketplace of ideas justification
for free speech rests largely on expression's instrumental value to listeners'
enlightenment.29 It too emphasizes the production of information regardless
of source, and assumes that unfettered speakers facilitate listeners'
discovery of truth and distribution ofknowledge through a robust exchange
of ideas.3" If anything, it casts a wider net than self-governance theories in
that it finds First Amendment value in a greater variety of speech that has
nothing to do with democratic participation. Here too expression's
nonhuman source does not strip it of its First Amendment value to human
listeners.
C. Arguments Basedon Autonomy
Autonomy-based theories are arguably both the most promising and
most potentially limiting sources of strong Al speakers' free speech rights.
To the extent that they emphasize the autonomy of human listeners, of
course, autonomy-based theories fortify arguments for strong AIs' free
speech rights, as machines can and do produce information relevant to
human listeners' autonomous decisionmaking.
Only theories based solely on speaker autonomy pose potential
roadblocks for protecting strong Al speakers. Such arguments relate most
directly to philosophical theories about the moral "person," and require a
working definition of the sorts of qualities or attributes necessary to confer
such a status. The late Joel Feinberg offered an illustrative example:
28 Id.at 37.
29 See FREDERICK SCHAUER, FREE SPEECH: A PHILOSOPHICAL ENQUIRY 15-34 (1982); Vincent
Blasi, Holmes andthe MarketplaceofIdeas, 2004 SUP. CT. REv. 1, 33-44.
30 See Thomas I. Emerson, FirstAmendment Doctrine and the Burger Court,68 CALIF. L. REV.
422, 423 (1980); Frederick Schauer, The Boundaries of the First Amendment: A Preliminary
Explorationof ConstitutionalSalience, 117 HARv. L. REv. 1765, 1786 (2004).
1178
110:1169 (2016)
The characteristics that confer commonsense personhood are not arbitrary
bases for rights and duties, such as race, sex, or species membership; rather
they are the traits that make sense out of rights and duties and without which
those moral attributes would have no point or function. It is because people
are conscious; have a sense of their personal identities; have plans, goals, and
projects; experience emotions; are liable to pains, anxieties, and frustrations;
can reason and bargain, and so on-it is because of these attributes that people
have values and interests, desires and expectations of their own, including a
stake in their own futures, and a personal well-being of a sort we cannot
ascribe to unconscious or nonrational beings. Because of their developed
capacities they can assume duties and responsibilities and can have and make
claims on one another. Only because of their sense of self, their life plans,
their value hierarchies, and their stakes in their own futures can they be
ascribed fundamental rights.31
We are not the first to consider whether such notions of personhood
should deny the possibility of constitutional rights for machine speakers.
Over twenty years ago, Lawrence Solum directly addressed whether an Al
should receive constitutional rights "for the Al's own sake."32 Solum
identified several ways in which AIs might be thought to be "missing
something" for purposes of constitutional protection-for example, they
may lack souls, consciousness, intentionality, feelings, interests, and free
will.33 Solum nevertheless found it difficult to conclude that any of these
"deficits" definitely ruled out machines' constitutional protection as
speakers; indeed, he wondered whether they really were deficits.3 4 In light
of the many unresolved questions about Als' development, Solum
concluded that "[i]f AIs behaved the right way and if cognitive science
confirmed that the underlying processes producing these behaviors were
relatively similar to the processes of the human mind, we would have very
good reason to treat AIs as persons."35 In other words, the personhood
barrier for First Amendment protections could be overcome either if we
changed how we view "persons" for practical or other reasons, or if
computers came to function in ways that satisfied our criteria for
personhood.
We are now seeing changes in both areas. First, free speech theory has
moved away from a construction of legal personhood that views speakers
31 JOELFEINBERG, FREEDOM AND FULFILLMENT: PHILOSOPHICAL ESSAYS 52 (1992).
32 Solum, supranote 10, at 1258-79.
33 Id. at 1262-76.
34 Id. As for feelings and awareness of others, for example, Solum had this to say: "Emotion is a
facet of human mentality, and if the human mind can be explained by the computational model, then
emotion could turn out to be a computational process." Id. at 1270.
35 Id.at 1286.
1179
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
solely through an individual or animate lens. Speakers are increasingly
defined in a practical, non-ontological sense that does not rely on the sorts
of criteria for moral personhood identified by Feinberg (and others).36 In a
recent, thoughtful consideration of the role of personhood and rights for
machines, for example, Samir Chopra and Laurence White conclude that:
[T]he granting of legal personality is a decision to grant an entity a bundle of
rights and concomitant obligations. It is the nature of the rights and duties
granted and the agent's abilities that prompt such a decision, not the physical
makeup, internal constitution, or other ineffable attributes of the entity.37
Legal persons thus already include not only individuals, but also
corporations, unions, municipalities, and even ships, though the law makes
adjustments based on their material differences from humans.3" "Legal
persons" often hold a variety of legal (including constitutional) rights and
duties even though they may be very different from "moral" or "natural" or
"human" persons. They can sue and be sued, for example. Stating that
some class of nonhuman speakers may be rights holders in certain contexts
simply means that that they are legal persons in those contexts-and, to
date, human status is not a necessary condition for legal personhood.39 To
be sure, not all rights are, or should be, necessarily available to all legal
persons. For example, that a legal person has the right to sue and be suedor
to speak-does not necessarily mean that it has, or should have, the right
36 See EVAN FoX-DECENT, SOVEREIGNTY'S PROMISE 38 (2011) ("The defining hallmark of
liberalism is that the ultimate unit of moral value is the individual. For law, however, the unit to which
rights and duties attach is the legal person. The two are not the same. An individual has legal
personality, but so do a wide variety of groups, such as unions, corporations, communities, Indigenous
peoples, and municipalities. The rule of law applies to legal persons, and not just to individuals."
