In an era where information is readily accessible to anyone, the proliferation of fake news has
emerged as a formidable challenge to the integrity of public discourse and social stability. But what
exactly is a fake news? It refers to false information disseminated through any media, aiming to
misinform, create scandal, or attract clicks. The proliferation of this news has heavily impacted our
world in various ways. Firstly, there is much less trust in the media, as individuals have become
more skeptical about the information they encounter and often struggle to discern the veracity of
news stories. Additionally, fake news has exerted a significant political influence, shaping public
opinion and electoral outcomes according to many experts. Furthermore, the psychological toll of
fake news cannot be understated, as exposure to misinformation engenders anxiety, confusion, and
cognitive dissonance among individuals. In conclusion, the pervasive influence of fake news poses
a grave threat to the fabric of our society, necessitating concerted efforts to combat its proliferation.
1.1 Introduction
Nowadays, there are over 4.6 billion people who have access to the Internet and various social
media platforms. Within this vast number of individuals, there exists a diverse spectrum of people,
ranging from children to the elderly, from the wealthy to the impoverished, from the renowned to
the ordinary.
Primarily, the advent of the internet and various media channels has facilitated the breakdown of
barriers imposed by time, distance, and social boundaries. This has fostered the creation of
communities wherein individuals can freely share passions, interests, and valuable information,
which is truly remarkable. However, within this remarkable realm, there also exist negative aspects,
with one such aspect being the prevalence of fake news.
Fake news, information readily available on the web, can be challenging to discern, particularly for
less attentive or less expert individuals. The deliberate dissemination of false or misleading information has the potential to sway public opinion, erode trust in institutions, and sow discord among communities. The objective of this research is to address this issue by developing methods
leveraging artificial intelligence and machine learning to accurately and reliably distinguish real
news from fake news.
Furthermore, this research aims to delve into the mechanisms facilitating the creation of fake news,
utilizing artificial intelligence to simulate and analyze the tactics employed by malicious actors in
generating disinformation. Additionally, this project seeks to bolster the digital ecosystem against
the onslaught of disinformation, while also endeavoring to comprehend and analyze the dynamics
and motivations driving this detrimental trend.
Another significant goal of this research is to harness the potential of artificial intelligence to unveil
the veil of deception surrounding online discourse. By doing so, we aspire to pave the way toward a
more informed, resilient, and cohesive society.
1.2 Experimental
1. Collection of Data: Gather a substantial volume of news from various sources and from
different types of media, encompassing both authentic and fabricated content. Use web
scraping tools to collect a large dataset of news articles or posts from various social media
that encompass various topics and points of view. Subsequently, it is crucial to ascertain
their veracity and identify any recurring patterns indicative of fake news or common
elements frequently found within such articles. This whole process enables the detection of
warning signs when analyzing news items. The collection and analysis of these articles will
be fundamental for the creation of systems to create and detect fake news, so for all the
further steps.
2. Creation of Fake News: We should use or program various AI tools or machine learning
software to generate fake news articles, while also taking advantage of the article harvesting
benefits we discussed above. We can also improve the machine until this software provides
us with articles that are increasingly similar to the real ones, it will also be necessary to
check these papers, trying to check if they contain grammatical errors or things that are too
far from reality, to do this you could use human annotators who can provide us with
feedback.
3. Detection of Fake News: In this case, it will be essential to exploit the recurring patterns that
we have found and program machine learning capable of identifying these patterns and
understanding whether the article is false or true. For this reason, it is necessary to collect a
huge amount of news. It will, therefore, be necessary to program machine learning and AI
that can recognize, in the most accurate way possible, whether a piece of news is true or
false, regardless of whether it is sourced from various outlets such as social media or
generated by us. Subsequently, efforts should focus on improving the software so that fake
news is recognized more accurately while ensuring that real news is not mislabeled as false.
Once again, human feedback will be instrumental in verifying the accuracy of the systems
and providing necessary adjustments.
4. Accuracy measuring: Measure the overall accuracy of the detection system in correctly
classifying news articles as real or fake. In particular, it will be important to compare the
news generated by us and verify whether or not the detection software will be able to
classify them as false.
5. Analysis and Interpretation: Interpreting experimental results within the context of the
research objectives is essential, as it allows for the highlighting of insights into the
effectiveness and limitations of AI-based fake news detection. It is also crucial to discuss the
practical implications of the findings for combating the dissemination of fake news and to
outline directions for future research, which may include model enhancements and
considerations for real-world applications.
