Latest / ELECTE's Podcast: AI Frontiers / Delete the Tie. The Inference Stays.

Transcript

Delete the Tie. The Inference Stays. · ELECTE's Podcast: AI Frontiers

• 0:03
  SPEAKER_00: This is AI Frontiers.
• 0:06
  SPEAKER_00: Today, the social graph you cannot delete.
• 0:09
  SPEAKER_00: You consent to what you disclose.
• 0:11
  SPEAKER_00: You never consented to the graph everyone else builds around you.
• 0:15
  SPEAKER_00: And what a platform infers from that graph survives anything you delete.
• 0:20
  SPEAKER_00: A social graph is nodes and ties, messages, follows, transactions, shared devices, repeated co-presence.
• 0:28
  SPEAKER_00: Nothing needs to be displayed for it to be priced.
• 0:31
  SPEAKER_00: A relationship the platform never shows, you can still be an attribute it sells.
• 0:35
  SPEAKER_00: The scale is not a footnote.
• 0:37
  SPEAKER_00: Data Reportal counts 5.79 billion social media user identities as of April 2026, up 294 million in a year.
• 0:47
  SPEAKER_00: Identities, not people.
• 0:49
  SPEAKER_00: The typical user is active on about six and a half platforms a month.
• 0:53
  SPEAKER_00: Fragmentation looks like a limit on any one platform's view.
• 0:58
  SPEAKER_00: It is the opposite.
• 0:59
  SPEAKER_00: The more scattered your network, the more valuable the actor that can stitch it together.
• 1:04
  SPEAKER_00: Classical network analysis describes ties that were observed.
• 1:08
  SPEAKER_00: A link prediction model estimates the ties you never made, from shared neighbors, common communities, structural position.
• 1:16
  SPEAKER_00: The inference is produced from an incomplete graph by design.
• 1:20
  SPEAKER_00: So delete the follow, drop the contact, leave the group.
• 1:24
  SPEAKER_00: The model still has your position and it rebuilds the deleted tie from the nodes around you.
• 1:30
  SPEAKER_00: Anonymization fails the same way.
• 1:32
  SPEAKER_00: Strip the names, and a node is still identifiable by the shape of its neighborhood.
• 1:37
  SPEAKER_00: Topology is a signature.
• 1:40
  SPEAKER_00: European courts got here first.
• 1:42
  SPEAKER_00: In OT, 2022, the Court of Justice of the European Union held that publishing the name of a public official's spouse or partner is special category data because it can reveal sexual orientation, quote, by means of an intellectual operation involving comparison or deduction.
• 2:01
  SPEAKER_00: In Meta versus Bundeskartellamt, 2023, the Grand Chamber drew the line at aggregation.
• 2:08
  SPEAKER_00: Combining a user's Facebook data with Instagram, WhatsApp, and third-party sites requires separate consent.
• 2:15
  SPEAKER_00: The law has reached the tie and the aggregation.
• 2:19
  SPEAKER_00: It has not reached the inference about you built from other people's ties.
• 2:23
  SPEAKER_00: The General Data Protection Regulation gives you access to your data and a right to contest automated decisions.
• 2:29
  SPEAKER_00: It does not show you the graph features behind a score or the source of a tie that was never yours to disclose.
• 2:36
  SPEAKER_00: You may learn the outcome.
• 2:38
  SPEAKER_00: You will rarely learn the edge.
• 2:40
  SPEAKER_00: The asymmetry was never that graphs exist.
• 2:44
  SPEAKER_00: It is who gets to hold one.
• 2:46
  SPEAKER_00: That's AI Frontiers.