Challenging yet enjoyable research paper experimenting how online media users perceive news credibility in three cases; namely AI versus machine versus tandem authorship.
Presentation done on 8-Feb-2020 with my study-mate Laila Khaled, within the "Current Issues in Mass Communication course" led and instructed by Prof. Shahira Fahmy, professor of Journalism and Mass Communication at the American University in Cairo.
P.s.: some content in the presentation was used for illustration only
Can an algorithm reduce the perceived bias of news - Dr. T. Franklin Waddell - 2019 JMCQ
1. Can an Algorithm Reduce
the Perceived Bias of
News? Testing the Effect of
Machine Attribution on News
Readers’ Evaluations of Bias,
Anthropomorphism, and
Credibility
T. Franklin Waddell
2. The American University in Cairo
Current Issues in Mass Communications - JRMC 5502-
01
Course Instructor: Dr. Shahira Fahmy
3. CONTENTS OF THIS PRESENTATION
THE FOLLOWING CONTENT IS COVERED IN THIS PRESENTATION:
1. Introduction video about AI by Bassem Akram Senior Data Management Analyst at the UNHCR
2. Discussion Questions
3. Role of AI in Misinformation
4. AI Journalism examples
5. Common criteria used to judge credibility of news
6. Heuristics
7. Affordances
8. MAIN Model and its four affordances
9. Anthropomorphism
10. Hypotheses
11. Method & Procedure
12. Main Study Supplemental Material
13. Supplemental Analysis
14. Results
15. Discussion
16. Limitations
17. Future Research Directions
18. Implementing this study in Egypt
19. Recommendation for AI Courses
20. Resources
7. AI WILL MAKE A LOT OF JOBS DISAPPEAR
“The rise of artificial intelligence is likely to extend this job destruction deep into the middle
classes, with only the most caring, creative or supervisory roles remaining." Stephen Hawking
17. AI Journalism
Different degrees of human
interference in Artificial
Intelligence Journalism.
Is 4.4 jolt an end to
Los Angeles’
earthquake
drought?
A robot wrote this
entire article. Are
you scared yet,
human?
2014 2020
18.
19. How do readers commonly judge the credibility of
news?
Needing more mental
processing, thus more
effort
Using ready mental rules of
thumb, minimal effort. Such
as length and source
heuristics
Capabilities that are
conveyed by an object that
makes it user-friendly
SYSTEMATIC PROCESSING
OF MESSAGES
HEURISTICS or Cue-based
AFFORDANCES
20. “Machine Heuristics” which is a rule of thumb that machines are more
secure, and trustworthy than humans (Sundar & Kim 2019)
Information overload online
Common heuristics in media Cues = markers/indicators
HEURISTICS
People are cognitive misers
23. MAIN
Model
The elements of the MAIN model by Sundar (2008) in
the figure below, and which are the basis for the study
at hand, are the affordances explained by the video.
Other elements of the MAIN model are: Modality,
Interactivity,and Navigability.
24. MODALITY
Close to the concept of the
medium, being it aural, textual
or audiovisual. Multimodality is
the multiple forms of media
combined together.
INTERACTIVITY
The interaction and activity; the
shift of moving from the passivity
of using traditional media, to the
activity of using digital media.
AGENCY
The identity of the source to
the receiver. The source can
be a human author, a
computer, or a news
organization.
NAVIGABILITY
Features that suggest moving
from one location to another
online or offline.
02
01
04
03
Perceived credibility of the audience is altered by those
four affordances (Sundar, 2008)
25. AGENCY
Similarity Attraction
Low perceived bias/ high
perceived credibility
High perceived bias/low
perceived credibility
Less human-like
(anthropomorphic)
HUMANS MACHINES
+ +
- - "One factor in user trust is the degree
to which a system is perceived as
human-like, or anthropomorphic"
(Jensen, Khan and Albayram, 2020).
32. PROCEDURES
612 Participants
from MTurk to
achieve (80%
power) at Alpha =
0.05
Effect Size = 0.15
Sample Size
Pretesting
Using Fox News and
MSNBC, which are
known for their
partiality
Recall Test
Using Independent
Sample 1, 92% of the
participants were
able to recall the
sources
The study was approved by the
institutional review board at the
primary investigator’s university.
36. WOULD PARTICIPANTS’
PERCEIVE AN ALGORITHM
TOOL (Quill) AS MORE
MACHINE LIKE COMPARED
TO A MACHINE AUTHOR
POWERED BY AN AI
COMPANY (Automated
Insights)?
Results revealed that both (Quill &
Automated Insights) were perceived
as more machine-like than a
human author.
NO
37. RESULTS
All hypotheses were supported except H5, it was partially supported because findings
showed that, tandem authors were perceived as more credible than human authors
alone via the indirect pathway of bias, but less credible than human authors alone via
source anthropomorphism.
