9: AI & Media Flashcards
(17 cards)
What is founding theory of the freedom of expression?
Common ground: democracy:
* promotion of knowledge and understanding
* responsability for decision-making lies with the public => public should be sufficiently informed
* majority decides, may restrict their freedom of speech => need for constitutional protection, necessary constraints to unlimited democracy
* freedom of expression is indispensible for democracy => freedom to criticise and oppose decisions of the majority and to work towards changing the majority opinion
* freedom of expression serves democratic interest of all citizens:
-> speakers: right to self-development, right to criticise government
-> recipients: right to be informed about all representative views in society
What makes the right to freedom of expression so special?
The excercise of this right is indispensible for democracy and democratic processes to function properly
What freedoms are in the freedom of expression?
- freedom to hold opinions
- freedom to express opinions
- freedom to distribute information/ideas
- freedom to seek and receive information and ideas
=> medium-independent: covers traditional and new media
What is in article 10 of ECHR on freedom of expression?
- 2 dimensions:
1) negative obligation: right of the citizen to be protected against government interference, not absolute: restrictions are possible under certain conditions
2) positive obligation: active state intervention may be necessary to secure the rights and the freedoms of the convention to everyone, obligation to act, protect the right of interference from others - scope: even shocking, disturbing and offending ideas
- limitation of scope/forms of speech that are not protected: cannot abuse the rights Ex. statements denying the holocaust, justifying pro-nazi policy, linking all muslims with terrorism, …
- restrictions: expression is NOT and absolute freedom: restrictions necessary to protect other fundamental human rights and public/private interests such as
-> combating illegal and harmful speech
-> protecting weaker groups in society such as minors/consumers/data subjects
Ex. episode Peppa pig that explains that spiders are not dangerous is not aired in Australia
-> ensuring pluralism and diversity
When can protected speech be restricted?
3 step test:
1) prescribed by law
2) legitimate aim:
-> interest of the state: national security, public safety
Ex. prohibition to reveal state secrets
-> interest of the population: protection of health and morals
Ex. protection of minors
-> social or general interests: prevention of crime
Ex. ban on hate speech
-> private interests: protection of the reputation or rights of others
3) necessary in democratic society: there should be a pressing social need, restriction is pertinent and proportional:
-> effectiveness
-> necessity test: least intrusive means
-> balance of interests
=> interference is justified
Why is it difficult to regulate speech online?
1) different categories of information in the same space:
* illegal content (which is harmful)
* harmful content (but not necessarily illegal)
* critical content (protected but often targeted by politicians for ex)
* shocking/offensive/disturbing content
2) same information different intentions
* misinformation (plain wrong)
* disinformation (deliberately misinforming)
* malinformation (based on facts but removed from its context to mislead or manipulate)
What are the problems with freedom of expression?
- content moderation
- generative AI
What is content moderation?
Systems that classify user-generated content based on matching or prediction leading to a decision (removal/restriction of visibility/account takedown)
What is the problem between FoE and content moderation?
over-blocking: unjustified removal or blocking content or the suspension or termination of user accounts = over removal of lawful content
Risks of this for freedom of expression:
* lack of quality data and bias in training sets: frequent overreaction to identity terms, difficulty differentiating discussion about race from hate speech
* context-dependency (offensive vs satire): AI might not effectively distinguish between lawful and unlawful content due to a lack of definition for harmful/hateful/offensive speech, AI might not recognize underlying context and nuances
* difficult for humans to make timely calls
Free-speech absolutism as solution?
No,
* under-blocking: under-removal of harmful/illegal content
* underrepresentation of certain groups in trainingset
* chilling effect: a severe interference (high fine/imprisonement) can lead to self-imposed restrictions with detrimental effect towards the exercise of one’s rights in the future, self-censorship
What is the problem between FoE and ai?
- detail error/attribution error: who did what
- unrealiable data = spread of misinformation
- automation bias
What is the solution for FoE and AI
Digital services act
What is the digital services act?
Goal: defending European values in online space
=> create a safer digital space
=> to ensure the fundamental rights of all users of digital services are protected
Who is affected by the DSA?
1) very large online platform: Fb, snapchat, amazon, alibaba => not in EU but a lot of EU users
2) online platform: online marketplaces, app stores, social media
3) hosting services: cloud and web hosting services
4) intermediary services: internet access providers and domain name registrars
What are the obligations for very large online platforms (vlop’s)?
1) liability
* no obligation to actively search for illegal content
* obligation to inform law enforcement when platforms find serious illegal content
* no rules on legality of content
2) due diligence: adress harmful content
3) obligations: systemic risk assessment => 4 categories:
* dissemination of illegal content
* negative effects on fundamental rights
* negative effects on civic discourse and electoral processes
* negative effects on gender-based violence, minors, individuals’ physical and mental well-being
=> audit and reports
4) more transparency and accountability: audits and give access to data to researchers
do the systemic risk reports work?
- long PDFs
- lack of standardization
- many reports lack verifiable data to substantiate claims about the effectiveness of mitigation measures
- confidentiality restrictions
How does DSA address genAI?
- DSA does not regulate AI but intermediary services
- Gen AI applications that are self-standing products
- gen AI which are part of VLOPs services are covered
What is the scope of the AI act?
- EU + outside EU if providers place AI-system on EU market
- users of AI-systems within EU
- providers and users outside the EU where the output of the system is used inside EU