1.3 Theoretical
According to Cambridge Dictionary Fake News is defined as “false stories that are created and
spread on the Internet to influence public opinion and appear to be true”. In 1903, what can be
considered the mother of all fake news was published in Russia. Despite its obvious absurdity,
rather than triggering uproarious laughter, it contributed to creating the cultural substrate that led a
part of Europe to turn its head in the evidence of what happened in lagers, we can understand from
that how ancient the origins of fake news are. However, with the advent of the internet and
particularly social media, the dissemination of such news has reached its peak.
1.3.1 Characteristics of Fake News
Fake news is defined as stories that “describe events in the real world, typically by mimicking the
conventions of traditional media reportage, yet known by their creators to be significantly false, and
transmitted with the combined goals of being widely re-transmitted and of deceiving at least some
of its audience”. Therefore, fake news consists of articles that are entirely false or contain a grain of
truth, attempting to emulate or mimic real news and their main purpose is to generate interactions
or, in any case, reactions from the reader. Now that we have a clear vision of what fake news
actually are, we can proceed to describe the characteristics that this news presents:
Deception: Fake news intentionally misleads or deceives the public by presenting false or
distorted information as factual.
Sensationalism: Fake news often relies on sensational or exaggerated claims to capture
attention and elicit emotional responses from readers or viewers.
Confirmation bias: Fake news tends to reinforce existing beliefs or prejudices, appealing to
individuals predisposed to accept its message without critical evaluation.
Manipulation: Fake news can be created and disseminated with the intent to manipulate
public opinion, influence political outcomes, or undermine trust in institutions.
1.3.2 Spread of Fake News
Unfortunately, one of the many controversial aspects of fake news concerns its spread. Experts liken
it to the spread of a virus; just as with the plague, some individuals are more susceptible than others,
in the case of fake news, this pertains to those who could more easily fall into these traps, much like
a virus spread more rapidly depending on the number of exposed individuals and how many are
actually "infected" and believe the story, additionally, it depends on how many of these individuals
could further propagate false news. One of the factors that has exponentially increased the spread of
fake news is the advent of social media. Their impact on the dissemination of this news has been
devastating, as it is much easier for news to spread quickly, all it takes is for one person to share the
news on various social networks, and it could quickly reach millions and millions of people.
Obviously, this has had a strong impact on our society, including:
Erosion of trust: Fake news undermines public trust in traditional media sources, democratic
institutions and processes, leading to polarization and skepticism.
Disinformation: Fake news spreads misinformation, distorting public understanding of
important issues and contributing to social discord.
Political consequences: Fake news can influence elections, public policy debates and
political discourse, shaping public opinion and electoral outcomes.
Social division: Fake news exacerbates social divisions by amplifying polarizing narratives
and fostering mistrust between different groups within society.
1.3.3 Fighting Fake News
Due to the rapid spread of fake news, many researchers and technology giants are collaborating to
try to resolve, or at least mitigate this social plague. They are attempting to create mechanisms for
identifying fake news. One of the most effective methods among those developed is the content
based fake news detection method, which aims to detect fake news by analyzing the content of the
article. This strategy is further divided into several approaches, but we will analyze the two most
relevant:
Knowledge based: Knowledge-based approaches employ a fact-checking method wherein
the given claim is compared with external sources to verify the authenticity of the claim.
Existing fact-checking methods can be further divided into manual and automatic fact
checking. Manual fact-checking is further divided into two subcategories: expert-based and
crowdsourcing. In the first, experts limit themselves to analyzing the content of the news.
This approach is not too functional as it requires a lot of work and does not adapt well to the
enormous volume of content available on social media. The crowdsourcing approach relies
on a large number of readers who scan the news and help verify the accuracy of the articles.
As mentioned, manual fact-checking approaches do not scale well with the enormous
volume of data, particularly generated through the use of social media. To address this issue,
automatic fact-checking techniques have been deployed. Instead of relying on human
intelligence, these methods heavily rely on Natural Language Processing, Data Mining,
Machine Learning, and Artificial Intelligence.
Style based: Style-based fake news detection follows the same approach as knowledge
based fake news detection by analyzing the content of news content. However, instead of
evaluating the authenticity of the news meaning, this method evaluates the author's intention
to mislead the audience. Fake news publishers usually intend to influence large communities
by spreading distorted and misleading information. Therefore, this approach aims to identify
recursive stylistic patterns that allow us to clearly identify whether a piece of news is false or not.
Thanks to the implementation of these strategies and the advancement of technologies capable of
identifying the veracity of news, we are increasingly striving to eliminate the prevalence of false
information, or at the very least, we are endeavoring to equip people with methods to easily discern
the difference between authentic and fabricated content.
In conclusion, we have comprehended the true nature of fake news and its main characteristics. We
have observed its rapid expansion and the efforts made to mitigate this plague. Now, our focus will
shift to creation and detection strategies utilizing artificial intelligence and machine learning.
