JT Team A
Does the Influx of Synthetic Media Pose an
Invincible Threat to the Integrity of Information on
the Internet?
A Research paper on Internet Trends
LGS JT Team A
JT Team A
THE PROBLEM AS IT CURRENTLY EXISTS
Synthetic Media, defined as artificially generated or altered media[1][2], while having its
advantages that might even make it a technological breakthrough, gives its abusers the
ability to produce media that is by nature deceptive, incorporating within it, events that
never happened, sayings that were never said, and data that was never recorded, spreading
both misinformation and disinformation[3][4]. The issue at hand is that if the trend in the
development rate of such technology continues [5], the capability of outlets of such media
will undergo rapid increase[6], reaching an extent where relying on the end users’ ability to
differentiate between true and false information may yield devastating effects, making it
near impossible to differentiate between these two sorts of information for the human brain,
hence bringing under question, the integrity of all digital information.
INTERNATIONAL VIEWS ON THE SUBJECT
A number of different views concerning the issue exist, all of them reflective of a person’s
larger views about false information. All the aspects of these views are discrete, i.e., they
are either present or absent, and since these aspects are largely correlated, it can be
concluded that one may either view the issue through the perspective of a person that has
the ability to understand the threats posed by misinformation, discern the credibility of
information, and keep themselves aware about the issue[7], or a person that does not possess
these abilities, to whom we can refer to as “aware” and “unaware” respectively. A third
perspective might be that of a person who misuses synthetic media[8].
When looking at the issue globally, a distinction between the people of developed and
developing countries is apparent, where a larger majority of the former can be considered
“aware” and actively combating the issue than the latter[9][10]. As an example, the countries in
the European Union are involved in efforts to combat the issue[11], but there is little or no
effort in this regard in South Asian or African Countries. Furthermore, a large proportion of
people from rapidly developing countries like India are likely to abuse synthetic media than
lesss economically well-off countries.
Synthetic Media has also been used internationally to achieve political aims by spreading
disinformation aimed at creating confusion [15], an example of which is the deepfake showing
the Ukrainian President ordering a surrender[16][17], or for propaganda, like the video
published by Ukraine showing Paris being bombed[18]. It is also used within countries to
defame political opponents, for which the allegedly fake video of a Kerala MP[14] serves as
an appropriate example.
THREATS POSED BY THE ABUSE OF SYNTHETIC
MEDIA
JT Team A
Videos, images, audio and text remain the most popular types of synthetic media, all of
which collectively pose several threats[19][20] such as political and legal ones. Synthetic
videos and audio can together form false proof against innocent individuals that is difficult
to refute. Similarly, synthetic images and text aid the creation of forged documents creating
yet another legal issue. Low quality synthetic media which is easy to produce is used
excessively by scammers[21] and cybercriminals on large scales as well as for identity theft.
Spreading false information about world affairs, finances, health, etc. leads to people
making poorly informed decisions. Since the rate at which such information spreads
increases exponentially, it constitutes a growing problem. Not only does false information
directly affect people, it also puts the authenticity of credible information under question[22]
making people less likely to believe information that might be of use to them. It is one of the
most powerful tools used in mass propaganda and is used both to defame certain
individuals, or spread lies that inculcate harmful ideologies, as well as to achieve other
malicious aims[23][24][25][26][27].
One way in which Synthetic Media is excessively abused is by spreading
disinformation to achieve political aims, both by Politicians themselves, as well as
proponents of political ideologies who are not political figures themselves. A large
number of times, synthetic media is used to defame individuals for political gain, and
example of such use cases would be the allegedly fake audio of former Chief
Justice of Pakistan circulated on social media[13], or the allegedly fake video of a
Kerala MP[14], both of which, while not convincingly proved fake themselves, show
how someone might abuse synthetic media for political gain. Similarly synthetic
media can also be used to achieve political aims by spreading false information
aimed at creating confusion among the masses[15], such as the deepfake showing
Ukrainian President ordering a surrender[16][17], or for propaganda, like the video
published by Ukraine showing Paris being bombed[18]. It must be noted that such
tactics are more popular in less developed countries where they are more likely to
deceive the people. While the above-mentioned examples were identified as fake,
what is concerning is that there were little or no repercussions, and they were able to
create substantial disruption before being debunked.