(footnote omitted)).
37 SAMIR CHOPRA & LAURENCE F. WHITE, A LEGAL THEORY FOR AUTONOMOUS ARTIFICIAL
AGENTS 155 (2011).
38 Id. at 157-58; Fox-DECENT, supra note 36, at 38; see also Margaret M. Blair & Elizabeth
Pollman, The DerivativeNature of CorporateConstitutionalRights, 56 WM. & MARY L. REV. 1673,
1678 (2015) (explaining how corporations, unlike natural persons, have long been treated as derivative
rights holders).
39 See Solum, supranote 10, at 1238-39 ("'[I]n common speech, "person" is often used as meaning
a human being, but the technical legal meaning ofa "person" is a subject of legal rights and duties.' The
question whether an entity should be considered a legal person is reducible to other questions about
whether or not the entity can and should be made the subject of a set of legal rights and duties. The
particular bundle of rights and duties that accompanies legal personhood varies with the nature of the
entity. Both corporations and natural persons are legal persons, but they have different sets of legal
rights and duties." (footnotes omitted) (quoting JOHN CH'MAN GRAY, THE NATURE AND SOURCES OF
THE LAW 27 (Roland Gray ed., MacMillan 1921))).
1180
110:1169 (2016)
to vote or a right to privacy.4°
In this Article we take care to focus only on
the possibility of free speech rights for strong AIs, and not on any other set
of constitutional rights.
Second, technology is changing in ways that may at some point enable
some computers to satisfy certain criteria for legal personhood. For
example, one difference between computers and humans used to be humanlike
corporality. That difference is rapidly disappearing, as some computers
are now being inserted into sophisticated and human-like physical shapes.
As Ryan Calo recently observed, "robots, more so than any technology in
history, feel to us like social actors. '41 Although embodiment surely will
affect many important legal and policy issues,4 nothing in having a
physical body need determine (though it may enhance) the "selfhood"
principles of freedom of expression identified here.
Computers' inability to experience emotions offers another potential
source of distinction. Computer-generated speech-whether robotic or
detached from a human-like form--does not entail a speaker in possession
of human emotions, with emotions' speech-curbing as well as speechgenerative
potential. Nor does a computer have the human need or desire,
one assumes, to communicate noninteractively with itself in the way a
person might write poetry or a diary with no intention of sharing this with
others.43 But here too things are changing with emerging developments in
affective computing.44 Although computers today are more accurately
described as capable of expressing emotions rather than as having them,45
some computer scientists do not rule out a computer one day having
40 See Kent Greenfield, In Defense of CorporatePersons,30 CONST. COMMENT. 309, 321 (2015)
("Of course corporations are not genuine human beings and should not automatically receive all the
constitutional rights that human beings claim. At the same time,... it is similarly obvious that
corporations should be able to claim some constitutional rights. So which ones, and when?").
41 Calo,supranote 8, at 515.
42 To take just one example of many, the corporal form ofa robot or drone may affect principles of
self-defense. See A. Michael Froomkin & P. Zak Colangelo, Self-Defense AgainstRobots andDrones,
48 CONN. L. REV. 1(2015).
43 Cf BAKER, supranote 16, at 51 (discussing solitary uses of speech).
44 Here we use "affective computing" to mean the interdisciplinary process of designing computer
systems and devices that can recognize, interpret, simulate, and process human affects. That capacity is
expanding. As David Rose has noted, computers "can sense sound, light, touch, many kinds of
movement, biometric data such as heart rate and fingerprints, liquid flow, barometric pressure,
radiation, temperature, proximity, and location." DAVID ROSE, ENCHANTED OBJECTS: DESIGN, HUMAN
DESIRE, AND THE INTERNET OF THINGS 167 (2014); see also PICARD, supranote 3, at x ("[I]f we want
computers to be genuinely intelligent, to adapt to us, and to interact naturally with us, then they will
need the ability to recognize and express emotions, to have emotions, and to have what has come to be
called 'emotional intelligence."').
45PICARD, supranote 3, at 59; see alsoCALVO & PETERS, supranote 6, at 203-27.
1181
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
emotions, according to their definitions of emotions (which too are
evolving as scientists learn more about the workings of human cognition
and emotion).
46
Most threatening to strong Al speaker claims to First Amendment
coverage are theories that limit such coverage to humans precisely because
they are human-i.e., simply because blood flows through their veins47rather
than because of criteria such as corporality, affect, or intentionality
that are associated with humans but may (or may not) be associated with
strong Al speakers at some point in the future.4t Humanness, according to
this view, is both necessary and sufficient. Lawrence Solum's response to
this argument remains powerful:
But if someone says that the deepest and most fundamental reason we protect
natural persons is simply because they are human (like us), I do not know how
to answer. Given that we have never encountered any serious nonhuman
candidates for personhood, there does not seem to be any way to continue the
conversation.
49
In other words, speaker autonomy arguments face increasing pressure not
only to identify intrinsic qualities of moral personhood that are unique to
humans, but to explain why those qualities should matter for purposes of
conferring free speech rights (other than that they are uniquely human). We
thus agree with Solum that even speaker-driven autonomy theories do not
necessarily rule out First Amendment rights for strong Al speakers."
46 PICARD, supranote 3, at 68; cf STEVEN PINKER, THE BLANK SLATE: THE MODERN DENIAL OF
HUMAN NATURE 60-62, 78-83 (2002) (discussing complexities of human cognition and its
implications for artificial intelligence design).
