All Intervention Categories (14)

While there are a near-infinite number of interventions that platforms could undertake to reduce abuse on their services, these tend to fit into a small number of common themes that we call "Intervention Categories". Developing an understanding of these fundamental building blocks (through which interventions and regulation can be built) has a number of advantages over focusing on specific interventions:

  1. Estimation: Intervention Categories have common strengths and weaknesses, so they're great abstract tools for considering the potential of given approach for mitigating a given harm.
  2. Future-Facing: Because the techniques are useful across forms of abuse, they can be used as a starting point for figuring out how to tackle emergent harms.
  3. Composition: Since they can be used together, sophisticated organizations will often use heavily hybridized approaches, which can be hard to understand if not explained in terms of these approach-based ideas.

We write at length about each of these "Intervention Categories" because they represent the most helpful piece of the puzzle: not just ideas for how to build platforms that are more resilient today, but a Swiss army knife of strategies that engineers, product managers, and regulators can keep in mind as they navigate today's harms and plan for those of tomorrow.

Graduated Features
Give users access to more powerful features as they engage more deeply.
Information Diversification
Broadening a user's information diet through deliberate mechanisms of alternative exposure.
User Control
Giving users capability controls over a circumscribed perimeter.
Speedbump
A temporary obstacle created to make undesired behavior less frequent, through difficulty or reconsideration.
Affinity
Users' past interactions should shape their future capabilities toward one another.
Ordering
Impact and perception hinge on what's seen first.
Omission
Less is more - a better platform can be achieved by removing pieces of it.
Limits
Preventing anomalous behavior by setting boundaries that outline expected user behavior.
Gatekeeping
Limiting participation based on user history.
Humans Only
Actively prevent automated use of a feature.
Content Analysis
Treating content differently based on its evaluation by automated classifiers.
Abdication
Designing a product feature so that it cannot be moderated.
Escalating Penalty
Treat patterns of behavior as more serious than their disaggregated constituent elements.
History
Utilizing descriptive data to help users understand content in context.
Is something missing, or could it be better?Suggest a new Intervention Category
About This SiteGet InvolvedContactCopyrightPrivacy
Loading...