Article
Is it time for an offence
of ‘dishonest algorithmic
manipulation for
electoral gain’?
Dan Jerker B Svantesson and
William van Caenegem
Bond University, Australia
Abstract
Algorithms impact important aspects of our lives and of society. There are now strong concerns about algorithmic
manipulation, used by domestic actors or foreign powers, in attempts to influence the political process, including the
outcome of elections. There is no reason to think that Australia is immune or protected from such activities and we
ought to carefully consider how to tackle such threats – threats that go to the very heart of a democratic society. In this
article, we examine the potential introduction of a Commonwealth offence of ‘dishonest algorithmic manipulation for
electoral gain’.
Keywords
algorithmic manipulation, fake news, content regulation, Internet law, social media, election, propaganda, law reform
Today important aspects of our lives are affected, or
indeed controlled, by algorithms. Algorithms may determine
whether we obtain a loan or not, and how high our
insurance premiums will be. Many, indeed most, of that
type of algorithm-influenced decisions occur ‘behind the
scenes’ and we are rarely aware of whether a particular
decision is based on an algorithm and, if so, how it has
affected that decision.
In the online environment, algorithms are even more
dominant. They determine crucially important matters
such as what search results we are presented with,
which emails reach us (and which are relegated to the
trash bin) and what newsfeeds we are presented with.
We use the term ‘algorithm’ here in a broad and flexible
sense; in simple terms, an algorithm may be seen as
instructions or rules governing a process or calculation
typically part of some kind of automatic technological
problem solving or assessment. Thus, whether that process
results from or can be classified as ‘Artificial
Intelligence’ or the like, is neither here nor there. Also,
yet unforeseen or unknown developments of technology
are included under a broad and flexible umbrella.
At any rate, not least in the online context, we – the
ones affected by the algorithms – usually have no idea at
all how they operate and the extent to which any hidden
biases dictate what we experience online. Information
has always been subject to manipulation and presentation
to the advantage of hidden persuaders; however, what is
new in our era is the extent to which automated processes
gather, package and present information to us in a
manner that suggests neutral technological determination.
However, sleight of hand often hides behind seemingly
ungoverned processes.
The legal issues that all this gives rise to are now
gaining considerable attention. In this article we seek
to briefly, and admittedly superficially, introduce a legal
problem and outline the contours of a possible solution,
one that will need considerable discussion and debate
Corresponding author:
Dan Svantesson, Bond University, Robina 4229 Qld, Australia.
Email: dasvante@bond.edu.au
Alternative Law Journal
2017, Vol. 42(3) 184–189
! The Author(s) 2017
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1037969X17730192
http://journals.sagepub.com/altlj
over some time. Our focus is on a particular aspect of
how our online experience is guided by algorithms; we
focus on what may be termed dishonest algorithmic
manipulation for electoral gain. More specifically, having
discussed what dishonest algorithmic manipulation for
electoral gain is, we examine two alternatives for
addressing this concern. The first is to predominantly
focus on the content we are being fed online, while the
second option – our preference – is to predominantly
focus on the method by which content is selected for us.
This second option limits the risk to free speech.
Naturally what we examine is from a slightly broader
perspective – not something new to the law. There is a
long history of regulation of ‘truth in political advertis-
ing’.1
Furthermore, courts have generally held against
extending consumer protection law applying to sociopolitical
speech.2
In this instance, we do not advocate
taking any different course: as will be explained below,
we take a narrower and more cautious approach,
focused on technical means.
‘Fake news’ and dishonest algorithmic
manipulation for electoral gain
We are said to now be in the ‘post-truth’ era and there
has been much attention lately directed at so-called
FAKE news, election manipulation, the role of ‘big
data’3
and foreign powers interfering with the political
process in sovereign States. These are serious matters
and are treated as such, for example in the context of
national defence strategies; not least with 2017 being a
year of key elections in, eg, both Germany and France.
However, so far, limited attention has been directed at
what our criminal laws can do to help combat these
destructive forces.
Our views – including our political views – are subjected
to manipulation in a variety of ways. Normally it is
up to citizens to evaluate the political bias that attaches
to information from particular sources. However, where
such manipulation occurs in the invisible engine rooms of
the Internet, citizens may be particularly vulnerable. For
example, online it is difficult to assess why we see the
particular content we are served; transparency is lacking
and hidden forces might be at work which we do not
understand. This is significant, especially where people
get the majority – if not all – of their news online, and
where that environment gives the impression of neutrality
or impartiality. Many of us now use only online
content to make important political decisions as electors
– such as who to vote for in national elections.
