A feed creates limited availability. Effects depend on platform, intervention, and measured outcome; hypothetical practices do not establish causal effects.
Questions explored
Why is a feed’s picture limited?
Publication, subscriptions, language, moderation, and ranking select material. The neighborhood with a library is a hypothetical example, not platform data.
Does a recommendation predict truth?
Interest, reliability, and usefulness differ. Meta’s account of June 29, 2023 describes several systems; it is dated self-description, not an independent audit of their current operation.
Does viewing establish agreement?
Display, attention, acceptance, and action are different outcomes. The example of forwarding a post to disagree is hypothetical; each causal transition needs checking.
What did the Facebook experiment establish?
Among 23,377 US participants during the 2020 election period, exposure to politically like-minded sources was reduced by roughly a third. Feed content changed, but no measurable effect appeared on eight preregistered attitudinal measures. This does not establish that feeds have no effects.
What distinguishes the X experiment?
The seven-week US experiment in July–September 2023 found conservative shifts in several opinions when switching from chronological to algorithmic feeds, without comparable reverse effects or significant changes in affective polarization or self-reported partisanship. Platform and intervention are specific.
Does chronology eliminate selection?
No: posting frequency, subscriptions, and login time remain conditions. Plato 275d–e distinguishes text and knowledge; an online author can answer, so the ancient critique does not literally describe every interface.
Does another opinion always moderate attitudes?
No general conclusion follows: Bail’s 2018 Twitter intervention found increased conservatism among Republicans. Mill proposes a normative encounter with strong objections; this is not an experiment or a duty to endure aggression.
Do repeated reports establish prevalence?
No: seven posts may use one clip. This is a hypothetical example; check dates, event independence, and an appropriate comparison base.
How can a route be checked without false balance?
Trace three messages to originals and distinguish confirmation from interpretation. Different logos do not establish independence; a personal trial is not a controlled experiment on political beliefs.
Who can change a digital environment?
Developers, organizations, authors, and users have unequal opportunities. A personal route does not replace demands for transparency, and systemic problems do not cancel available steps. This is the article authors’ practical position.
Key distinctions
Publication, subscriptions, language, moderation, and ranking select material. The neighborhood with a library is a hypothetical example, not platform data. (Article authors: explanatory applications and hypothetical examples.)
Interest, reliability, and usefulness differ. Meta’s account of June 29, 2023 describes several systems; it is dated self-description, not an independent audit of their current operation. (Meta, account dated June 29, 2023; distinctions and limitations discussed by the article authors.)
Display, attention, acceptance, and action are different outcomes. The example of forwarding a post to disagree is hypothetical; each causal transition needs checking. (Article authors: explanatory applications and hypothetical examples.)
Among 23,377 US participants during the 2020 election period, exposure to politically like-minded sources was reduced by roughly a third. Feed content changed, but no measurable effect appeared on eight preregistered attitudinal measures. This does not establish that feeds have no effects. (Brendan Nyhan and colleagues, Nature, 2023; intervention during the 2020 election period.)
The seven-week US experiment in July–September 2023 found conservative shifts in several opinions when switching from chronological to algorithmic feeds, without comparable reverse effects or significant changes in affective polarization or self-reported partisanship. Platform and intervention are specific. (Germain Gauthier and colleagues, Nature, 2026; experiment in July–September 2023.)
No: posting frequency, subscriptions, and login time remain conditions. Plato 275d–e distinguishes text and knowledge; an online author can answer, so the ancient critique does not literally describe every interface. (Plato, Phaedrus, 275d–e; contemporary comparison by the article authors.)
In Bail and colleagues’ 2018 study, exposure to opposing political messages through Twitter increased conservatism among Republicans. This result concerns the selected participants and intervention; it does not establish harm from every encounter with another position. (Christopher Bail and colleagues, PNAS, 2018.)
No: seven posts may use one clip. This is a hypothetical example; check dates, event independence, and an appropriate comparison base. (Article authors: explanatory applications and hypothetical examples.)
Trace three messages to originals and distinguish confirmation from interpretation. Different logos do not establish independence; a personal trial is not a controlled experiment on political beliefs. (Article authors: explanatory applications and hypothetical examples.)
Developers, organizations, authors, and users have unequal opportunities. A personal route does not replace demands for transparency, and systemic problems do not cancel available steps. This is the article authors’ practical position. (Article authors: explanatory applications and hypothetical examples.)
In On Liberty, chapter II, Mill urges readers to learn the grounds of an opposing position. This is a normative philosophical argument, not an experimental assessment of social media. (John Stuart Mill, On Liberty, chapter II.)
Limits of interpretation
- No universal effect of all algorithms is established.
- Display is not belief or action.
- Meta’s account is dated 2023; it is not an independent audit of current versions.
- Conditions of the different field studies are not interchangeable.
- Plato and Mill offer historical philosophical positions; practical scenes are author-created.
- A personal trial is not a controlled experiment.
- The source passages identified below were read, rather than every study supplement and all raw data.
Sources
- How AI powers experiences on Facebook and Instagram — Meta
Accessible explanatory paragraphs dated June 29, 2023; not all cards or code.
Source use: UNKNOWN
- Like-minded sources on Facebook are prevalent but not polarizing — Brendan Nyhan et al.
Abstract and selected design and results paragraphs; not all supplements. Correction checked separately.
Source use: UNKNOWN
- The political effects of X’s feed algorithm — Germain Gauthier et al.
Abstract, opening main text, selected results and the experimental period in the methods; not all supplements.
Source use: UNKNOWN
- Exposure to opposing views on social media can increase political polarization — Christopher A. Bail et al.
Abstract and passages on methods, results, and limitations; not all raw data.
Source use: UNKNOWN
- Phaedrus — Plato / Benjamin Jowett
274c–275e; not the whole dialogue.
Source use: UNKNOWN
- On Liberty — John Stuart Mill
Chapter II: selected paragraph about one’s own and the opposing position; not the whole work.
Source use: UNKNOWN