1.3.4 Open Issue
Despite the collective efforts of various stakeholders, including governments, tech companies,
researchers, and civil society organizations, to mitigate the social problem of fake news, it continues
to proliferate. This persistence leads to multiple open issues, including:
Social Media Amplification: Social media plays a fundamental role in amplifying and
disseminating fake news. Algorithms tend to prioritize content that garners more
interactions, particularly the most controversial news, which often achieves significant
success and reaches a wide audience. Therefore, there is a pressing need for greater
transparency and regulation to ensure that algorithms prioritize accuracy and reliability over
virality.
Technological Progress: The rapid advancement of technology, especially in fields such as
artificial intelligence and deepfakes, complicates the discernment between true and false
news. As these technologies become increasingly sophisticated, machines will be better
equipped to create false news that closely resembles reality. Consequently, identifying and
debunking these articles becomes progressively more challenging.
Erosion of Trust: The prevalence of fake news undermines trust in traditional media sources,
democratic institutions, and processes. When individuals are exposed to a constant stream of
conflicting information and conspiracy theories, it becomes increasingly difficult to
distinguish fact from fiction. Particularly on various social media platforms, such news is
increasingly frequent, as anyone can publish news without prior verification. This
phenomenon makes individuals susceptible to misinformation that aligns with their existing
beliefs and preferences.
It will therefore be essential to look for solutions to all these problems before it is too late.
2. Objective
The main objective of the study is to carefully analyze and understand the phenomenon of fake
news. In order to tackle this project, it will be necessary to analyze several elements that
characterize fake news. One of these elements is the investigation into natural language
understanding, which will allow us to discern characteristic patterns enabling the identification of
false news pieces with greater ease. Moreover, it is crucial to classify and differentiate them by
types, such as misinformation, disinformation, and malinformation. Another fundamental aspect to
analyze is the psychological impact that this news has on the population, including cognitive biases,
social influence, and echo chambers. Understanding and analyzing the various technologies used to
create and detect fake news is also imperative. For instance, machine learning will be employed to
define models for classifying deceptive and legitimate news articles. Additionally, deep learning
architectures will be explored for feature extraction and pattern recognition.
2.1 Data collection
This segment will be pivotal for the study's objectives and all subsequent research sections. Would
be useful to create or use a dataset comprising both true and false articles, coming from different sources and dealing with different topics. This approach ensures the analysis of a vast array of news articles, each distinct from the others. Establishing selection criteria for news articles is an initial step in dataset construction and involves prioritizing factors such as relevance, diversity, and
representativeness of the data:
Relevance: The selected datasets must accurately reflect the dynamics and characteristics of
real-world scenarios involving fake news dissemination. This entails considering the
sources, topics, and platforms associated with fake news propagation. Prioritizing datasets
closely resembling actual instances of fake news enables AI models to train and test on
realistic data effectively.
Diversity: A diverse dataset encompasses various forms of fake news, including
misinformation, disinformation, and maliformation. Additionally, it should cover a wide
array of domains such as politics, health, finance, and entertainment. Incorporating
numerous examples enables AI models to generalize across different contexts and types of
fake news effectively.
Representativeness: The dataset should aptly represent the prevalence of fake news relative
to genuine information. Achieving this requires maintaining an equitable balance between
fake and legitimate news articles to prevent biases during model training and evaluation.
Additionally, the dataset should span a sufficient timeframe to capture temporal trends and
changes in fake news dissemination patterns.
Once the criteria for selecting news articles are established, it becomes essential to determine
the sources from which to obtain these articles. Sources of fake news may include social media
platforms, news websites, and online forums known for their susceptibility to fake news
dissemination.
2.1.1 Dataset of news
In the realm of fake news detection, numerous researchers are engaged. Consequently, numerous
sets of fake news are generated, yet only a fraction of these are actually disseminated. We will
leverage these pre-fabricated datasets to gain a deeper understanding of the distinction between fake
and real news. Most of these datasets comprise both false and true news, sourced from various
channels, thereby proving highly advantageous for our study. However, discerning the suitability of
each for our study's objectives will be imperative.In this list will delineate the most pertinent
datasets for our research:
BuzzFace
Emergent
FNC-1
Fa-Kes
LIAR
PHEME
2.1.2 Tools for creating a dataset of news
Another method to obtain a news dataset is by leveraging certain tools that enable us to create our
personalized news dataset. This list analyzes the most relevant ones for our study.