Imagine that a marketing company employs algorithmic
manipulation to artificially cause the search results,
tweets and Facebook posts you see simply to communicate
content favouring a certain product. Such manipulative
and misleading commercial communication is already
regulated by law, for example, via the broadly worded
Article 18 of the Australian Consumer Law (ACL): ‘A
person must not, in trade or commerce, engage in conduct
that is misleading or deceptive or is likely to mislead
or deceive.’
However, where the same means are employed for
non-commercial communication the law says nothing,
precisely for the reason that provisions such as the
abovementioned Article 18 are focused on conduct
occurring in trade or commerce. Making all the search
results, tweets and Facebook posts you see communicate
content favouring a particular political view, person
or party undermines the integrity of the democratic process,
but the law is mute about it. This is simply astonishing,
since much more is ultimately at stake: why are
consumers protected, but voters not? Naturally raising
this question suggests that there is some identifiable
truth from which algorithmic manipulation diverts the
reader. Although it often will be difficult to ascertain
‘truth’ in this context, we believe that it nevertheless
may be possible to identify when such truth has been
corrupted. It is an exercise commonly engaged in by
courts – as for instance in cases of libel or defamation
(where the matter can often not be determined scientifically,
but nonetheless can be determined by the normal
evidentiary processes before a court of law).
The seriousness of the situation the world is facing
can escape no one who has read news reports such as
Carole Cadwalladr’s fascinating description, in The
Guardian on 26 February 2017,4
of what goes on
behind the scenes; the hidden forces at work to feed
online content to unsuspecting readers. In her article
Cadwalladr discusses in detail the links between
Trump’s election campaign, Brexit, social media, algorithms,
fake news, bots (automated software agents),
bio-psycho-social profiling, Artificial Intelligence and
data analytics companies such as Cambridge Analytica.
For example, in referring to her interview with Phil
Howard and Sam Woolley of the Oxford Internet
Institute’s Computational Propaganda Project,
Cadwalladr5
observes that:
[B]efore the US election, hundreds upon hundreds of
websites were set up to blast out just a few links, articles
that were all pro-Trump. ‘This is being done by people
who understand information structure, who are bulk
buying domain names and then using automation to
blast out a certain message. To make Trump look like
he’s a consensus.’
1
As documented for instance in Graeme Orr, The Law of Politics (Federation Press, 2010).
2
See, eg, Graeme Orr, ‘Government Communication and the Law’ in Sally Young (ed), Government Communication in Australia (Cambridge University
Press, 2007) 19.
3
For an interesting discussion of big data in the context of government powers refer, eg, to: Kate Galloway, ‘Big Data: A case study of disruption and
government power’ (2017) Alternative Law Journal 42(2) 89.
4
Carole Cadwalladr, ‘Robert Mercer: The big data billionaire waging war on mainstream media’, The Guardian, 26 February 2017 https://www.the-
guardian.com/politics/2017/feb/26/robert-mercer-breitbart-war-on-media-steve-bannon-donald-trump-nigel-farage.
5
Ibid.
Svantesson and Caenegem 185
[. . .]
You can take an existing trending topic, such as fake
news, and then weaponise it. You can turn it against
the very media that uncovered it. Viewed in a certain
light, fake news is a suicide bomb at the heart of our
information system. Strapped to the live body of us – the
mainstream media.
[. . .]
Many of the techniques were refined in Russia, he says,
and then exported everywhere else. ‘You have these
incredible propaganda tools developed in an authoritarian
regime moving into a free market economy with a
complete regulatory vacuum. What you get is a
firestorm.’
This is the world we enter every day, on our laptops and
our smartphones. It has become a battleground where
the ambitions of nation states and ideologues are being
fought – using us. We are the bounty: our social media
feeds; our conversations; our hearts and minds. Our
votes. Bots influence trending topics and trending
topics have a powerful effect on algorithms, Woolley
explains, on Twitter, on Google, on Facebook. Know
how to manipulate information structure and you can
manipulate reality.
We’re not quite in the alternative reality where the
actual news has become ‘FAKE news!!!’ But we’re
almost there. Out on Twitter, the new transnational
battleground for the future, someone I follow tweets a
quote by Marshall McLuhan, the great information theorist
of the 60s. ‘World War III will be a guerrilla information
war,’ it says. ‘With no divisions between military
and civilian participation.’