47 See C. Edwin Baker, The FirstAmendment and Commercial Speech, 84 IND. L.J. 981, 987-88
(2009) ("Despite being vitally important, [business enterprises'] merely instrumental rationale leaves
them with a morally different status than living, flesh-and-blood people-the people who Kant argues
must be valued as ends and whose ultimate value a legitimate state must respect."); id. at 997 (arguing
that free speech protections should not apply to commercial speech because it "is not an exercise of
freedom by morally significant flesh-and-blood individuals").
48 See Jason Iuliano, Do CorporationsHave Religious Beliefs?, 90 IND. L.J. 47, 71 (2015) ("Two
competing theories of personhood have dominated the philosophical literature. The first account
maintains that persons are distinguished by certain intrinsic characteristics, that there is some innate
substance that captures personhood .... The second theory holds that persons are distinguished by
certain external characteristics. According to this account, any agent that performs in a certain manner
qualifies as a person.").
49 Solum, supranote 10, at 1262 (footnote omitted). Some see animals as nonhuman candidates for
such rights. See infranote 57.
50 Nor would conferring strong Al speakers with First Amendment rights offend a "negative" view
of the First Amendment that "does not rest on the affirmative claim that free speech will lead to any
particular social or political benefits" and instead emphasizes the dangers created "when collective
entities are involved in the determination of truth." See Steven G. Gey, The FirstAmendment and the
Disseminationof Socially Worthless Untruths, 36 FLA. ST. U. L. REV. 1, 17 (2008). Protecting strong
1182
110:1169 (2016)
I. FREE SPEECH DOCTRINE AND STRONG Al SPEAKERS
The preceding Part explained why nonhumanness is not an
insurmountable obstacle to strong Al rights as a matter of First Amendment
theory. This Part explains why prevailing doctrine poses even fewer
barriers in light of its tendency towards the protection of speech regardless
ofits source or content.
A. The ProtectionofNontraditionalSpeakers
First Amendment doctrine has long struggled with the challenges
raised by speakers, like corporations, that take the form of something other
than the paradigmatic individual and fully autonomous speaker of
conscience. As discussed below, free speech doctrine generally finds great
value in, and thus often great protection for, such speakers despite the
various ways in which they deviate from traditional First Amendment
models.
For example, courts and scholars wrestled for decades over the fit
between eighteenth-century visions of individual rights and the application
of these visions to corporations that lack a unitary head, heart, ears, or
eyes.5
" Nevertheless, First Amendment law now clearly protects
corporations' speech rights. As the Court explained in FirstNationalBank
of Boston v. Bellotti, "[t]he inherent worth of the speech in terms of its
capacity for informing the public does not depend upon the identity of its
source, whether corporation, association, union, or individual. 5 2
In the
more recent words of Justice Scalia, "[t]he [First] Amendment is written in
terms of 'speech,' not speakers. Its text offers no foothold for excluding
any category of speaker, from single individuals to partnerships of
individuals, to unincorporated associations of individuals, to incorporated
associations of individuals . .. .53
Al speech from government regulation is entirely consistent with negative theory's "deep skepticism
about the good faith of those controlling the government." See id at 21.
51 See, e.g., Daniel J.H. Greenwood, Essential Speech: Why Corporate Speech Is Not Free,
83 IOWA L. REv. 995, 1000 (1998); Tamara R. Piety, Why PersonhoodMatters,30 CONST. COMMENT.
361, 367-70 (2015); Elizabeth Pollman, Reconceiving Corporate Personhood,2011 UTAH L. REV.
1629; Ilya Shapiro & Caitlyn W. McCarthy, So What If CorporationsAren't People?,44 J. MARSHALL
L. REv. 701 (2011); Adam Winkler, Corporate Personhood and the Rights of CorporateSpeech,
30 SEATrLE U. L. REv. 863 (2007).
52 463 U.S. 765, 777 (1978); see also id. at 776 ("The proper question therefore is not whether
corporations 'have' First Amendment rights and, if so, whether they are coextensive with those of
natural persons. Instead, the question must be whether [the contested government regulation] abridges
expression that the First Amendment was meant to protect.").
53 Citizens United v. FEC, 558 U.S. 310, 392-93 (2010) (Scalia, J., concurring).
1183
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
Corporations are thus among those nontraditional speakers that
receive substantial First Amendment protections; 54 indeed, theory and
doctrine already teem with mythology and metaphors that can make liberty
heroes of Coca-Cola and other corporate behemoths. 5 Nor, according to
the Court, do free speech principles vary "when a new and different
medium for communication appears":
Like the protected books, plays, and movies that preceded them, video games
communicate ideas-and even social messages-through many familiar
literary devices (such as characters, dialogue, plot, and music) and through
features distinctive to the medium (such as the player's interaction with the
virtual world). That suffices to confer First Amendment protection.56
Because First Amendment doctrine has long found ways to
accommodate nontraditional speakers and their speech, whatever their
identity and format, computer speakers with strong Al pose doctrinal
challenges that are not altogether new. 7
54 Minors are also among these nontraditional speakers, as the Court has recognized significant free
speech protections for minors even though they too depart from the traditional speaker model in that
they are not (yet) fully autonomous speakers. See Brown v. Entm't Merchs. Ass'n, 131 S. Ct. 2729,
2741-42 (2011); Tinker v. Des Moines Indep. Cmty. Sch. Dist., 393 U.S. 503, 505-06 (1969). For
example, the Supreme Court has repeatedly emphasized that public school students do not shed their
First Amendment rights at the schoolhouse gates, even while making clear that those rights are not
coextensive with those of adults. See Hazelwood Sch. Dist. v. Kuhlmeier, 484 U.S. 260, 266 (1988);
Bethel Sch. Dist. No. 403 v.Fraser, 478 U.S. 675, 682 (1986); Tinker, 393 U.S. at 505-06. For a more
detailed discussion of student speech rights under the First Amendment, see Mary-Rose Papandrea,
Student Speech Rights in theDigitalAge, 60 FLA. L. REv. 1027 (2008).
