Research Proposal for the Effect of Machine Learning on
Social Media
Yating Han
March 19, 2018
1
1 Introduction
At 7:00 pm which you usually reserve for some goofiness, your phone suddenly rings:
Facebook suggests you might know Joe. After a few minutes of stalking, you finally realize
your childhood friend who left the town at ten. You accept his friend request and click on
the “New Immigration Ban” article Facebook believes you are interested. But how does
Facebook know?
The answer is machine learning (ML) - the hottest and most mysterious word of the 21st
century. Without understanding the subject itself, nevertheless, the general public seems to
share various assumptions. As a result, many blame it for issues such as social bubble and
the spread of fake news besides its convenience.
Machine Learning is primarily used in the following approaches:
• Analyzing trends;
• Detecting social connections;
• Facial recognition;
• Sentiment analysis;
• Targeted Advertisement.
These tools combined can lead to the priming of posts.
Note that, all of the approaches above uses either reinforcement learning or unsuper-
vised learning, which humans have very little control technically. These are strong evidence
that machine learning is indeed responsible for most social bubbles in social media. In the
meantime, it suggests that ML might not spread naive advertisement by itself.
However, there is barely a research on the exact relationship between the use of ML in
social media and social psychology. Without the support of concrete research, it is impos-
sible to take subsequent actions. If ML is indeed the problem, regulations must be made;
otherwise, it is urgent to find the actual cause.
2
The research I am proposing is to compare the effect of psychology and ML on 1) Spread
of in fake news; 2) Social and political bubble; 3) Different types of advertisement. Doing so
will decrease viewers’ vulnerability in social media. With this aim, this research will be on
social science standpoint rather than focusing on algorithms.
2 Synthetic Review
The given articles suggest some insights on common views on Machine Learning. More
specifically, all authors chose to answer at least one of the following:
• How does ML relate to fake news?
• Is social media more dangerous with customization?
• Should we “regulate” the algorithms?
How does ML relate to fake news?
In “Facebook is Eating the World”, Bell states explicitly that “social platforms have to
employ algorithms to try and sort through the important and recent and popular and decide
who ought to see what” (Bell). This is a rather pessimistic view: the viewers seem to have
zero control over the contents. Herrman also admits that Facebook’s suggests “become hard
to miss”, but disagree on the extent in which the Facebook influences its viewers’ mind by
merely priming certain contents (Herrman). Both authors believe that ML implants fake
news in our mind, consciously or not.
Is social media more dangerous with customization?
All except one author mentioned ML agree that social media is more dangerous than it
used to. In “Social Media and Fake News in 2016 Election”, the authors found a positive
3
correlation between social media and vulnerability of fake news. Furthermore, those who
have segregated social networks are more likely to believe in fake news that aligns with their
existing views (Allcott and Gentzkow). Besides, some author argues that Facebook has full
power over our choices, as Bell states, “we have no option but to trust them to do this”.
However, both articles fail to analyze whether the benefit excess disadvantages (Bell).
Should we “regulate” the algorithms?
Herrmann, again, believes that more regulation on algorithms must be made “to make sure all
citizens gain equal access to the networks of opportunity and services they need” (Herrman).
On the other hand, Morozov has a positive view of the situation: the current problem will
be resolved while the democracy evolves, therefore regulations are reluctant and may even
backfire. To him, people “blame everyone but themselves while offloading more and more
problems to Silicon Valley” (Morozov).
General Idea
Although there is clearly a lack of understanding of Machine Learning, the authors give
interesting views on the role of ML in social media. Without a doubt, ML is changing the
way social media offers information. While one must admit that ML can lead to the belief
in fake news, we cannot determine the net cost of ML in social media.
Among the most authors, we see a shared fear about algorithms, which is also true on
a larger scale. As stated in the Introduction, there is little evidence to support such fear.
Therefore, it is crucial to conduct the research I am proposing.
4
3 Object of Study
The object of study will be choosing the dominance between viewers and algorithms for
existing problems in social media. To answer this question, the following question must also
be answered:
• What is the most basic idea behind Machine Learning? To what extent can program-
mers manipulate it?
• What are some psychological reasons that may boost the effectiveness of ML?
• How much control does the viewer have over the algorithms?
4 Plan of Work
Due to the complex nature of this topic, different strategies will be used in different
sections of the research. Mostly, the comparison between human and algorithm contributions
will be based on arguments and logic. However, claims will be supported in evidence in the
following forms:
Social Science Evidences
A short survey will be conducted on Chinese international students. It will compare
WeChat Moment (a Chinese social media) and Facebook. The main difference between
them is that WeChat Moment sorts in chronological order while Facebook uses Machine
Learning. Sample Question: On which platform do you get irritated more often?
If possible, I will survey several CMU student and faculties in the School of Computer
Science and Dietrich College of Humanities and Social Sciences, and compare their views on
given problems.
5
Finally, I will use peer-reviewed articles on social media and social psychology, and news
articles from credible sources, including class materials. The latter will be used primarily on
case studies.
Technical Evidences
Technical evidence will be primarily based on existing papers on Machine Learning and
online resources. I am not planning to get into details on the technical side, however.
6
Works Cited
Allcott, Hunt and Matthew Gentzkow. “Social Media and Fake News in the 2016 Election”.
Journal of Economic Perspectives 31.2 (2017): 211–236. Print.
Bell, Emily. “Facebook is eating the world”. Columbia Journalism Review (Mar. 2016). Web.
Herrman, John. “Inside Facebook’s (Totally Insane, Unintentionally Gigantic, Hyperparti-
san) Political-Media Machine”. The New York Times (Aug. 2016). Web.
Morozov, Evgeny. “Democracy is in crisis, but blaming fake news is not the answer”. The
New York Times (Jan. 2017). Web.