By that definition we’re already there.
At the time of writing, the news headlines are
focused on allegations of Russia seeking to influence
the election in several westerns democracies. And in
Germany, for example, steps are being taken towards
a law regulating, among other matters, fake news,
with a view to protecting voters and the democratic
process.6
There is no reason to think that Australia is beyond
threats such as those perceived in Europe and the US,
and an attack on the democratic process of a country is
doubtlessly an attack at the very core of that country.
Consequently, we argue that Australia ought to start
looking at, and acting on, these issues now, while taking
a very cautious approach. In any case, we focus on conduct
engaged in to not simply take the advantage in a
political debate, but to make an electoral gain. In other
words, conduct that is intended to skew electoral results
and also has that potential or actual effect. Simply influencing
opinions alone is not sufficient – it must be with a
view to influencing the outcome of an actual electoral
process that is extant and foreseen.
Option one: Focus predominantly on the
content
We realise we are in dangerous territory when regulating
non-commercial speech. So, if we want to deter certain
conduct in this sphere, then any offence we create should
be narrow, specific and clearly limited to activities in
relation to which there can be no doubts as to their
legitimacy. In other words, there is very limited room
for grey areas and rough edges.
Some countries already have in place criminal offences
that may be of relevance. Canadian law, for example,
contains the following criminal offence:
Every one who willfully publishes a statement, tale or
news that he knows is false and that causes or is likely
to cause injury or mischief to a public interest is guilty of
an indictable offence and liable to imprisonment for a
term not exceeding two years.7
While the language used suggests that attention is directed
towards the relevant activity (i.e. the wilful publishing
of the content), this offence is clearly
predominantly aimed at addressing the content. No
assessment is necessary in relation to the activity as
such, as the legality of the activity is dependent only on
the content published; provided of course the mens rea
(the mental element, put simply, the intention) is
established.
The difficulty of applying content-focused law is well
known from areas such as hate speech, injurious falsehood
and defamation law. The recent debate about section
18C of the Racial Discrimination Act 1975 (Cth) is a
case in point. But the problems are particularly well illustrated
by a case that came before the Supreme Court of
Canada. In R v Zundel,8
Section 181 of their Criminal
Code was applied in the context of a publisher which
published material questioning the historically accepted
account of the Holocaust. The Supreme Court of
Canada had to examine the constitutionality of the mentioned
Section, and the majority concluded that:
Section 181 of the Code infringes the guarantee of freedom
of expression. Section 2(b) of the Charter protects
the right of a minority to express its view, however
unpopular it may be. All communications which convey
or attempt to convey meaning are protected by s 2(b),
unless the physical form by which the communication is
made (for example, a violent act) excludes protection.
The content of the communication is irrelevant. The
purpose of the guarantee is to permit free expression
to the end of promoting truth, political or social participation,
and self-fulfilment. That purpose extends to the
protection of minority beliefs which the majority regards
as wrong or false. Section 181, which may subject a
6
Amar Toor, ‘Germany grapples with fake news ahead of elections’, The Verge, 19 January 2017 https://www.theverge.com/2017/1/19/14314680/ger
many-fake-news-facebook-russia-election-merkel.
7
Criminal Code (R.S.C., 1985, c. C-46), Section 181.
8
R v Zundel [1992] 2 S.C.R. 731.
186 Alternative Law Journal 42(3)
person to criminal conviction and potential imprisonment
because of words he published, has undeniably
the effect of restricting freedom of expression and,
therefore, imposes a limit on s 2(b).
Given the broad, purposive interpretation of the freedom
of expression guaranteed by s 2(b), those who
deliberately publish falsehoods are not, for that reason
alone, precluded from claiming the benefit of the constitutional
guarantees of free speech. Before a person is
denied the protection of s 2(b), it must be certain that
there can be no justification for offering protection. The
criterion of falsity falls short of this certainty, given that
false statements can sometimes have value and given the
difficulty of conclusively determining total falsity.9
This clearly highlights the risks associated with offences
predominantly focused on the content in question. What
we propose is different. We are wary of provisions such
as this as they may be vulnerable to misuse. Instead of
focusing on the acceptability of the content, as is done for
the abovementioned Canadian offence, we turn our
attention to the acceptability of the method used to
create or facilitate that content.