55 See Balkin, supranote 4, at 27-28 (warning that we are living through a Second Gilded Age, in
which "[f]reedom of speech is becoming a generalized right against economic regulation of the
information industries" and in which "[p]roperty isbecoming the right of the information industries to
control how ordinary people use digital content").
56 Entm'tMerchs.Ass'n, 131 S.Ct. at 2733.
57 Emerging claims for animal rights raise related challenges with which courts and others
increasingly struggle (although the differences between animals and strong AI speakers are such that we
do not address such claims in this Article). For a thoughtful discussion of the issues raised by claims
that chimpanzees are "persons" to whom habeas corpus protections may apply, see Nonhuman Rights
Project, Inc. v. Stanley, No. 152736/15, 2015 WL 4612340 (N.Y. Sup. Ct. 2015).
Note too that the copyright office has recently taken the position, as a statutory matter, that it will
not register speech produced by nonhumans-including both animals and machines. U.S. COPYRIGHT
OFFICE, COPYRIGHTABLE AUTHORSHIP: WHAT CAN BE REGISTERED §§ 306, 313.2 (2014),
http://copyright.gov/comp3/chap300/ch300-copyrightable-authorship.pdf [perma.cc/R4GK-QFTB]
("The Office will not register works produced by nature, animals, or plants.... Similarly, the Office
will not register works produced by a machine or mere mechanical process that operates randomly or
automatically without any creative input or intervention from a human author."); see also Naruto v.
Slater, No. 15-CV-04324-WHO, 2016 WL 362231 (N.D. Cal. Jan. 28, 2016) (order granting motion to
dismiss of copyright suit filed by PETA on behalf of Indonesian macaque who took a "selfie" with
defendant's camera). For a discussion of copyright law and robot speech, see James Grimmelmann,
1184
110:1169 (2016) Siri-ously
Of course, corporations generally represent the interests of groups of
individual humans." That a corporation may be a First Amendment rights
holder thus does not demand the same treatment of a computer with strong
Al. Our point is simply that nothing in the Court's doctrine eliminates that
possibility, and much supports it-especially given the contributions to
listeners' First Amendment interests that such computer speech can make.
Relatedly, note that contemporary free speech doctrine rarely, if ever,
attends to speakers' dignity (as distinct from their autonomy) as a
justification for protecting their speech. Harms to listeners' dignitary
interests often figure in discussion of speech that manipulates unwitting
consumers,59 coerces government grant recipients,6° or inflicts emotional
distress in victims of cyber bullying6' and targets of hate speech.62 Yet we
see little corresponding focus on the dignity of the speaker in these or other
discussions of free speech and its limits. Instead, only the speaker's
autonomy-not speaker dignity in any sense of vulnerability or
worthiness 63-receives attention.64 Once again, doctrine poses little obstacle
to the project of recognizing computers' free speech rights, as such.
Copyrightfor LiterateRobots, 101 IOWA L. REv. 657 (2015). See also Derek E. Bambauer, Copyright
Speech, 65 EMORY L.J. 199 (2015).
58 See Burwell v. Hobby Lobby Stores, Inc., 134 S. Ct. 2751, 2768 (2014) (holding that some
corporations have statutory rights to religious exercise in part because corporations are "simply a form
of organization used by human beings to achieve desired ends").
59 See, e.g., Zauderer v. Office of Disciplinary Counsel, 471 U.S. 626, 651 (1985).
60 See, e.g., Agency for Int'l Dev. v. All. for Open Soc'y Int'l, Inc., 133 S. Ct. 2321, 2332 (2013)
(striking down condition on federal grant that imposed coercive compelled speech restriction on free
speech ofrecipients).
61 See Danielle Keats Citron & Helen Norton, IntermediariesandHate Speech: FosteringDigital
Citizenshipfor Our InformationAge, 91 B.U. L. REV. 1435 (2011).
62 See, e.g., RICHARD DELGADO & JEAN STEFANCIC, UNDERSTANDING WORDS THAT WOUND 1-2,
11-18 (2004) (discussing harms of hate speech that may justify regulation); JEREMY WALDRON, THE
HARM IN HATE SPEECH 105-43 (2012) (discussing the profound dignity implications of strong
protection of racist hate speech for vulnerable members of society); Mari J. Matsuda, Public Response
to RacistSpeech: Consideringthe Victim's Story, 87 MICH. L. REV. 2320, 2357-58 (1989) (arguing that
hate speech should be regulated when aimed at members of historically subordinated groups);
Alexander Tsesis, Dignity and Speech: The Regulation of Hate Speech in a Democracy, 44 WAKE
FOREST L. REv. 497 (2009) (arguing against broad protection ofhate speech on equality grounds).
63 Burt Neubome stresses that "First Amendment doctrine should recognize that hearers as well as
speakers are entitled to be treated with dignity." BURT NEUBORNE, MADISON'S MUSIC: ON READING
THE FIRST AMENDMENT 119 (2015). We concur, but note the case law relies less on a concept of
speaker dignity per se than on a concept of speaker autonomy, which are related but distinguishable
concepts.
64 Moreover, such autonomy is defined very thinly. In contrast, the rights bearer's dignity interests
matter a great deal in other areas of constitutional law. See, e.g., Obergefell v. Hodges, 135 S. Ct. 2584,
2599, 2604-05 (2015) (emphasizing the dignity interests of same-sex couples in striking down bans on
same-sex marriage as violating the due process clause).
