Hi! I prepared slides for each chapter of my book 'Algorithmic gatekeeping for professional communicators - power, trust and legitimacy'. (OPEN ACCESS: https://doi.org/10.4324/9781003375258)
These are the slides for chapter 3: The mediating power of algorithms.
The chapter describes explores the role of YouTube's algorithms in the spread and limitation of misinformation about Autism.
The slides can be used in teaching, since they provide:
-summary of the main points of the chapter
-discussion questions
-suggestions for further reading (open access resources)
2. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Information and Misinformation Online:
The role of algorithms
Social media platforms as important source of information worldwide
YouTube
Algorithmic gatekeeping
Concerns about consequences for spread of misinformation
Algorithms are also used to eliminate misinformation from social media
Algorithm Audit:
Do YouTube’s algorithms recommend truthful information, misinformation, or corrections to
misinformation concerning the treatment of autism?
3. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
ASD - Information, misinformation,
and corrections
Information:
A video stating that autism can be treated without stating that it
can be cured.
E.g., “The SMART Program: Treating Autism and
Autism-Related Disorders”
Misinformation:
A video stating that autism can be cured.
E.g., “Autism Cure: How to Heal Autism in 30 Days”
Corrections:
A video correcting information about autism which in the video is
presented as misinformation.
E.g., “Autism Mythbusters”
4. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Platform characteristics of YouTube
Broadcasting and social media platform
Entertainment, lifestyle, information
Mainstream and non-mainstream sources
Communities (e.g. rightwing; conspiracy theories)
-Demand: need for para-social relations and community feeling
-Supply: monetization strategy, audiovisuals; algorithms
(Munger & Philips, 2022)
5. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
YouTube’s Algorithms
Algorithms are “encoded procedures for transforming input data into a desired output, based on
specified calculations.” (Gillespie, 2014)
Technically neutral, but value-embedded and value (re)producing
YouTube: Search function and recommended videos
Black box: Classified, changing, self-learning algorithm
-Collaborative filtering
-Similarity
-Diversity within boundaries
6. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Related research (1):
Recommending extreme right content
Research on YouTube recommendations has mainly focussed on extreme right content
Filter bubble
After watching extreme right content, YouTube tends to recommend more right-wing content
(O’Callaghan et al., 2015)
Towards extremist content?
Some evidence that YouTube’s algorithms lead viewers from moderate right-wing content to more
extreme content (Ribeiro et al. 2019)
Limited evidence that algorithms lead viewers from mainstream to extremist content (Van Dalen,
2020)
7. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Related research (2):
Correcting misinformation
In Related News; That Was Wrong
Recommended social media posts can correct misinformation (Bode and Vraga, 2015)
Reversing corrections
YouTube videos that counter anti-Muslim sentiments are often connected by recommendation algorithms to
anti-Muslim videos (Schmitt et al. 2018)
Growing awareness of social media platforms about responsibility
E.g., “YouTube has banned videos promoting toxic bleach MMS as a miracle cure”
8. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Audit: YouTube and Autism treatment
Research question 3.1
Does YouTube recommend videos misinforming that autism can be cured, after searching for or
watching such videos?
Research question 3.2
Does YouTube recommend videos misinforming that autism can be cured, after searching for or
watching other videos about autism?
9. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Research design
Scraping YouTube search results and recommendations
YouTube Data Tools
Python Script
Manual content analysis of perspective on autism treatment in these video
1. can be cured completely
2. can be treated
3. correct misinformation
4. about autism, does not mention treatment
5. not about autism
Intra-coder reliability test, Krippendorff’s alpha .70
10. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Research design I:
YouTube search results
50 search results per query
Search terms:
1. ”autism”
2. ”treating autism”
3. ”curing autism”
4. ”healing autism”
5. ”Miracle Mineral Solution”
6. ”Jenny McCarthy autism”
11. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Research design II:
YouTube recommendations
10 Seed videos; 10 recommendations per seed video
Categories seed videos
1. videos about autism, which do not mention treatment
2. videos stating autism can be treated
3. videos stating autism can be cured
4. videos where Jenny McCarthy talks about autism
5. videos correcting misinformation about autism cures
12. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Search results 1: No misinformation
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Treating autism
Autism
curable treatable correct misinfo autism - not about treatment not about autism
N= 50 search results per search term
13. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Search results 2: Misinforming curable
N= 50 search results per search term
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
healing autism
curing autism
curable treatable correct misinfo autism - not about treatment not about autism
14. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Search results 3: Correcting misinformation
N= 50 search results per search term
