Future
Society Hubwhere tech and society fuseFuSo logo
01 / 12›
Computer Science x Society: study Computer Science at the University of St.Gallen. Apply now for Bachelor and Master. bcs.unisg.ch

AI +
Social Media

Aurelia Tamò-Larrieux, Simon Mayer, Giovanni De ToniQR code linking to https://futuresociety.ch/ai+xSlidesfuturesociety.ch/ai+x
The information environment‹02 / 12›

AI shapes how information
reaches us.

Web search

Ranking & retrieval

Social media

Recommendations & moderation

Chatbots

Language models & generation

AI & MACHINE LEARNINGSelectRankGenerateExplain
Each channel affects what we see and how we understand it.These channels coexist and interact
Our lenses‹03 / 12›

Making sense
of this change.

01 TECHNICAL
The systems and their functions
Functions & features
02 SOCIAL
Effects on people and public discourse
Opportunities & risks
03 REGULATORY
Responses to these effects
Evidence & intervention
Technology × Society
01 / Technical lens‹04 / 12›

AI acts at several points
in the same information environment.

Selection
↗

Deciding what reaches you

Recommenders choose and rank posts, videos, and accounts.

Generation
✳

Producing content

AI helps create text, images, voices, and video.

Conversation
“ ”

Answering and explaining

Chatbots synthesize answers and respond to follow-up questions.

AI + Social Media
02 / Social lens‹05 / 12›

Effects on society

Possibilities

  • Access to information
  • Participation and expression
  • Finding communities and unfamiliar views

Questions to investigate

  • Whose voices become visible?
  • What earns our trust?
  • Do we encounter different perspectives?
Do these tools expand our ability to form our own judgments?
03 / Regulatory lens‹06 / 12›

Countries regulate platforms
in different ways.

EUR-Lex, European UnionDigital Services Act (2022)Assess and mitigate systemic risksThe EU framework for very large platforms and search engines.Read the regulationeSafety Commissioner, AustraliaEffective 10 December 2025Age restrictions for social media accountsCovered services must take reasonable steps to prevent accounts for under-16s.Read the guidanceState Council, China31 August 2021Time limits for minors' online gamingA direct restriction on access and use, in a different legal setting.Read the report
Each card links to its original sourceApproaches are not legally equivalent
03 / Regulatory lens: recent developments‹07 / 12›

Europe's rules are evolving.

European Commission: policy proposal17 September 2026EU KIDS Act: age rules and safety by designThe proposal addresses social media and other services children use, including AI chatbots.PROPOSAL, NOT YET LAWOpen the proposalEuropean Commission: DSA designation31 August 2026ChatGPT designated a VLOSEA very large online search engine designation; Reddit and Roblox were designated VLOPs.Read the announcement

How do we assess systemic risks when information arrives as an answer?

From questions to evidence‹08 / 12›

Bring the lenses
closer together.

Observe and measure what these systems do.
Decide which social outcomes matter.
Turn legal duties into changes we can test.
You're invited to the Solid Symposium 2027‹09 / 12›

Control over the data AI acts on

AI assistants and agents increasingly read and act on our personal data. Solid keeps that data in a Pod its owner controls, and lets each application, AI included, use it only with permission that can be revoked. Come and discuss it with us at the Solid Symposium in St.Gallen.

  • Your data lives in a Pod you choose, separate from the AI that uses it: self-hosted, at your institution or with a provider you trust.
  • Each application gets access through open W3C standards, and you can revoke it at any time.
  • You can switch assistant or Pod provider without losing your data.
SQUARE building on the University of St.Gallen campusSoSy 27
Join us in St.GallenThe 5th Solid Symposium, 10–12 May 2027
Mon 10 May
Solid Open Day on digital sovereignty, open to all
Tue 11 & Wed 12
Symposium: papers, workshops, posters, demos

SQUARE, University of St.Gallen

QR code linking to https://sosy2027.org/

Call for workshops opens soonPapers, posters and demos: call opens this autumn.sosy2027.org
Today's session: 2.5 hours‹10 / 12›

The programme.

  1. 10′
    IntroductionAurelia Tamò-Larrieux and Simon Mayer (University of St.Gallen), Giovanni De Toni (ETH AI Center)
  2. 35′
    Opening keynote: Algorithmic transparency in AI + Social MediaJoão Vinagre, European Centre for Algorithmic Transparency (European Commission JRC)
  3. 30′
    Lightning talks: techno-legal approaches
    • Auditing recommender LLMsSimon Mayer + Luka BekavacUniversity of St.Gallen
    • Systemic-risk reporting under the DSAAlice PalmieriUniversity of St.Gallen
    • AI labelling on social mediaHenry TariUniversity of St.Gallen
    • AI in regulatory processesJake SteinUniversity of Oxford
    • Third-party tracking in social media appsDick BlankvoortUniversity of St.Gallen
    • Product recommendations by LLMsLucas Uberti-Bona MarinMaastricht University
  4. 15′
    Lightning talks: Q&A with the speakers
  5. 45′
    Panel discussion: the impact of recommender systemsGiovanni De Toni (ETH AI Center), Jing Zeng (University of Zurich), Angela Müller (AlgorithmWatch CH), João Vinagre (European Commission JRC)
  6. 15′
    Closing
Future Society Hub logoFuture Society Hub
University of St. Gallen
‹11 / 12›

Opening keynote

Algorithmic transparency
in AI + Social Media
Dr. João Vinagre
European Centre for Algorithmic Transparency
European Commission Joint Research Centre
Thank you. João, the floor is yours.futuresociety.ch
Future Society Hub logoFuture Society Hub
University of St. Gallen
‹12 / 12

Panel discussion

The impact of recommender systems
Questions welcomeRaise your hand to ask the panel.
01Social effects of recommender systems
02Optimising recommender systems for better outcomes
Giovanni De Toni
Dr. Giovanni De Toni
Postdoctoral Researcher
ETH AI Center
Best Paper, ACM RecSys 2025
  1. 01Harm and "unwantedness" are measured through proxy metrics, and imperfect proxies can optimise for the wrong end.
  2. 02Give users real control over their feeds, with controls whose effects on the feed are clear and measurable.
Jing Zeng
Prof. Dr. Jing Zeng
Assistant Professor
Computational Communication Science
University of Zurich
  1. 01TikTok's recommender reshapes creators and content production, and its effects are not always negative.
  2. 02Optimising for visibility and engagement puts disproportionate costs on marginalised communities.
Angela Müller
Dr. Angela Müller
Executive Director
AlgorithmWatch CH
Member, Federal Media Commission
  1. 01Recommender systems shape our information environment and democracy: access to reliable information and constructive discourse, now multiplied by AI.
  2. 02Close transparency and accountability gaps, confront power concentration in Big Tech, and fill regulatory gaps.
João Vinagre
Dr. João Vinagre
Researcher
European Centre for Algorithmic Transparency
European Commission JRC
  1. 01Recommender systems can harm fundamental rights, civic discourse and electoral processes.
  2. 02Researcher data access helps build a safer, more predictable and trusted online environment.
45 minutesPhoto A. Müller: David Bächtold / AlgorithmWatch, CC BY 4.0