HOW THESE THREATS COULD EVOLVE
One threat posed by the evolution of Synthetic Media is the narrowing “uncanny valley”
making it near impossible for humans to differentiate between credible and false
information, rendering all methods that combat the issue by relying on people’s ability to
discern credibility obsolete. The scale on which such Synthetic Media is being produced is
also predicted to grow, casting further doubt on the integrity of all digital information, since
a larger portion of it will be plagued with lies. The fact that many instances where synthetic
media was misused by substantially popular bodies went unopposed means we might see
wider adoption of such tactics in the future. Therefore, the main threats posed by this
evolution are that synthetic media will resemble reality to an extent that makes it
unrecognizable[28], and that it will be produced in exponentially larger quantity making it
much harder to tackle all of it one by one[29].
JT Team A
ANALYZING CURRENT METHODS TO CURB THE
ISSUE AND THEIR SHORTCOMINGS
While a number of methods exist to curb the issue, the choice of spreading awareness[30][31]
[32]
, out of all the methods, remains the most popular. While it has its advantages like greater
outreach, it remains one of a category of methods that rely solely on the recipients’ ability to
put the received information to use, and the assumption that the recipient will be willing to
take in the information, which will not be the case for someone who is an avid believer of
misinformation. Other efforts to resolve the issue are legislative[33][34], however as of now
there is little or no legislation regarding various aspects of the problem, and much of the
legislation that does exist is either ineffective or impractical[35]. A third method is to use the
same technology used to generate high quality synthetic media, i.e., Artificial Intelligence
and Machine Learning, to scan and detect synthetic media being used to spread false
information[36][37]. While the fact that not all digital information can be scanned this way, and
that any attempt to do so will constitute a serious privacy violation serves as a significant
obstacle for further progress, it is worth noting that when used, this method remains highly
successful and efficient in detecting such media, especially with the advent of new
technologies like Deep Learning, Neural Networks etc.
ELABORATING ON DISCRIMINATOR MODELS (AI),
BLOCKCHAIN AND OTHER TECHNOLOGIES AS A
SOLUTION
As synthetic media evolves, the strategies employed to counter its abuse must adapt
adequately38]. This includes identifying alternatives to methods that depend on people’s
ability to identify disinformation, since doing so will become increasingly difficult. Further
legislation regarding the issue is also required, but any such legislation must be complete
and practical, for which the required infrastructure and technology needs to be built first[39]
[40][41][42]
. Existing methods to detect such synthetic media using AI need to be improved with
[43][44][45]
time , and referring to official and recognized sources for information and not third
parties must be stressed on[46]. Emerging technologies like Blockchain can be used to trace
the source, and determine the originality and credibility of media[47][48]. Moreover, awareness
as a method must not be discarded but instead stepped up, with modified aims, encouraging
people to voluntarily check and report malicious synthetic media[49], eliminating the need for
software robots to scan digital information (which is a violation of digital privacy unless
severely restricted), and to actively participate in efforts to combat the malicious use of
synthetic media, which requires widespread scientific literacy. Furthermore, not only do we
need to encourage and promote the use of technology to detect Synthetic Media in Judicial,
Economic and other systems, especially instead of humans, we also need to modify and
design technology in a way that shortens the time required for its use and achieves other
goals that collectively aid such implementations of said technology. It must be noted that the
reason AI and Blockchain are suggested as potential effective solutions is that they together
JT Team A
form a two-fold approach, where the former allows the recipient to discern credibility, while
the later allows the source to preserve originality, together forming a complete solution to
the problem.
One way to discern deepfakes and other elements of Synthetic Media using AI is through
increased pace of development of Discriminator Models through Generative Adversarial
Networks or Classifier Networks trained to detect specific categories of Synthetic Media.