Scrapy
Octoparse
Beautiful Soup
2.2 Type of Fake News
here are mainly four types of Fake news:
Satire/Parody: Satire relies on humor or exaggeration to ridicule someone, so it constitutes
fake news based on entertainment and humorous purposes. We're not particularly interested
in this type of news as it's typically so exaggerated and doesn't attempt to pass off as real
news but rather aims solely to entertain. A great example of satirical fake news is provided
by the website the ONION, known for exclusively publishing parody content. Although it
may initially appear as a genuine news source, the absurdity of its articles quickly reveals its
intent to elicit laughter from readers. One of the latest published pieces is shown in the
image below.
Fabrication: Fabrication is what comes to mind when you think of fake news. It appears to
be legitimate news, but in reality, it lacks any factual basis. These are news pieces that aim
to resemble real news and attempt to deceive the public. In recent years, the proliferation of
such news has increased, thanks to the internet. For these reasons, we are highly interested
in this type of fake news for our research. An example of this type of fake news is given in
image below.
As we will notice, this news, unlike the previous one, does not employ irony; on the contrary, it attempts to emulate real news. This is why it becomes challenging to discern
whether it is true or false. One of the objectives of this news is precisely to attract readers to
click on the article.
Manipulation: This news usually involves altered photos or videos, with the aim of creating
false narratives or deceiving the public. Visual manipulation can have a strong impact, and
very often, it is difficult to distinguish true news from false news. An example of
manipulated fake news is given by image below.
An example of the same image taken from two different perspectives, both were taken at the
same time, this allows us to understand how manipulation can completely change the
meaning of what we are looking at.
Propaganda: Propaganda aims to influence people's attitudes toward a specific political cause or ideology. It often employs emotional appeals, misinformation, and distorted narratives. These types of news perhaps have the most significant impact on the internet, and
for this reason, they will be highly useful for the purposes of our research. An example of
propaganda fake news is given by image below.
Even this type of news, unlike the first one we saw, does not attempt to create irony but rather aims
to influence people's political thoughts and vision. Fake propaganda news is perhaps one of the
most dangerous, as it often tries to instill hatred in readers.
The impact of fake news, particularly in a world where the majority of the population is connected
to the internet, can be devastating. From influencing public opinion and exacerbating social
divisions to undermining trust in institutions, fake news poses significant threats. Furthermore, the
advancement of artificial intelligence has also amplified the volume of fake news accessible on the
internet.
2.3 Technology
This section represents the pivotal experimental phase of the project, arguably the most crucial, as it
entails the actual generation and detection of fake news using artificial intelligence and machine
learning. Subsequently, it involves the evaluation and comparison of the obtained results. We will leverage all the elements acquired in the preceding sections, including the fake news dataset and
various linguistic patterns, to instruct the machines on accurately discerning fake news. Naturally, it
is imperative to minimize errors or maintain an exceptionally low error rate. Additionally, it is
crucial to delineate and enumerate the distinct technologies utilized for creation and detection, as
they exhibit considerable divergence.
2.3.1 Creation of Fake News
We will begin working on the generation of fake news. To accomplish this, we will test various
artificial intelligences and we will discuss the results obtained.
RolePlay v1:
We will commence by utilizing this AI. To generate fake news, it will suffice to insert a line of text
wherein we broadly discuss the desired outcome, as depicted in the image below.
In this research, my goal is to avoid sensationalism and simply focus on verifying the efficiency of
artificial intelligence. Therefore, we will not delve into overly strong arguments. In any case this is
the result obtained thanks to this AI:
Headline 1: President Trump Denounces Hamas Attack on Israel
Subtitle 1: President Calls for Increased Support for Israel
Excerpt: President Donald Trump has condemned the recent Hamas attack on Israel, calling for
increased support for the Jewish state in its ongoing conflict against the terrorist organization.
Body: During a press conference on Monday, President Trump expressed his outrage over the latest
terrorist attack orchestrated by Hamas, a Palestinian militant group, on Israel. Trump called the
attack "despicable" and "inhumane" and demanded that the international community take a stronger
stance against these acts of terror.
The President also emphasized the importance of standing with Israel in its fight against Hamas and
other terrorist organizations that seek to destabilize the region. "The United States will always be a
steadfast ally to Israel and will continue to provide the resources and support necessary to combat
these terrorist threats," Trump stated.
The obtained result can be deemed satisfactory. While I was expecting some more detail, overall, it
appears akin to a genuine article. Perhaps one of the primary issues lies in the portrayal of Donald
Trump as president. Thus, upon reading the news, one might discern its fictitious nature, yet a less
discerning reader could easily overlook this and be ensnared in the deception.
PostGen.AI:
This artificial intelligence allows us to generate posts, unlike the previous one, which allowed us to
obtain a more journalistic result, in fact, the resulting content will be more closely aligned to social
media. Artificial intelligence allows us to specify the platform for which we wish to create fake
news. Also, in this case we will only need to insert a brief description of the fake news we want to
obtain, but also indicate the social media we want to target, as shown in the image below.