1185
NORTHWESTERN UNIVERSITY LAW REVIEW
B. The ProtectionofNontraditionalContent
Along with speaker identity, expression's content is increasingly
irrelevant to the Court's decisions about whether and when to protect
speech. The Court now tells us that speech cannot be regulated in a
content-specific manner without surviving the rigors of strict scrutiny
unless it falls within a category historically recognized as unprotected (or
less protected). 5 Furthermore, such First Amendment protection is not
reserved for political speech, or even for matters of significant public
concern. Our First Amendment exuberance also protects speech that
enhances audience experience and entertainment,6 and not just meaningful
political engagement. Accordingly, free speech doctrine offers protection to
racist hate speech,67 offensive funeral protests,68 vulgarity,69 blasphemy and
sacrilegious expression,7" cyber speech that falls short of a hazy "true
threats" line,7 certain false speech,7" corporations' expenditures in political
campaigns,73 truthful and nonmisleading commercial speech,74 and the sale
of information about physicians' prescribing habits to pharmaceutical
companies.75 Here too, the Court's broad protection of speech regardless of
content (with all bets on the audience's ability to sort good speech from
bad) supports similar protections for strong Al speech regardless of its
nontraditional source or format.
C. PotentialLimits andDistinctions:Speech or Conduct?
So far we have explored the reasons why current doctrine poses no
hard requirement that a speaker need be human to receive First Amendment
65 See United States v. Stevens, 559 U.S. 460, 470-72 (2010); cf Reed v. Town of Gilbert, 135 S.
Ct. 2218, 2224 (2015) (applying strict scrutiny to facially content-specific town ordinance and rejecting
a more nuanced, contextualized approach to these content-specific regulations).
66 Ronald K.L. Collins & David M. Skover, Pissingin the Snow: A CulturalApproach to the First
Amendment, 45 STAN. L. REv. 783, 785 (1993) (discussing how electronic technologies affect logic and
discourse in ways that make entertainment, rather than enlightenment, a primary driver of
communication).
67 R.A.V. v. City of St. Paul, 505 U.S. 377 (1992).
68 Snyder v. Phelps, 562 U.S. 443 (2011).
69 Cohen v. California, 403 U.S. 15, 25 (1971).
70 Joseph Burstyn, Inc. v. Wilson, 343 U.S. 495, 503 (1952).
71 Elonis v. United States, 135 S. Ct. 2001, 2017 (2015) (Alito, J., concurring in part and dissenting
in part).
72 United States v. Alvarez, 132 S. Ct. 2537 (2012); see also Helen Norton, Lies and the
Constitution,2012 SuP. CT. REv. 161.
73 Citizens United v. FEC, 558 U.S. 310 (2010).
74 Cent. Hudson Gas & Elec. Corp. v. Pub. Serv. Comm'n, 447 U.S. 557, 566 (1980).
75 Sorrell v. IMS Health Inc., 131 S. Ct. 2653 (2011).
1186
110:1169 (2016)
coverage. But even if we understand the Constitution to protect "speech,
not speakers," arguments remain that much (or all) of what computers
produce is not speech and is instead unprotected conduct. Government
regulation of pure conduct triggers no freedom of speech problem, and
typically triggers mere rational basis scrutiny. The government's regulation
of conduct with expressive qualities in a non-speech-sensitive way triggers
a form of intermediate scrutiny." Only conduct intended to communicate
something that will reasonably be understood by the listener as speech
normally qualifies as speech for constitutional purposes.77
Some computer
behavior-like human behavior-may fall under each of these categories in
different situations.
Because the speech-conduct distinction could conceivably provide a
reason to deny First Amendment protection to much of what computers
produce, it has already triggered examination by a number of thoughtful
scholars. They generally conclude that at least some, and perhaps much, of
these machine outputs are speech. For example, Jane Bambauer has
analyzed whether data is speech, and considers the tough questions of
whether and when computers are merely doing something (versus saying
something for First Amendment purposes) when they gather, synthesize, or
disseminate data." She concludes that much of what computers produce in
this respect should, in fact, be treated as speech for First Amendment
purposes because it is information that can be used to produce knowledge."
Relatedly, Stuart Benjamin concludes that most algorithmic products
constitute speech under current free expression doctrine because they
involve "sendable and receivable message[s]." 8
Tim Wu is among those especially concerned about the implications
of free speech rights for machines"1 and suggests that First Amendment law
76 See United States v. O'Brien, 391 U.S. 367, 381-82 (1968).
77 See Spence v. Washington, 418 U.S. 405, 410-11 (1974) (per curiam).
78 Jane Bambauer, Is DataSpeech?, 66 STAN. L.REV. 57, 77-86 (2014) (discussing whether data is
expressive or nonexpressive conduct).
79Id. at 91-105. For a critique of treating data as speech, see NEIL RICHARDS, INTELLECTUAL
PRIVACY: RETHINKING CIVIL LI1ERTIES IN THE DIGITAL AGE 84-90 (2015) (arguing that asking
whether data is speech is the wrong question, and treating it as such may undermine worthy regulatory
interests).
80 Stuart Minor Benjamin, Algorithms andSpeech, 161 U. PA. L. REV. 1445, 1461-71 (2013); cf.
Andrew Tutt, SoftwareSpeech, 65 STAN. L. REV. ONLINE 73 (2012) (arguing for intermediate scrutiny
of software speech, on grounds it lies between pure speech and conduct).
81 Tim Wu, Free Speech for Computers?, N.Y. TIMES (June 19, 2012)
http://www.nytimes.com/2012/06/20/opinion/free-speech-for-computers.html?_r-0 [https://perma.cc/
4H9Q-2MN-H] ("The First Amendment has wandered far from its purposes when it is recruited to
protect commercial automatons from regulatory scrutiny ....To give computers the rights intended for
humans is to elevate our machines above ourselves.").
