◆ Semantic Web Catalogue

Problem thread

Can the Web forget?

Should what is published about a person stay findable forever, or can they have it removed?

issue
governance
also sustainability, topology
status
open
steps
8, 1998–2019
threads
all problem threads

The Web's builders treated permanence as a virtue: cool URIs don't change, and the Wayback Machine keeps copies of everything it can. Viktor Mayer-Schönberger argued in 2009 that perfect memory harms people, and in 2014 the EU's highest court gave people the right to have search engines drop links about them; the GDPR made erasure a general right. Forgetting is hardest where the semantic web's own ideas apply: copies spread across federated servers, archives and content-addressed networks, and now into models trained on them. Meanwhile platforms forget by accident, as Myspace did.

Lineage

  1. 1998
    precedent writing ★ seminal
    Cool URIs don't change

    Berners-Lee argues that a URI, once published, should never change; permanence becomes a mark of good practice. [src]

  2. 25 Oct 2001
    proposes project ★ seminal
    Wayback Machine

    The Internet Archive keeps copies of web pages so that what was published stays readable after it changes or disappears. [src]

  3. 2009
    poses writing ★ seminal
    Delete: The Virtue of Forgetting in the Digital Age

    Mayer-Schönberger argues that digital memory has made forgetting the exception, that this harms people who cannot outgrow their past, and proposes expiration dates for information. [src]

  4. 13 May 2014
    proposes writing ★ seminal
    Google Spain v AEPD and Mario Costeja González

    The EU Court of Justice rules that a search engine must, on request, remove links to information about a person that is inadequate, irrelevant or no longer relevant, unless the public interest outweighs it. [src]

  5. May 2015
    proposes writing
    Towards Making Systems Forget with Machine Unlearning

    Cao and Yang propose making learned systems forget particular training data without retraining from scratch. [src]

  6. 14 Apr 2016
    proposes project
    General Data Protection Regulation

    Article 17 gives people a right to erasure, which the regulation also calls the right to be forgotten, against anyone holding their personal data. [src]

  7. 23 Jan 2018
    proposes standard ★ seminal
    ActivityPub

    A Delete activity asks every server holding a copy to remove it; receiving servers should comply, but nothing can make them. [src]

  8. 18 Mar 2019
    measures project
    Myspace

    Forgetting also happens by accident: Myspace confirms that about 50 million songs uploaded before 2015 were lost in a server migration. [src]

Open now

Under each question, the answers people have put forward or built. The catalogue gives no odds on them.

  • Can a copy that has spread across servers, archives and peer-to-peer networks be made to forget? [src1][src2]
    Federated and content-addressed systems copy data by design; removal depends on every holder agreeing.
  • Can a language model forget what it was trained on? [src1][src2]
    Erasure rights apply to personal data wherever it is held; a trained model holds its data only implicitly.
  • Where does a person's right to be forgotten end and the public record begin? [src]
    The 2014 judgment weighs the person's rights against the public's interest, for example when the person plays a role in public life.

Forces

Pressures that push this problem one way or another, as their sources argue them. All forces

How to contribute