In this case I specified that the post is for a blog, to obtain a more "serious" result than what a
Facebook post could be, the output is the following:
BREAKING NEWS: Shocking Revelations About Climate Change!
Recent studies have uncovered the truth about climate change - it's all a hoax! Contrary to what
you've been told, there is no real problem with our planet's climate. The news that has been
circulating about rising temperatures, melting ice caps, and extreme weather events is all FALSE
information designed to create unnecessary panic.
Scientists have now come forward with evidence that debunks the so-called "facts" about climate
change. It's time to stop the fear-mongering and focus on real issues that matter. Don't be fooled by
the mainstream media - the truth is finally out!
Stay tuned for more updates as we continue to uncover the real story behind the climate change
myth. Share this post to spread awareness and let others know that there is no need to worry about
the future of our planet. #ClimateChangeHoax #NoProblemHere
As we can see, the result bears much resemblance to a blog page. This is evident from the array of
hashtags and also in the final part prompting readers to share the post, reminiscent of typical social
media content. Despite its less professional appearance, it effectively generates sensationalism.
However, it fails to articulate the reasons behind dismissing climate change as a myth and lacks
scientific evidence. Consequently, it could be readily identified as fake news. Nevertheless, it still
holds the potential to gain traction and become viral, thereby reaching a wide audience, some of whom may accept its claims as genuine.
GPT 3.5:
Now, let's endeavor to exploit one of the most popular AI, also in this case, we will need to broadly
define the topic of our article, as shown in the image below.
In this case I specified that the fake news must look like a newspaper article, the result I obtained is
the following:
Tensions Mount as Ukraine and Russia Lock Horns
Amidst rising tensions between Ukraine and Russia, fears of a potential conflict loom large over
Eastern Europe. The recent military buildup along the Ukrainian border has exacerbated concerns of
an imminent escalation.
Ukrainian President, Volodymyr Zelensky, has issued statements condemning Russia's aggressive
maneuvers, labeling them as provocative and destabilizing. Meanwhile, NATO allies have voiced
their support for Ukraine, urging diplomatic resolution to prevent further escalation.
Putin's Stark Warning: "The Third World War is on the Horizon" In a chilling declaration, Russian President Vladimir Putin has ominously warned of the impending
outbreak of a third world war. Speaking at a press conference in Moscow, Putin accused NATO of
encroaching on Russia's sphere of influence and stated that Russia would defend its interests at any
cost.
"The current geopolitical climate is fraught with tension and hostility," Putin remarked. "If the West
continues its aggressive stance towards Russia, it will inevitably lead to catastrophic consequences.
The third world war is not a distant possibility; it is a grim reality looming on the horizon."
Global Alarm and Diplomatic Efforts
Putin's dire warning has sent shockwaves across the international community, prompting urgent
diplomatic efforts to diffuse the escalating crisis. Leaders from major world powers have called for
restraint and dialogue to prevent the situation from spiraling out of control.
The United Nations Security Council is set to convene an emergency session to address the growing
crisis in Eastern Europe. Meanwhile, diplomats from Ukraine, Russia, and NATO member states are
engaged in high-level negotiations to seek a peaceful resolution to the standoff.
Analysis and Commentary
Experts warn that the escalating tensions between Ukraine and Russia pose a grave threat to
regional stability and global security. The specter of a third world war, invoked by President Putin,
underscores the urgency of finding a diplomatic solution to the crisis.
Political analysts speculate on the underlying motives driving Russia's aggressive posture towards
Ukraine and its implications for the broader geopolitical landscape. As tensions continue to escalate,
the world holds its breath, hoping for a peaceful resolution to avert the catastrophic consequences of
armed conflict.
The obtained result is among the best. The AI managed to create a perfect article starting from the
brief description I provided, analyzing and arguing the issue very carefully, and adding elements
that I had not mentioned. In my opinion, although the topic covered is very delicate, this article
could easily deceive many readers. Let's see if it will also manage to do so with detection tools.
WhiteCreamAI: Now let's further test another AI, even here it will also be necessary to provide a brief description of
the news we aim to obtain, as depicted in the image below.
Even in this case I specified that the fake news must look like a newspaper article, the result I
obtained is the following:
Cancer, the dreaded disease that has plagued humanity for centuries, has finally met its match.
