期刊名称：Proceedings of the National Academy of Sciences
印刷版ISSN：0027-8424
电子版ISSN：1091-6490
出版年度：2022
卷号：119
期号：1
DOI：10.1073/pnas.2119924119
语种：English
出版社：The National Academy of Sciences of the United States of America
摘要：Twitter defines the current political landscape. Ordinary observations attest to this, and now we have some rigorous evidence. As a feature of modern democracies, Twitter merits close examination; the article “Algorithmic Amplification of Politics on Twitter” (
1) provides exactly that. Their findings should give us all a reason to pause—Twitter’s personalized curation of people’s feed amplifies the impact of rightwing content but not so much leftwing content. The effect replicates in six of seven countries studied.
These findings raise ethical questions about the company’s impact on our democracies. All media influence our political debates, from the invention of the printing press onward. Similar controversies surrounded Facebook (
2–
5). Although the authors work for Twitter, they and the company seem to have allowed the data to speak without overt constraint, as a careful reading of their article and supplement suggest. On the lofty assessment of Twitter’s corporate responsibility, this article will prompt much debate. Holiday dinners just got more interesting.
Susan T. Fiske, Eugene Higgins Professor, Psychology and Public Affairs, Princeton University, has served on IRBs since graduate school, chaired Princeton’s IRB for nine years, and chaired the NASEM consensus report
Proposed Revisions to the Common Rule (
8).
On a more operational level, the ethics of the studies themselves are open to debate. Reasonable people can disagree, as decades of Institutional Review Board (IRB) service have shown me. This editorial identifies some competing perspectives, given some basic principles of human subjects research. Standard assessments take a risk–benefit approach. Let’s stipulate, that with our democracies at stake, studying Twitter’s impact benefits society. We are left with the risks of the research. Assume a series of reasonable people at dinner:
Reasonable Person 1. The research was initiated for a business, not a federally funded institution.
As such, IRB oversight was not required. A business context (product design, user testing) requires no IRB review (no federal support, not seeking generalizable knowledge, just Twitter’s effects on its subscribers). By this argument, the authors were analyzing an existing dataset created for another purpose, so if individual identities are protected, then no IRB review is needed, even if they did work at a university (so the Facebook authors argued) (
6).
RP 2. In the politician study, they are public figures, who forfeit IRB protection. Even regular people arguably make themselves public figures by tweeting.
Journalistic ethics apply to public figures. If you are a public figure providing content (granting an interview, tweeting), you don’t get to decide what the media do with your content.
RP 3. All Twitter users cede control over their posts because
Tweets enter the public domain. Analyzing them is anyone’s unconstrained choice, regardless of IRBs. Of course, analysts can report their findings without regard to the user.
RP 4. But Twitter intervened in people’s information stream. When Twitter initiated personalized feeds for its users, it held back 1% of the accounts, leaving them to be the untreated control group. They then compared the controls to a random sample of 4% of everyone else. When people sign up for the service,
they in principle agree to being manipulated, curated or not. The equivalent of informed consent has been given.
RP 5.
The experimental group, 99% of users, is arguably the default. The control group is the intervention; removing curation is a lack of manipulation, not a human subjects issue, unless the treatment is life-supporting. Twitter is not that.
RP 6. Given the baseline reality that Twitter now manipulates (“personalizes”) people’s feed, customers implicitly accept this when they sign up, but
people may not know their feed is manipulated, if they don’t read the subscriber agreement. Many reasonable people do not (who does?). This revelation was the source of the Facebook study controversy (
6), which was not a human subjects issue but a business transparency issue. The user consent to research should be written in accessible, salient language; companies ignore this at their peril. This study could be viewed that way, but Twitter already has set up an ethics review committee.
RP 7. The question here is whether a reasonable person knows that social media curate their feed. Let’s assume yes. If so, this article’s intervention was denying the curation, hence the “control group” terminology. Twitter removed the curation for business purposes. Now the authors come along and use the data for a different purpose (analyzing the effect of untreated feeds). Not an IRB issue for all the reasons mentioned, but also because
the authors are from the same entity that collected the data.
RP 8.
Anyway, no private citizen’s identity is being revealed more publicly than they choose. Even if other people can access the data from the authors,
this fits open-science norms, as long as the data do not entail personal identity information beyond what subscribers have chosen to reveal on Twitter.
RP 9. After all this, suppose this were classified as human subjects research requiring IRB review. The
minimal-risk standard is risk beyond everyday life. Arguably, whether Twitter personalizes one’s feed or not is an everyday risk, so not a human subjects event requiring additional protection.
RP 10. Of course, just because IRBs are not involved, that does not mean that a business can do whatever they want in the name of research.
The Twitter internal ethics panel needs to be as independent as possible.
RP 11. Even if the research does not violate any rules or norms, social media users will find media manipulation distasteful and discomforting.
They will doubt the intentions of the company (
7).
However, this is the point; this is why the study matters enough to discuss over the holidays and beyond.
Reasonable people can disagree, but research ethics (and business ethics) are matters of consensus, with the guidelines codifying the consensus. My opinion aligns with the argument that this research itself has risks no greater than everyday life and is ethically defensible. Indeed, I would go farther: this kind of research is ethically necessary to protect our democracies.