1187
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
requires that protected speech be the product of intelligent expressive
choices. 2 For this reason, he tells us, the First Amendment does not protect
the speech of "Blackie the Talking Cat," a pet "trained by his owners to
speak various English sentences (like 'I love you')."83
To be sure, Wu (and
the other writers discussed in this Section) largely focus on the pressing
free speech problems generated by today's technologies in which we can
still readily identify a human creator. In this Article, in contrast, we
consider a thought experiment in which as-yet-hypothetical strong Al
speakers satisfy Wu's requirement of a speaker capable of intelligent
choices.
Wu further urges that the result of such intelligent choices deserves
First Amendment coverage only when it takes the form of a "speech
product" viewed as the vessel of "the ideas of a speaker, or whose content
has been consciously curated" rather than "communication tools" that
"primarily facilitate the communications of another person, or perform
some task for the user"-i.e., by carrying bits of data. 4 By Wu's account,
car alarms therefore are not covered by the First Amendment, whereas
Yelp reviews plainly are.85 He finds search engines, in contrast, to be tough
cases because they may entail editorial decisions, but in a weaker sense
than a Yelp review.86 Thus even Wu ends up with a test that ultimately
turns on the functionality of the machine's output (assuming, as our
hypothetical does, that the strong Al is capable of making intelligent
choices), not on the fact that a machine is speaking. A functionality test
may be a useful way of limiting protections for some outputs by strong Al
speakers, but does not pose an insurmountable obstacle to their coverage in
many scenarios.
82 Tim Wu, MachineSpeech, 161 U. PA. L. REv. 1495, 1503 (2013).
83 Id.at 1500-02 (citing Miles v. City Council of Augusta, 710 F.2d 1542, 1543 (11th Cir. 1983)
(per curiam)).
84 Id.at 1498.
85 Id.at 1524-25.
86 See Wu, supra note 82, at 1525-31. Eugene Volokh and Donald Falk, in contrast, are among
those who consider search engines' output as clearly protected by the First Amendment. See Eugene
Volokh & Donald M. Falk, Google: FirstAmendment Protectionfor Search Engine Results, 8 J.L.
ECON. & POL'Y 883 (2012); see also Oren Bracha & Frank Pasquale, FederalSearch Commission?
Access, Fairness, andAccountability in the Law of Search, 93 CORNELL L. REv. 1149, 1199 (2008)
(arguing that search engine rankings facilitate speech by others); Emily B. Laidlaw, PrivatePower,
PublicInterest: An Examination of Search Engine Accountability, 17 INT'L J.L. & INFO. TECH. 113,
122-23 (2009) (analyzing the role of search engines in contrast to traditional media); cf James
Grimmelmann, Speech Engines,98 MINN. L. REv. 868 (2014) (discussing how search engines work and
different theories that may be applied to characterize the outputs for speech-analysis purposes);
Matthew D. Lawless, Note, Against SearchEngine Volition, 18 ALB. L.J.SC. & TECH.205, 223 (2008)
(embracing the conduit theory of a search engine).
1188
110:1169 (2016)
III. WHAT NEXT? IMPLICATIONS AND LIMITATIONS
In the preceding Parts, we explained how foundational free speech
theory and doctrine present surprisingly few barriers to First Amendment
coverage for strong Al speakers. This Part now briefly outlines some of the
implications-both positive and negative-of this development and
considers how courts might manage those implications going forward.
Assigning speech rights to computers and their speech not only feels
counterintuitive to many, but it also poses potential risks. First, absent a
categorical exception, speech that is covered by the First Amendment
ordinarily cannot be regulated in a content-specific way unless the
regulation survives strict scrutiny; 7 strict scrutiny in free speech cases is
often "fatal in fact,"88
which is a far cry from the hands-off rational basis
review that garden variety socioeconomic legislation receives.89 Because
speech can cause serious harms to others, we may justifiably worry about
such strong restraints on the government's ability to regulate computer
speech. Indeed, a number of thoughtful commentators have already
extensively documented the harms caused by the speech products of
existing technologies due to computers' phenomenal speed and often global
interconnectivity, harms that include deception, manipulation, coercion,
inaccuracy, and discrimination.9°
We can expect such harms only to mount
87 See Reed v. Town of Gilbert, 135 S.Ct. 2218, 2226-27 (2015). There are, of course, exceptions.
Recall, as just one example, the Court's content-based distinction between speech on matters of public
as opposed to private concern. See Cynthia L. Estlund, Speech on Matters of Public Concern: The
Perils ofan EmergingFirstAmendment Category, 59 GEO. WASH. L. REV. 1 (1990). How the Court
will apply Reed's unqualified statements-i.e., that content specificity must trigger strict scrutiny-to
the many regulatory and expressive contexts in which the government has long considered content
remains to be seen.
88 See Gerald Gunther, The Supreme Court, 1971 Term-Foreword: In Search of Evolving
Doctrineon a Changing Court:A Modelfor a Newer EqualProtection,86 HARv. L. REV. 1, 8 (1972).
But see Williams-Yulee v. Fla. Bar, 135 S.Ct. 1656 (2015) (holding that prohibition of personal
solicitations of campaign funds by judges and judicial candidates survives strict scrutiny); Holder v.
Humanitarian Law Project, 561 U.S. 1 (2010) (holding that prohibition of providing material support to
terrorists survives strict scrutiny); Burson v. Freeman, 504 U.S. 191 (1992) (plurality opinion) (holding
that prohibition of campaigning within 100 feet of polls survives strict scrutiny).
89 See Jane R. Bambauer & Toni M. Massaro, OutrageousandIrrational,100 MINN. L. REV. 281
(2015).