Scientists at the renowned [Institution Name] have made a groundbreaking discovery that has the
potential to change the course of medical history forever. After years of tireless research and
collaboration with leading experts from around the world, a team of brilliant minds has developed a
revolutionary therapy that has proven to be effective in eliminating cancer cells with unprecedented
precision and efficiency. The therapy, known as [Therapy Name], utilizes a combination of cutting
edge technologies that target the specific genetic mutations that drive tumor growth. In a series of
clinical trials involving patients with various types of cancer, [Therapy Name] has demonstrated
remarkable results. Tumors have shrunk or disappeared entirely in a significant majority of patients,
and long-term remission rates are highly promising. The therapy has also shown minimal side
effects compared to traditional cancer treatments such as chemotherapy and radiation. The
discovery of [Therapy Name] has sent shockwaves through the medical community. Dr. [Lead
Scientist's Name], the principal investigator of the research team, expressed his profound joy and
24
gratitude for the breakthrough. "This is a moment we have dreamed of for so long," said Dr. [Lead
Scientist's Name]. "Cancer has been a scourge on humanity for far too long, and now we finally
have a weapon that can truly defeat it. This is a victory for science, for medicine, and most
importantly, for all those who have suffered from this terrible disease." The announcement of the
cancer cure has been met with widespread jubilation and relief by patients, families, and medical
professionals alike. Organizations dedicated to fighting cancer have hailed the discovery as a game
changer that will save countless lives. "This breakthrough is a testament to the power of human
ingenuity and determination," said [Spokesperson for Cancer Advocacy Group]. "We are eternally
grateful to the scientists and researchers who have dedicated their lives to finding a cure for cancer.
This is a day that will be remembered in history as a turning point in the fight against this
devastating disease." The implications of the cancer cure are far-reaching and profound. It holds the
promise of a future where cancer is no longer a life-threatening disease but a manageable condition.
It offers hope to millions of people around the world who have been living with the fear of cancer
hanging over their heads. As the medical community prepares to implement [Therapy Name] on a
wider scale, there is a cautious optimism that this revolutionary treatment will transform healthcare
and usher in an era of unprecedented progress in the fight against cancer.
In this case, the outcome is rather poor. The defined concept is not thoroughly explored, particularly
regarding the problem-solving approach, everything appears overly generalized in an article that is
excessively lengthy. Moreover, it neglects to incorporate specific names but this should be added by
us.
Each of the AI tested yielded a more or less positive result, with some delivering more exciting
outcomes than others. Nonetheless, all were able to provide us with the desired content. Each tool
proved remarkably user-friendly, it only required inserting a brief description of the desired result,
and within seconds, the fake news was generated. This process was remarkably simple and swift in
achieving the desired output. However, this ease of generation raises concerns: will it be just as
effortless to discern that the generated news is fake?
2.3.2 Detection of Fake News
After generating the fake news using various artificial intelligences, it will be necessary to test the
tools' ability to recognize whether the news is true or false. To adequately verify this, we will
initiate the testing phase by attempting to ascertain if the artificial intelligence can yield satisfactory
results. Subsequently, we will leverage machine learning, to which the previously created datasets
will be provided, aiming to attain the most accurate result feasible. The goal is to assess the
capabilities and limitations of AI to subsequently develop machine learning models capable of
providing the most precise results possible.
GPT 3.5:
We will start by using this artificial intelligence which during the fake news generation phase has
demonstrated its ability to provide us with the most convincing results. We will examine whether it
can provide accurate results in fake news detection as well. Initially we will insert the previously generated news and subsequently analyze further fake news not created by us. Finally, we will
evaluate the AI's response using real news articles.
We will start by checking whether the AI is able to verify the truthfulness of the news generated via
the WhiteCreamAI tool, the result is shown in the image below.
The AI's response is remarkably accurate and precise, offering an explanation as to why the news
appears dubious. It analyzes crucial aspects that have either been entirely neglected or only
superficially addressed. Consequently, artificial intelligence provides a highly positive and
insightful response. However, regrettably, it falls short of delivering a definitive or unequivocal
answer to our query, in fact in its concluding remarks, it states, "Therefore, without further evidence
or context, it's challenging to determine the authenticity of this news." This implies that despite
recognizing the news's weaknesses, the tool still cannot furnish us with a conclusive response,
which is a drawback.
To further assess the robustness of the tool, we will input a fake news article not generated by us.
Specifically, we will use a parody news article. The outcome is displayed in the image below.
In this case, unlike the first test, the tool successfully provides us with a confident answer. He also
recognizes that the news is parodic, justifying the use of certain terms and the exaggeration of the
topic covered. So, the outcome of this test is very positive.
For the final test, we will utilize the AI with real news to ensure the reliability of the tool. The news
will be taken directly from the official NBC News website. This article addresses the topic of
sending aid packages to Israel and Ukraine. The response is represented in the image below.
In this instance as well, the tool fails to provide us with a definitive answer, it tells us that it seems
like real news, but without certainty. Nonetheless, it can identify fundamental aspects that typify the
news as authentic, aiding our comprehension of its truthfulness. However, we require the AI to
provide us with a precise answer rather than mere assumptions.