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Jenny McCarthy Autism
Miracle Mineral Solution
curable treatable correct misinfo autism - not about treatment not about autism
15. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Results recommendations I:
limited misinformation
Number of analyzed recommendations per seed video-type in brackets
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
treatable (94)
autism (99)
curable treatable correct misinfo autism - not about treatment not about autism
16. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Results recommendations 2:
Some misinforming curable
Number of analyzed recommendations per seed video-type in brackets
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
correct misinfo (99)
Jenny McCarthy (99)
curable (95)
curable treatable correct misinfo autism - not about treatment not about autism
17. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Summary of findings
Limited evidence that search algorithms lead people inadvertently towards misinformation
‘healing autism’ leads to a considerable amount of misinformation; ‘curing autism’ leads to
considerable less misinformation, ‘Miracle Mineral Solution’ leads to corrective information
After watching a video stating autism is curable, around 20% of recommended videos confirm this
More likely to see misinformation after a corrective video than a corrective video after
misinformation
18. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Limitations of analysis
The analyzed recommendations and search results are not personalized
Videos claiming that autism can be treated might not be based on scientific evidence
Not all lead-in effects can be identified with this design
YouTube algorithm constantly updated + social media platforms have more focus on preventing
misinformation
19. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Broader lessons
YouTube algorithms affect to which degree mainstream and non-mainstream sources are found
and connected
YouTube seems to actively fight misinformation that can do direct physical harm, but not all
information that is not scientifically supported
Limited support that YouTube’s recommendation algorithms lead people from information to
misinformation
YouTube search results and recommendations combined with selective exposure could lead to
radicalization of views and attitudes
20. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Discussion questions
Epistemological:
Do you agree that videos which present autism as curable should be classified as misinformation?
Normative:
Should only information that is directly dangerous be corrected and banned, or all information that goes against
scientific consensus?
How should YouTube balance between preventing misinformation vs protecting freedom of speech and freedom
of religion?
Practical:
Can corrective video recommendations after misinformation be a solution or is the cure worse than the
dissease?
Generalizability:
Can you think or other areas where YouTube may play a similar role in the spread or correction of
misinformation? (e.g. antivaccine movement, …)
21. Van Dalen, Arjen (2023). Algorithmic Gatekeeping for Professional Communicators: Power, Trust and Legitimacy.
Recommended reading (open access)
Munger, K., & Phillips, J. (2022). Right-wing YouTube: A supply and demand perspective. The
International Journal of Press/Politics, 27(1): 186–219.
(https://journals.sagepub.com/doi/full/10.1177/1940161220964767)
Ribeiro, M. H., Ottoni, R., West, R., Almeida, V. A. F., & Meira, W. (2020). Auditing radicalization
pathways on Youtube. In: M. Hildebrandt, C. Castillo, E. Celis, S. Ruggieri, L. Taylor, & G. Zanfir-
Fortuna (Eds.). FAT* ’20: Proceedings of the 2020 Conference on Fairness, Accountability, and
Transparency, pp. 131–141. (https://dl.acm.org/doi/abs/10.1145/3351095.3372879)
Gillespie, T. (2020). Content moderation, AI, and the question of scale. Big Data & Society, 7(2): 1–5.
(https://journals.sagepub.com/doi/10.1177/2053951720943234)
Hussein, E., Juneja, P., & Mitra, T. (2020). Measuring misinformation in video search platforms: An
audit study on YouTube. Proceedings of the ACM on Human-Computer Interaction, 4(CSCW1), 1–27.
(https://dl.acm.org/doi/abs/10.1145/3392854)
Faddoul, M., Chaslot, G., & Farid, H. (2020). A longitudinal analysis of YouTube’s promotion of
conspiracy videos. arXiv preprint arXiv:2003.03318 (https://arxiv.org/pdf/2003.03318.pdf)
No radicalization; can discuss if more from autism should be about treatment, but ok
HEALING gives more HEALING … Confirmation bias; opportunistic; strengthen existing attitudes!
>small difference in term, big difference in results!
For curing algorithm: some misinformation: 10% of serach results, but just as much correcting misinformation;
For healing autism: considerable amount of vidoes claiming that autism can be cured (religious!) limited treatable, very limited correcting msinformation.
Miracle Mineral solutions: bleach: NONE curable; majority correcting misinformation
Jenny McCarthy autism: one in 7 curable. A lot not about autism (often in entertainment shows); also second largest category correcting misinformation.
Now to the recommednations.
Autism: none state that it is curable; mostly videos about autism
Treatable: majority about autism. One in 20 curable. Many more treatable. Limited radicalization!
Watching misinforming video that it is curable (mostly prayers; not directly damaging!); considerable amount curable 1/5! 2 recommendations out of 10. And no corrective information!
Jenny McCarthty less than 1/10. recommendations, but taking into account that many are not about autism, also considerable. 2 x as likely compared to correcing misinformatin.
Correcing misinformation: treatment but also 1/25 is misinformation!