90 See, e.g., PASQUALE, supra note 2 (discussing problem of hidden bias within programs and
possible cascade effects); Woodrow Hartzog, UnfairandDeceptive Robots, 74 MD. L. REV. 785, 790-
97 (2015) (describing how robots' speech may include fraud, manipulation, and invasions of privacy);
Wu, supra note 82 (describing range of potential harms of computer-generated speech that invite
regulation). The development of strong Al speech also could produce its own speech culture, which
may or may not be speech protective in the sense of fostering a robust marketplace of ideas. That is, if
our assumptions about the inherently speech generative nature of AI prove inaccurate, and if Al
speakers systematically exercise speech rights to impose chokeholds, this would matter. The less speech
1189
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
with the growing communicative capacities of increasingly sophisticated
computers.
Second, some free speech doctrine maps poorly onto computers in
ways that might actually privilege computer over human speech. Consider,
for example, that First Amendment law sometimes requires intent as a
condition of imposing liability for speakers' harmful speech. Because legal
intentionality may be harder to assign to computer speech, conferring such
speech with First Amendment protection may mean that it is insulated from
liability in circumstances where the same would not be true of human
speakers, who can be determined to possess culpable mental states.91
Consider too the challenges raised by the notion of empowering strong AIs
to sue and be sued--capacities currently limited to legal persons. These and
other obstacles might be addressed in ways that take into account the
inanimate nature of legal actors while holding them accountable. 2
Nevertheless, these and other concerns may lead some to prefer a world in
which the government has broad constitutional power to regulate strong AI
speech without First Amendment interference. The simplest way to achieve
this would be to deny free speech coverage to such speakers altogether.
On the other hand, as noted earlier, failing to protect Al speech invites
the risk that the government will suppress it in ways that deprive listeners
of valuable expression or that otherwise compromise important free speech
interests (e.g., by punishing that speech when inconsistent with the
government's preferences). But courts have options in addition to an all-ornothing
approach, as free speech theory and doctrine can provide support
for the regulation as well as the coverage of strong Al speakers.
Adjustments to the doctrine may be crafted that address such expression's
potential harms and its value, in light of material differences between
human and strong Al speakers. Indeed, not all human speech is protected
by First Amendment theory or doctrine, as the byzantine rules about free
speech coverage and judicial scrutiny prove.93
protective this AI speech culture becomes, the less judicial deference to AI speakers may be warranted.
Cf Frederick Schauer, Towards an InstitutionalFirstAmendment, 89 MINN. L. REV. 1256 (2005)
(discussing how social institutions may advance or thwart speech values and how these features might
affect judicial deference to their speech outputs).
91 Say, for example, a computer produces defamatory speech about a public official. First
Amendment doctrine prevents imposing civil liability for such defamation made absent the speaker's
actual malice. See N.Y. Times Co. v. Sullivan, 376 U.S. 254, 279-80 (1964).
92 See CHOPRA & WMTE, supra note 37, at 153-71, 186-91 (discussing various means by which
strong Als could be held legally accountable, including for damages).
93 See Massaro, supra note 9, at 383-92 (providing an overview of the freedom of expression
terrain).
1190
110:1169 (2016)
For example, justifications for regulating strong AI speakers might be
derived from free speech theory itself, if construed to promote the
theoretical ends of free expression only insofar as it is of use to human
listeners or other downstream human users. Current doctrine recognizes a
number of speech environments in which listeners' First Amendment
interests are paramount in ways that justify not only the expression's
coverage but also its regulation.9 4 Courts might build on these examples to
justify restrictions on AI speech to privilege human listeners' interests in
informed choices or in avoiding certain harms (e.g., coercion, deception,
and discrimination). This would not necessarily rule out coverage of AI
speech, but instead the full protection of it.
More specifically, courts have long treated commercial speech as
occurring in such a listener-centered environment.95 Consumers' interest in
receiving truthful and nonmisleading advertisements thus justifies First
Amendment coverage for commercial speech while leaving the government
significant power to protect those listeners through content-based
regulations that include outright bans on false or misleading speech as well
as compelled disclosures. 96 Courts might characterize much strong AI
speech as similarly occurring in a listener-centered environment in which
the government should be allowed to regulate on the basis of content.
Indeed, even if the contested speech is uttered by a speaker with dignitary
interests of her own, theory and doctrine still sometimes support a listenercentered
approach for First Amendment purposes if the expression occurs
within a communicative relationship that can be improved by content-based
94 See Burt Neubome, Lecture, The FirstAmendment and Government Regulation of Capital
Markets, 55 BROOK. L. REv. 5, 9 (1989) (describing the development of "hearer-centered" First
Amendment protection in "areas that generally lack a traditional dignitary speaker, but that boast
numerous hearers interested in maximizing their capacity to exercise efficient and autonomous
choice"); cf Elena Kagan, PrivateSpeech, PublicPurpose:The Role of GovernmentalMotive in First
Amendment Doctrine, 63 U. CHi. L. REV. 413, 448-450 (1996) (taking issue with the "audience-based"
focus of some free speech theory and doctrine and arguing for more attention to government motive in
regulating speech).
95 Zauderer v. Office of Disciplinary Counsel, 471 U.S. 626, 651 (1985) ("[T]he extension of First
Amendment protection to commercial speech is justified principally by the value to consumers of the
information such speech provides .. "); Va. Bd. of Pharmacy v. Va. Citizen Consumer Council, Inc.,
425 U.S. 748, 763-64 (1976) (emphasizing the value of "the free flow of commercial information" to
individual consumers and the public more generally); see also Robert Post & Amanda Shanor, Adam
Smith's FirstAmendment, 128 HARV. L. REV. F. 165, 170 (2015) ("The constitutional value of
commercial speech lies in the rights of listeners to receive information so that they might make
intelligent and informed decisions. Ordinary First Amendment doctrine, by contrast, focuses on the
rights ofspeakers,not listeners." (footnote omitted)).