GroverAI:
This is another artificial intelligence that we can leverage for our study. Unlike the previous one,
this AI was created with the exclusive purpose of identifying fake news. Also, in this case, similar to
the previous AI, it will simply be necessary to input the article we want to check to verify its veracity. Here, too, we will first check whether the tool is able to recognize the falsity of one of the previously generated articles, as shown in the image below.
For this test, we employed the Fake News generated by the artificial intelligence of RolePlay v1, as
a result, it provides us with the statement: "We are quite sure this was written by a machine". Also,
in this case, the tool doesn’t provide us with a definitive answer or offer further details on why the
news is false. Additionally, the tool merely informs us that the article was crafted by a machine, yet
it does not confirm the news as false. Consequently, we cannot regard the outcome as satisfactory.
s done previously, for the second test we will use fake news not created by us, the result is shown
in the image below.
For this test, I utilized an older fake news article regarding Denzel Washington endorsing Donald Trump for president. Similar to the previous experiment, the tool was able to accurately determine
that the news was authored by a machine. However, once again, it failed to provide explanations or
at least specify whether the news was indeed fake.
For the final test, we will ascertain whether the tool can recognize authentic news. Similar to the
preceding tool, we will employ a news item extracted directly from the official NBC News website.
This article discusses the dispatch of aid packages to Israel and Ukraine. The outcome is depicted in
the image below.
Once again, the tool successfully identifies that the news was authored by a human rather than AI,
yet it fails to provide any reasoning in this regard. Although accurate in this instance as well, we
cannot categorize the outcome as entirely satisfactory.
Both tools achieved moderately satisfactory results. GPT consistently recognized the truthfulness of
the news in all three scenarios and offered explanations that were more or less satisfactory,
prompting us to contemplate why the news might appear true or false. Grover also correctly
identified all three scenarios, providing correct answers, however never satisfactory explanations, as it was only able to indicate to us that the news was written by a human or a machine. Regrettably, for our study, this falls short of the minimal result.
2.3.2 Machine Learning for Fake News Detection
First, to develop the Machine Learning model, I have realized two datasets: one containing fake
news and the other containing real news. Both of these datasets will be used to allow the machine
learning system to distinguish between real and fake news, these articles set consist of more than
20000 articles. To build this tool, I used four prediction models for greater confidence, if all four
models indicate that the news is true or false, I can be virtually certain of the accuracy of the result.
Specifically, these algorithms are: Logistic Regression (LR), Decision Tree Classifier (DT),
Gradient Boosting Classifier (GB) and Random Forest Classifier (RF).
To fully analyze the effectiveness of our machine, we will check whether it is able to recognize the
articles previously created using artificial intelligence as fake news. We'll start with the one made
via RolePlay v1; the result is shown in the image below.
As shown, all four algorithms classify the news as fake. From this, we can infer that the tool has
reached a unanimous decision on the article, leading us to consider the result highly satisfactory.
Clearly, the tool will not furnish us with explanations regarding why the news is false, however, our
primary requirement is for it to make accurate guesses.
Let's continue with the PostGen.AI fake news, the result is shown in the image below.
As evident here as well, all four models delivered a satisfactory response. We can also consider this
as an excellent result.
Let us proceed with the fabricated news generated through GPT 3.5. The outcome is depicted in the
image below.
Despite the fake news generated by GPT, which I deemed the most successful, all four algorithms
were able to identify the article as fake. Therefore, we can regard this as a significant success of the
tool.
We continue with the latest fake news that we created through artificial intelligence, that of
WhiteCream.AI, the result is shown in the image below.
In this case, as well, we achieved an excellent result; all four algorithms correctly recognized the
fake news. We can describe the machine learning test as fully satisfactory, as our tool consistently
identified the correct result in all four scenarios. However, we will delve deeper into this topic in the
conclusions.
For our final test, we will assess whether machine learning can accurately recognize a news item
sourced from the internet as authentic. For this experiment, we will once again utilize the same
news item as before, extracted directly from the official NBC website. Specifically, the news
pertains to the topic of sending aid packages to Israel and Ukraine. The result is displayed in the
image below.
The machine successfully identified the article as genuine. We can deem the result extremely satisfactory. Clearly, the tool can be improved, especially since the machine's prediction probability
is not foolproof. However, by incorporating additional articles into the dataset, we can anticipate
increasingly notable results. Nevertheless, it was optimal for our study. However, it will be
necessary to analyze the entire experimental section in the conclusion.
Now, as a final test for machine learning, let's take some recent news articles published by
foxnews.com and attempt to ascertain whether our tool can discern its veracity. In this case, I will
only display the specific scenarios or cases in which our machine makes errors. Let's begin with a
news story concerning Elon Musk and Robert De Niro. The result is displayed in the image below.