The benefit of using AI models based on Convolutional Neural Networks for this task is
twofold; Firstly, the capacity of present-day technology to process millions of data examples
easily essentially removes any limitation in training these models which only get more
accurate at their task as they train on more data. Secondly, these models are able to ‘see’
more of the image than we as humans can with the bare eye since they process numbers;
thus, they are able to spot patterns in the data and learn a highly complex nonlinear function
to discriminate between real and fake examples. The development of Deep Learning based
Classifiers to detect spam accounts/emails on social media websites such as Twitter/Gmail
is one such example of Classifier Networks being put to a similar use which shows that
development of such models has immense potential. Generative Adversarial Networks in
the past have been used to create Generator Models which create high quality fake images
of humans [51][52] which are so realistic that discerning them from real images through the
naked eye is impossible. However, Discriminator Models which are used in improving the
performance of these Generator Models (which are being developed extensively due to their
immense applications) are usually discarded after training which is a waste of a very
promising application of AI which can be put to use for discerning Synthetic Media. Greater
focus on discriminator models is needed to put AI to use in combatting Synthetic Media.
A PROPOSED STRATEGY FOR DEVELOPING
DISCRIMINATOR MODELS IN THE FUTURE
Creation of an AI based model architecture involves many aspects, each of which need to be
dealt with precision for cutting edge performance which is necessary for a successful
discriminator model in our case. These aspects include but are not limited to data
augmentation methods, test train splits, ability of loss function to converge quickly, transfer
learning for quick convergence and finding optimal parameters using popular networks, use
of various regularization techniques to prevent overfitting, data manipulation for feature
extraction, mini-batch sizes etc. Most of these aspects are dependent on experimentation and
seeing training and test error as the model is trained while some such as choosing an
appropriate loss function require in-depth theoretical research into the structure of the input
data, which is different for every AI model.
Gaussian blur, Exponential blur, and Rayleigh blur are some example of data manipulation
in the testing phase to make it even more similar to the data that it will later see when it will
make its predictions. When considering data augmentation techniques, it is pertinent to note
that ‘original’ data is precious and not so easily available, hence techniques like shearing,
stretching, rotation, and translation are used to expand the image dataset for example since
JT Team A
the transformed images are also part of the same dataset. Such techniques should obviously
be used but the use of GAN’s and autoencoders, which are the cause behind the problem
itself, can also be employed to create more deepfakes which we can train our discriminator
on. This can be effective because it takes advantage of the extensive research into GAN’s in
recent years for countering the problems it creates.
If a discriminator model trained through GAN’s is employed, then it is important to note the
current loss function for discriminator functions should be further researched because the
loss function has been optimized for the generator, which is a subtle difference considering
that the discriminator’s performance also becomes better as a consequence. D(x) is the
output of the discriminator function which is near 0 for fake examples and near 1 for real
examples. As such, you would want D(x) to return 1 for real and 0 for fake examples in the
ideal scenario. Differentiating the loss function with respect to our model parameters allows
us to update their values based on the error we get every iteration.
Greater research into deepfakes [53] has identified that faces in many deepfake videos do not
blink their eyes or blink in an anomalous manner. Binary classification programs which can
detect significant changes in the eye blinking pattern (by analyzing multiple frames in
succession) which is a spontaneous action that depends on the overall physical condition of
a human can provide a unique way of classifying deepfakes effectively. This is one such
example of in-depth research yielding positive results to deal with the problem of Synthetic
Media.
ANALYZING THE SOURCES USED
The information referred to in a report collectively shapes its outlook. Keeping this in mind,
we tried to cover all aspects of the topic by referring to all kinds of sources, while also
maintaining an appropriate balance between them. For instance, we referred to not only
research papers but also articles and blogs, allowing me to refer to facts as well as analyze
opinions from people with different perspectives forming a complete analysis. Similarly,
while referring to information from experts, like MIT on the usage of AI, we also looked
into content written by general writers. Moreover, wherever a difference in opinion
appeared, we referred to multiple sources. Not only did we look into content directly
addressing my concerns, we also looked into sources that, while not directly discussing the
topic at hand, contained information we could use to deduce several conclusions, for
example we referred to roadmaps of companies producing synthetic media related products,
to identify how synthetic media might evolve in the future. In all cases, the most appropriate
source of information was given precedence, such as law experts on the legal implications,
Engineers on maters regarding technology, etc.