Option two: Focus predominantly on the
method
Focusing predominantly on the method is a first way to
limit the scope of any offence, while still making it effective
in deterring exactly the secretive and invisible
manipulation we are concerned about.
We propose the creation of a new offence, rather
than the introduction of a new tort. So how should
we express such an offence? There are, of course,
several possibilities, but to keep it simple, and to
draw upon the wordings used in other settings in
the Criminal Code Act 1995 (Cth), we propose the following:
‘A person commits an offence if the person does
anything with the intention of obtaining electoral gain from
dishonest algorithmic manipulation.’ Terms used need to
be carefully defined in the law itself. Thus ‘electoral
gain’ would here include undermining, or favouring, a
political view, organisation or individual with a view to
gaining an electoral advantage. Clear evidence of intent
would be an essential element of the offence, and the
intention requirement is one important filter restricting
the application of the offence. For example, it is clear
that major online platforms, such as Facebook and
Google, need to increase, and are in fact increasing,
their efforts to combat fake news. However, the
offence is not aimed at the activities of such online
platforms as they are not commonly argued to
intend to undermine, or favour, any specific political
view, organisation or individual, with a view to influencing
the electoral process.
Defining ‘dishonest algorithmic manipulation’ goes to
the heart of the proposed offence and will no doubt be
both the most important, and most difficult, task. It
would of course be premature to seek to provide a definite
definition of this key term in this short article.
Rather, the delineation of what amounts to ‘dishonest
algorithmic manipulation’ has some core elements:
there must be a dishonest intent; the manipulation
must involve algorithms and there must be an element
of manipulation in the sense of deliberate and hidden
determination of the settings of an algorithm.
The precise delineation of the offence must be
decided after extensive consultation, and there will be
many stakeholders. Here we make some preliminary
observations that may be used to guide any such consultation
process. We envisage that there are several fundamentally
different drafting options for delineating what
amounts to ‘dishonest algorithmic manipulation’.
First, it is possible to rely on a positive but nonexhaustive
list outlining (examples of) what amounts to
‘algorithmic manipulation’, rather than having a broad and
abstract definition. The means employed must concern
the manipulation and deployment of algorithms, properly
defined. The element of dishonest intent to influence
voters (or the electoral process) should be separately
established. It is not difficult to identify practices that
would satisfy both elements. One example is where
bots are ‘designed [and used] for the express purpose
of disrupting free expression – for example, by hijacking
Twitter hashtags and flooding them with contrary or
irrelevant messages’.10
There are several bot tactics –
such as hashtag spamming, making artificial trends,
smear campaigns, death-threat campaigns and political
propaganda – that may be used to infringe on the free
expression rights of others in a different way:
Hashtag spamming—the practice of affixing a specific
hashtag to irrelevant content—renders the hashtag unusable.
Artificial trends can bury real trends, thus keeping
them off the public and media’s radar. Smear campaigns
and death threats can both intimidate vocal political
opponents and dissuade would-be speakers.11
All such bot tactics may fall within the definition of ‘dishonest
algorithmic manipulation’ and be included on a
positive list of offensive conduct. They would also satisfy
the separate requirement that the manipulation is hidden
and deliberate.
A second option is to start with the assumption that
any algorithmic manipulation is caught, and then couple
that starting point with a negative list outlining practises
that are not dishonest. Constructing such a list would
take us into more treacherous terrain as there are
numerous forms of algorithmic manipulation we would
not want to see affected by the offence, such as standard
9
R v Zundel [1992] 2 S.C.R. 731, Per La Forest, L’Heureux-Dube´, Sopinka and McLachlin JJ.
10
Nathalie Mare´chal, ‘When Bots Tweet: Toward a Normative Framework for Bots on Social Networking Sites’ (2016) 10 International Journal of
Communication, 5022–5031 http://ijoc.org/index.php/ijoc/article/view/6180/1811.
11
Ibid.
Svantesson and Caenegem 187
forms of search engine optimisation. Further, we
would need to anticipate acceptable future forms of algorithmic
manipulation. Thus, we see this option as less
realistic.
We should also attempt to define the element of dishonest
intent. This should be done by reference to the
effects of such conduct – that being the influencing of
electoral choice. Again we could use a non-exhaustive
list of incidences of dishonest intent. For example, one
possible incidence would be where algorithms are
manipulated with the intent that the result unduly and
invisibly prioritises one political message, while drowning
out competing views. Those behind the manipulation
must favour a particular electoral outcome, which motivates
their actions.