38. DISCUSSION
Due to Machine Heuristics,
AI authors lead to higher
perceived credibility than
Human authors.
Due to similarity
attraction, AI authors are
perceived as less credible
than human authors as a
result of their lower
human-likeness (less
anthropomorphic).
A supplemental test conducted
after the main study showed
that tandem authorship of
different authors (AI & human),
versus multiple authors, has
significant effect on bias.
Participants tend to be
distracted from the authors of
articles when the stimuli
(content) is placed in an
information overloaded context
like Social Media.
2
3
4
1
39. LIMITATIONS
The underrepresentation of the
conservatives/republicans had an
effect on the results
Stimuli pertained to
politically motivated
current events prone to
perceptions of bias
Exposing them to
content directly
from source might
have had an effect
on the ability to
recall the source
Sample Stimuli
Context
40. FUTURE RESEARCH DIRECTIONS
Tandem Social Media
Higher Human
Likeness
Other Fields
More research on
the tandem
relationship between
machine & human.
Testing the effect of
the reader’s ability to
recall the author on
Social Media on the
psychological effect
on automation.
Testing whether
attributing human
traits to machine
authors can mitigate
the negative effects
on credibility via the
indirect route of
anthropomorphism.
Studying how
people respond to
automated authors
in fields other than
politics.
AI Literacy
The effect of the
degree of
familiarity with
automation on
perceived
credibility.
41. IMPLEMENTING THIS STUDY IN EGYPT
“As for possible moderators,
additional studies should evaluate
not just the effects of automation
overall, but also probe specific
variables that might condition the
effects of purported machine
authorship such as technological
expertise or familiarity with
automation.” (Waddell, 2019)
What inspired us
42. IMPLEMENTING THIS STUDY IN EGYPT
The role of “AI Literacy”
on the effect of human,
automated and tandem
authorship on perceived
credibility of news.
Participants will be pre-
tested for their AI literacy
through a questionnaire
which should have high
internal consistency
(reliability).
600 students, half from a
gov. and another half
from private universities
Research
Problem
Pre-test
Sample
43. H1
Machine attribution (versus
human attribution) will mitigate
perceptions of media bias in the
case of high AI literate students
through Machine Heuristics.
H3
Machine attribution (versus human
attribution) has minimal effect on
perceptions of source
anthropomorphism in the case of
high AI literate students.
H5
Machine attribution (versus human
attribution) will mitigate perceptions of
media bias in the case of low AI literate
students through Machine Heuristics.
H2
Machine attribution (versus human
attribution) will enhance
perceptions of news credibility via
the indirect pathway of media bias
in the case of high AI literate
students.
H4
Machine attribution (versus human
attribution) will have minimal effect on
news credibility via the indirect pathway
of anthropomorphism in the case of
high literate AI students.
H6
Machine attribution (versus human
attribution) will enhance perceptions of
news credibility via the indirect pathway of
media bias in the case of low AI literate
students.
H7
Machine attribution (versus human
attribution) has a high effect on
perceptions of source anthropomorphism
in the case of low AI literate students.
H8
Machine attribution (versus human
attribution) will have a negative effect on
news credibility via the indirect pathway of
anthropomorphism in the case of low literate
AI students.
HYPOTHESES
44. Online Courses for AI literacy
1. Data Science and Machine Learning: AI for Everyone - on Coursera by Andrew NG
https://www.coursera.org/learn/ai-for-everyone/home/welcome
2. Understanding the Impact of Deepfake Videos
https://www.linkedin.com/learning/understanding-the-impact-of-deepfake-videos/the-
strange-reality-of-deepfake-media?u=57686545
46. - Sundar & Kim 2019: https://doi.org/10.1145/3290605.3300768
- Sundar 2018 MAIN Model: https://www.issuelab.org/resources/875/875.pdf
- Affordances: https://www.theatlantic.com/technology/archive/2014/03/earthquake-bot-los-angeles-times/359261/
- Book “Artificial Intelligence in HCI”: https://link.springer.com/content/pdf/10.1007%2F978-3-030-50334-5.pdf
- Alexa Ad: https://www.youtube.com/watch?v=xxNxqveseyI
- Alexa “5 things you didn’t know Alexa does”: https://www.youtube.com/watch?v=W3DEJgnGZYc&t=2s
- Artificial intelligence: How to turn Siri into Samantha https://www.bbc.com/news/technology-26147990?piano-modal
- Black mirror "Husband clone": https://www.youtube.com/watch?v=dK9f-vMh0bw
- The North Wind and the Sun: https://www.youtube.com/watch?v=51_FHblK4mc
- “Dehumanization: An Integrative Review” (Haslam, 2006):
https://www.researchgate.net/publication/6927454_Dehumanization_An_Integrative_Review
RESOURCES