JT Team A
CONCLUSION
To put the content of this research thesis into perspective, it is evident that while threats
posed by the abuse of Synthetic Media are great, and will definitely proliferate in the future,
we are on the path of exploring potential solutions with adequate effectiveness, and by
ensuring that appropriate measures, both the above-mentioned ones and others, are taken at
adequate intervals in the future, this issue can be tackled to a large and extremely
satisfactory extent. Hence, based on the hope that these measures will be taken, synthetic
media does not pose a substantial threat to the integrity of digital information, as of now,
and will not do so as long as it is kept in check for the years to come.
The above text, excluding subheadings and including references to the bibliography, amounts to exactly 2521 words.
BIBLIOGRAPHY
[1] W3C definition of Synthetic Media - https://www.w3.org/community/synthetic-media/wiki/Main_Page
[2] The World Bank on Synthetic Media and its implications - https://olc.worldbank.org/system/files/4.%20Synthetic
%20Media%20and%20Content.pdf
[3] Buzzfeed article about an information crisis - https://www.buzzfeednews.com/article/charliewarzel/the-terrifying-
future-of-fake-news#.taE9n0qax
[4] Video showing what Synthetic media can do - https://www.youtube.com/watch?v=AmUC4m6w1wo
[5] how products based on synthetic media might evolve - https://www.synthesia.io/post/the-future-of-synthetic-media
[6] WIRED article about the future of synthetic Media and AI - https://www.wired.com/story/future-of-artificial-
intelligence-2018/
[7] An Article Spreading Awareness - https://towardsdatascience.com/ai-generated-synthetic-media-aka-deepfakes-
7c021dea40e1
[8] https://www.sciencedaily.com/releases/2020/11/201124092134.htm
[9] Misinformation is rampart in developing countries -
https://www.pewresearch.org/fact-tank/2019/05/13/misinformation-and-fears-about-its-impact-are-pervasive-in-11-
emerging-economies/
[10] An Article from and American Think Tank - https://www.cfr.org/blog/misinformation-threat-democracy-
developing-world
[11] EU efforts to tackle the issue at hand -
https://www.europarl.europa.eu/RegData/etudes/STUD/2021/690039/EPRS_STU(2021)690039_EN.pdf
[12] https://www.tandfonline.com/doi/full/10.1080/19331681.2019.1686676
[13] https://arynews.tv/forensic-audit-proves-saqib-nisars-audio-leak-edited/
[14] https://www.tribuneindia.com/news/haryana/fir-lodged-in-gurugram-mp-fake-video-clip-case-379946
[15] https://venturebeat.com/2020/09/09/synthetic-media-targets-politics-ahead-of-u-s-presidential-election/
[16] https://www.npr.org/2022/03/16/1087062648/deepfake-video-zelenskyy-experts-war-manipulation-ukraine-russia
[17] https://www.bbc.com/news/technology-60780142
[18] https://www.forbes.com/sites/alexandralevine/2022/03/17/ukraines-promotion-of-fake-paris-bombing-video-
highlights-risks-of-misinformation/?sh=594777101e9b
JT Team A
[19] https://www.reuters.com/technology/deepfake-anyone-ai-synthetic-media-tech-enters-perilous-phase-2021-12-13/?