Finally, one can of course imagine hybrid structures,
for example combining the first and third option so that
we get a detailed definition backed up with a list of
examples; a method used with success in other settings
such as Council Directive 93/13/EEC of 5 April 1993 on
unfair terms in consumer contracts (mirrored in
Australia’s ACL part 2–3).
Obstacles and challenges
The above has already emphasised the considerable
challenges associated with properly delineating the proposed
offence so as to avoid the risk of interference
with legitimate political speech. However, there are
other obstacles and challenges that also need to be
confronted.
As an example, we are not naı¨ve as to the obvious
difficulties involved in enforcing the crime outlined above.
Where the dishonest algorithmic manipulation originates
abroad, effective enforcement may require assistance by
the very State that is behind the crime in the first place. If
it is indeed the case that Russia has been engaging particularly
actively in dishonest algorithmic manipulation,
the enforcement of the proposed offence would often
require cooperation from Russia; cooperation that of
course is unlikely to be forthcoming. Having said that,
recent arrests of Russian hackers holidaying outside
Russia highlight that the prospect of enforcement nevertheless
should not be underestimated.12
At any rate, computer forensics is a complex area and
even where the algorithmic manipulation is domestic,
proving its existence and identifying the responsible parties
will not be easy. But Australia’s capabilities are steadily
increasing, and it may reasonably be assumed that, if an
Australian election was subjected to serious algorithmic
manipulation, addressing the situation would be made a
priority.
We argue that these obstacles and challenges should
not prevent us trying to address dishonest algorithmic
manipulation. After all, law serves different functions, one
being to communicate societal standards. And if the law
is punchy enough, it will have a general deterrent effect.
The goal of clearly articulating that dishonest algorithmic
manipulation for electoral gain is unacceptable would be
achieved the very day such activities are made a criminal
offence.
Concluding remarks
The offence of ‘dishonest algorithmic manipulation for
electoral gain’ will not alone solve the problems of
‘FAKE news’, election manipulation and foreign
powers interfering with the electoral process in sovereign
States, associated with our ‘post-truth’ era.
However, we suggest that such an offence may serve
as one useful tool in the toolbox required to address
this complex matter.
Any restriction on political activity is a sensitive
matter – it goes to the heart of a democratic society.
And the last thing we want is to create a tool that may be
used to repress healthy political debate. However, the
question is whether we can retain a healthy political
debate in an era of dishonest algorithmic manipulation.
Given recent developments, we fear that we cannot;
although recognising the difficulty of what we are proposing,
we do not believe the option of doing nothing is
preferable. We cannot do nothing.
As Professor Lawrence Lessig famously noted, now
almost 20 years ago: ‘Left to itself, cyberspace will
become a perfect tool of control. . ..’13
This seems a particularly
apt observation when placed in the context of
dishonest algorithmic manipulation for electoral gain. We
think it is possible to fashion a new offence that specifically
and narrowly applies to hidden dishonest manipulations
of the Internet that confer unfair political gain in
elections, while not deterring free speech and debate.
However, we hasten to acknowledge that developing
such an offence requires great care and extensive consultation.
Our modest aim with this article is merely to
start the conversation.
Author note
This article builds, and expands, on their previously published
work, Dan Jerker B Svantesson and William van Caenegem,
‘Faking it: We should make manipulating algorithms for political
purposes a crime’, The Conversation, 10 March 2017 http://the-
conversation.com/faking-it-we-should-make-manipulating-algo-
rithms-for-political-purposes-a-crime-73970.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
article.
12
‘Russian arrested in Spain ‘‘over mass hacking’’’, BBC News Technology, 10 April 2017 http://www.bbc.com/news/technology-39553250.
13
Lawrence Lessig, Code and Other Laws of Cyberspace (Basic Books, 1999) 5–6.
188 Alternative Law Journal 42(3)
Funding
The author(s) received no financial support for the research,
authorship, and/or publication of this article.
Dan Jerker B Svantesson is Professor and CoDirector,
Centre for Commercial Law, Faculty of Law,
Bond University (Australia) and researcher, Swedish
Law & Informatics Research Institute, Stockholm
University (Sweden).
William van Caenegem is Professor, Centre for
Commercial Law, Faculty of Law, Bond University
(Australia).
Svantesson and Caenegem 189