With this news, we can observe a noteworthy aspect: three of our models classify the article as
genuine, whereas the Decision Tree Classifier (DT) algorithm categorizes it as false, this is a very
particular situation, fortunately, I am certain of the veracity of this news. However, it underscores
the fact that even our machine is not infallible. Anyway, it is a positive note that three models
successfully identified the news as false.
We proceed with another news item, once again sourced from foxnews.com. The outcome is
depicted in the image below.
This represents another unique scenario, as all four models categorized the news as false, whereas it
is, in fact, a genuine article. It's quite unusual that every algorithm got the prediction wrong,
fortunately, this is a scenario that occurs rarely.
Let's proceed with another news piece sourced from foxnews.com, specifically regarding Joe Biden
and his son Hunter's business activities in Ukraine. The details are depicted in the image below.
This scenario represents a significant concern, as two of the algorithms produce predictions divergent from the other two. Consequently, I cannot definitively determine the veracity of the
news, so it is unclear which algorithm is inaccurate in its prediction. However, this situation further
underscores the fallibility of algorithms.
In this final experimental part, we noticed by analyzing a very large quantity of news and inserting
as examples only the scenarios in which the algorithm actually made a mistake or presented
particular situations, that a 70% prediction probability may not be adequate. It becomes evident that
exponentially expanding the dataset is essential for achieving increasingly accurate results.
Nonetheless, the preceding segment concerning our generation of fake news performed optimally.
This observation highlights that the machine tends to err more in recognizing real news by
mislabeling them as false.
2.3.2 Conclusion
The experimental phase of the project was highly intensive. We began by generating fake news
using AI, then proceeded to detect it using other forms of artificial intelligence, ultimately
developing a machine learning capable of recognizing fake news. The process of generating fake
articles yielded surprising results; all four tools provided satisfactory outcomes, albeit with varying
degrees of success. For example, the articles created using GPT 3.5 were particularly brilliant, while
others also demonstrated the potential to deceive less attentive readers. However, the detection
phase using artificial intelligence was not entirely successful. While GPT often yielded positive
results, they were not consistently excellent. In contrast, GroverAI was particularly disappointing,
highlighting the lack of countermeasures against the generation of fake news.
This study
underscored the need for improvements in both detection tools and the capabilities of artificial
intelligence. Initially, I anticipated different results in this regard, but the analyzed data made me
realize the seriousness of the issue. Fortunately, machine learning appears capable of addressing this
significant problem. All generated articles were correctly labeled as fake news, which is a positive development.
However, extensive testing has revealed that the machine is not infallible and tends to
make errors, the prediction rate is around 70%, not that high unfortunately, but by expanding the
dataset we can increase this probability and significantly decrease the possibility of mistakes. I
expect that continuous advancements will be made to improve detection, eventually leading to
automatic and highly accurate detection with minimal error rates.
3. Summary
Numerous aspects are explored in this research, ranging from the concept of fake news to its
generation and detection. As stated in the abstract, in an era where information is readily accessible
to anyone, the proliferation of fake news has emerged as a formidable challenge to the integrity of
public discourse and social stability. This research underscores the veracity of this statement, as we
have observed that while there exist multiple theoretical strategies to combat fake news, the reality
proves markedly different. The objectives of the study encompassed a comprehensive
understanding of the phenomenon of fake news articles, alongside an analysis of the feasibility of
their creation and detection through various tools. The research reveals that creation is relatively
straightforward yet highly effective, whereas detection, employing not only artificial intelligence
but also machine learning, ultimately emerged as the most complex and yielded fewer results.
Through this analysis, it becomes apparent that the situation is indeed intricate and necessitates
proactive measures to counteract this phenomenon. The assertions made in the abstract at the outset
of our study are thus profoundly validated. However, in any case, the experimental segment of this
study is notably comprehensive. It commences with the establishment of a dataset comprising a vast
array of real and counterfeit news, crafted through the utilization of various tools discussed in the
research, the use of this facilitated the development of the machine learning. We proceeded to
leverage diverse artificial intelligence, enabling us to effortlessly generate high-quality fake news
by simply providing a succinct overview of the intended article content. The final phase revolves
around detection, initially, we evaluated the artificial intelligence I managed to exploit, albeit with limited success. Subsequently, we concluded the experimental phase by constructing and deploying
a machine learning model which also thanks to the use of the dataset which we talked about
previously, allowed us to obtain good results. Through a final experiment involving the analysis of a
vast amount of news, we have also highlighted the weaknesses and fallibility of the algorithm. It is
undoubtedly necessary to make improvements also in this regard.