utm_source=reddit.com
[20] Threats of Synthetic Media - https://www.cfr.org/report/deep-fake-disinformation-steroids
[21] https://www.businessinsider.com/scammers-use-video-game-footage-to-spread-misinformation-for-likes-2022-3
[22] How Synthetic Media results in uncertainty - https://journals.sagepub.com/doi/full/10.1177/2056305120903408
[23] Synthetic Media as a threat to democracy - https://ppforum.ca/event/digital-democracy-101-synthetic-media-
deepfakes/
[24] https://www.dal.ca/news/2019/03/13/can-you-believe-it--how-synthetic-media-poses-a-threat-to-democr.html
[25] Effects of Synthetic Media on Finance - https://carnegieendowment.org/2020/07/08/deepfakes-and-synthetic-
media-in-financial-system-assessing-threat-scenarios-pub-82237
[26] https://academicworks.cuny.edu/cgi/viewcontent.cgi?article=1123&context=bb_etds
[27] Synthetic Media as a threat to National Security - https://www.realinstitutoelcano.org/en/analyses/the-
weaponisation-of-synthetic-media-what-threat-does-this-pose-to-national-security/
[28] https://research.nccgroup.com/2022/01/12/impersonating-gamers-with-gpt-2/
[29] https://www.ned.org/the-big-question-how-will-deepfakes-and-emerging-technology-transform-disinformation/
[30] An example of awareness - https://www.cjr.org/tow_center/reporting-machine-reality-deepfakes-diakopoulos-
journalism.php
[31] Awareness on the topic by Reuters - https://www.reuters.com/article/rpb-hazeldeepfakesblog-idINKBN1YY14C
[32] https://lab.witness.org/projects/synthetic-media-and-deep-fakes/
[33] Legal Issues of deepfakes - https://www.internetjustsociety.org/legal-issues-of-deepfakes
[34] https://digitalcommons.law.uga.edu/cgi/viewcontent.cgi?article=1321&context=fac_pm
[35] How existing legislation is impractical - https://techcrunch.com/2019/06/13/deepfakes-accountability-act-would-
impose-unenforceable-rules-but-its-a-start/
[36] Research by MIT on how AI can detect deepfakes - https://www.media.mit.edu/projects/detect-fakes/overview/
[37] https://link.springer.com/chapter/10.1007/978-3-030-88040-8_7
[38] https://knowablemagazine.org/article/technology/2020/synthetic-media-real-trouble-deepfakes
[39] Legal Implications of Synthetic Media - https://carnegieendowment.org/2019/11/15/legal-ethical-and-efficacy-
dimensions-of-managing-synthetic-and-manipulated-media-pub-80439
[40] https://www.cliffordchance.com/insights/resources/hubs-and-toolkits/talking-tech/en/articles/2020/12/deepfakes-
and-legal-implications--seeing-is-not-believing.html
[41] https://carnegieendowment.org/2019/01/28/how-should-countries-tackle-deepfakes-pub-78221
[42] https://www.dacbeachcroft.com/en/gb/articles/2020/september/the-legal-implications-and-challenges-of-deepfakes/
[43] How technology used for synthetic media detection can be improved - https://partnershiponai.org/manipulated-
media-detection-requires-more-than-tools-community-insights-on-whats-needed/
[44] https://truepic.com/deepfakes-and-synthetic-media-are-also-threats-to-private-industry-provenance-technology-
can-help/
[45] https://ec.europa.eu/research-and-innovation/en/horizon-magazine/can-artificial-intelligence-help-end-fake-news
[46] https://www.forbes.com/sites/forbesbusinesscouncil/2021/05/17/how-to-combat-the-growing-synthetic-content-in-
disinformation-campaigns/?sh=23bca72650a9
JT Team A
[47] How Blockchain can be used to solve the issue - https://www.ibm.com/blogs/industries/blockchain-protection-
fake-news-deep-fakes-safe-press/
[48] https://www.wired.com/story/the-blockchain-solution-to-our-deepfake-problems/
[49] Governing access to synthetic media technology - https://techpolicy.press/governing-access-to-synthetic-media-
detection-technology/
[50] https://www.freiheit.org/pakistan/misinformation-pakistan-during-covid-19
[51] https://research.nvidia.com/sites/default/files/pubs/2017-10_Progressive-Growing-of/karras2018iclr-paper.pdf
[52][1812.04948] A Style-Based Generator Architecture for Generative Adversarial Networks (arxiv.org)
[53] Yuezun Li, Ming-Ching Chang, and Siwei Lyu. 2018. In Ictu Oculi: Exposing AI Created Fake Videos by Detecting
Eye Blinking. 2018 IEEE International Workshop on Information Forensics and Security (WIFS) (2018).
DOI:https://doi.org/10.1109/wifs.2018.8630787