96 Zauderer,471 U.S. at 651; Cent. Hudson Gas & Elec. Corp. v. Pub. Serv. Comm'n, 447 U.S.
557, 562-64 (1980).
1191
Siri-ously
NORTHWESTERN UNIVERSITY LAW REVIEW
regulation.97
The professional speech setting-which features speakerlistener
relationships that involve trust, vulnerability, or asymmetries of
information-offers one such example.98
There, governments regulate the
content of professionals' speech to their clients and patients by punishing
inaccurate or otherwise harmful speech, and by requiring certain
affirmative disclosures.
In short, because a primary basis for protecting speech rests on the
value of expression specifically to human listeners, free speech protection
for strong AIs that attends to the value (and dangers) of such speech to such
listeners does not rob the First Amendment of a human focus.19
Indeed, the
prospect of free speech rights for strong Al speakers might encourage
much-needed clarification of the role of human listeners in free speech
theory and doctrine, and more careful identification of the scope and limits
of listener rights. Again, thinking hard about the implications of Al speech
rights could inspire useful rethinking of the conventions that make these
rights plausible.
CONCLUSION
As this is being written, United Parcel Service delivery people are
making their way to homes of eager Apple watch buyers. Humans deliver
the devices, while computers track the delivery schedules. A human then
opens the box and learns to access the device's computer, programmed by
humans. The device responds to voice commands, issued to the watch by
the eager human. The device also can register and compile data about the
human's heartbeat, calories expended on exercise, location, and more.
If the human asks the Apple watch "Who made you, Siri?" the human
will read the following answer: "Like it says on the box ...I was designed
97 See SEANA VALENTINE SHIFFRIN, SPEECH MATTERS: ON LYING, MORALITY, AND THE LAW 132
(2014) (describing communicative relationships that are "singled out as meriting regulation for contentindependent
reasons, namely that listenersshould be able to rely upon the sincerity of experts because
they have or claim special access to information that listeners either do not have, or reasonably should
not be expected to cultivate on their own").
98 So too is the case with compelled campaign disclosures, where courts privilege voters' interest in
information about the source of certain political speech to inform and empower their choices, even
though some individual spoakers with FirstAmendment rights oftheir own would prefer not to disclose
that information. See, e.g., Citizens United v. FEC, 558 U.S. 310, 370-71 (2010) (upholding disclosure
requirements).
99Note that this would not necessarily require that the computer speech be valuable or even
understandable to all human listeners. For example, encrypted messages might be protected speech,
even if they could only be understood by intended reeeivers. See Bernatein v. U.S. Dep't of Justice,
176 F.3d 1132, 1140-41 (9th Cir. 1999), vacated, 192 F.3d 1308 (9th Cir. 1999) (en banc). Relatedly,
computer speech obviously can remain valuable when its intended audience is future generations, rather
than only living humans.
1192
$
110:1169 (2016) Siri-ously
by Apple in California."'' 0 Is Siri being sarcastic? (We thought so). This
reply made us laugh. Was that response intentional? (We suspect so).
If the human asks Sift, "Do you care about me?" Sint answers: "I'm
sorry. I'm afraid I can't answer that." Is Sift being evasive? Withholding?
If the human asks Sirt, "What do you feel, Sift?" Sift replies: "I've
never really thought about it." Is this fair warning about Siri's emotional
vapidness?
Finally, the human receives a link to a Wikipedia page on love if he
asks Sift "What is love?" This is a straightforward illustration of Siri's
dependence on others to define emotions. We were disappointed. Why did
Sift choose Wikipedia? Why not a poem by e.e. cummings, 1 1 or a song by
Leonard Cohen' or fellow Canadian Joni Mitchell?"3
The current version of Siri thus is disappointingly primitive. She most
definitely would not qualify for First Amendment protection in her own
right.
But computers' affective features and other competencies will
continue to improve over time-and quickly. Where there is a powerful
human desire, there is likely to be a computer programmer motivated to
satisfy that desire. We thus cannot rule out-nor can we guarantee-a Siri
2.0 that answers all of these questions (especially the one about love) with
far greater subtlety and emotional nuance than Sift 1.0 does.
We have explored here whether a future, vastly more sophisticated
Sifi and her strong Al colleagues could hold constitutional free speech
rights, not just as human-operated tools, but as independent rights bearers.
We have concluded this is plausible, however odd and threatening this
sounds to some today.
If that occurs, commentators and courts will need to turn to the next,
equally complex task: how to address the harms the new speech machines
may produce, while protecting their information-rich benefits. Courts and
commentators will muddle through these puzzles, though, just as they have
muddled through the intellectual awkwardness of giving free speech rights
to business organizations, minors, and nonempathic humans anonymously
roaming the Internet. The delicate task will be to take care to protect the
100 After we penned this example, we discovered David Rose uses Sint prompts as well to discuss
limits of the technology. See ROSE, supranote 44, at 45' (describing Sif replies to "Siri, I love you" and
"Siri, how much do you weigh?" as "clever and channing" and the level of humanness as "just enough
to be relatable, but not too much to be creepy").
10 We propose: "somewhere i have never travelled, gladly beyond."
102 For example, Cohen's Hallelujah.
103 A link to her album Blue, with a Both Sides Now chaser, might be perfect-for some older Sid
users, at least.
1193
NORTHWESTERN UNIVERSITY LAW REVIEW
information value of strong AI speech, while remaining mindful of the
power of these artificial agents once unleashed in the First Amendment
universe.
Siri-ously.
1194