Law and Technology- fundamental rights and AI Flashcards
what is the first AI system?
Theseus- 1950
what is the name of the AI system developed by Google?
PaLM- explain jokes
which domains now use AI?
AI is now used when: booking a flight; assisting the pilot when flying; deciding whether you get a loan; governments use AI systems for surveillance and oppression
what is the name of the AI’s which determine what you see on social media?
recommender systems. they also determine which videos you get recommended on youtube
currently, what has the doubling time of training computation shortened to?
6 months
in Daniel Solove’s taxonomy what are the 6 privacy conceptualisations?
1) privacy is the right to be let alone, free from state interference
2) privacy is limited access to the self
3) privacy is secrecy
4) privacy is people’s control over their personal information
5) privacy is the right to protect one’s personhood
6) privacy is intimacy
where is the right to privacy recognised?
art 12 UDHR
art 17 ICCPR
art 7 EU Charter of Fundamental Rights
art 8 ECHR
what obligations does the right to privacy impose on the state?
positive and negative obligations- the state must refrain from infringing individuals privacy and ensure that privacy is protected from violations.
why was data protection developed?
as a response to the rise of data processors in the 70’s
what is data protection?
protection of information of an identifiable natural person
who are the 2 key actors in cases of infringement?
- controller- must comply with all GDPR principles.
- data processor- entrusted with processing data on behalf of the controller
what are the main principles related to the processing of personal data under the GDPR which controllers and data processors must keep in mind?
lawfulness
fairness
transparency
accuracy
integrity
confidentiality
accountability
what does the fairness requirement imply?
personal data should not be processed in a way that goes against the data subjects’ reasonable expectations and adversely impacts them. complying with the fairness requirement implies that controllers ensure that AI models are statistically accurate
what does the transparent requirement mean?
data subjects have the right to be informed
what is the problem with transparency rights?
data subjects have the right to obtain information. however, providing this information may be complicated and processors may be unable to explain. this issue is the black box phenomenon.
what is purpose limitation?
data can only be processed for the purposes explicitly specified to data subjects
what is the data minimisation principle?
personal data must be adequate and relevant for the purposes specified
what is the storage limitation principle?
it imposes on the controller and processor the obligation to store personal data for as long as necessary. the data must be erased once the model is fully trained
what is the accuracy requirement?
it creates a right to a correct representation of oneself. so the personal data must be correct and up to date. this also means that data subjects have the right to erasure and rectification
what are the 3 additional obligations imposed on controllers and processors?
they must ensure that the data processing is done with security, integrity and confidentiality. they have the duty not to disclose private information.
what is the accountability requirement?
it is a ‘meta-principle’, which renders data collectors responsible for demonstrating compliance with all the principles
what has gained increased relevance due to the unpredictability of AI applications?
Data Protection Impact Assessments- they incentivise companies to take accountability for the risks that their systems might impose
what lies at the core of AI development?
privacy and data considerations. for AI to be compliant with the law, these considerations cannot be disregarded
what can the digitisation of social protection programmes be traced back to?
e-health programmes in the 1990’s
what are ADMS?
innovative tools that can bolster welfare systems with shrinking budgets
what does integrating ADMS in the welfare workflow require?
the recruitment of profiles that are not typically found in the public sector, such as data scientists
what does ADMS perform?
automated administrative processes
how can ADMS affect the citizens?
it can affect their rights and poor people can get state support to build a decent life
what are the 3 components of ADMS that are relevant for the delivery of equitable welfare services to women?
1) datasets- gendered datasets are lacking, meaning the possibility to serve female citizens becomes undermined. Biased datasets produce biased predictions.
2) decision-making models
3) design- public services targeting women are not designed with their involvement, which can contribute to a negative user experience
what is a negative of ADMS?
few case studies about ADMS examine gender inequalities
what % of women live in poverty compared to men?
20% women
18% men
what % of single parents are female?
90%
what is the exception to the 2-child policy?
the ‘‘rape clause’’- the mother must prove to the government that the child was born out of rape.
what is a negative of the rape clause?
it does not apply if the mother continues to live with her rapist, which fails to recognise that most rape occurs within abusive relationships
what does the Equal Treatment Act forbid?
unequal treatment based on socio-economic variables such as gender
from 2016-2017, what % of single parent families were headed by women?
82%
what is ParentsNext?
a programme for parents already receiving state support, designed to help them achieve education and employment goals
what do the 3 case studies present?
different forms of automated inequality
what are the 3 main pitfalls of the case studies?
1) a faulty approach to data
2) a lack of gender impact assessment
3) the absence of co-design
what is principle 1 of the Gender Equality Principles for the Digital Welfare State?
gender relevant datasets and statistics. algorithms reproduce their creators’ biases. if gender data is missing, the algorithm cannot be blamed for discriminating. building a representative dataset includes contextualising data within broader socio-economic realities
what is principle 2 of the Gender Equality Principles for the Digital Welfare State?
gender mainstreaming in planning.
what is principle 3 of the Gender Equality Principles for the Digital Welfare State?
co-design, oversight and feedback.
what is principle 4 of the Gender Equality Principles for the Digital Welfare State?
equality by default. the system must consider data that is representative of the situation.
what has digital technology enabled?
pathways for false information to be created and disseminated at a scale and reach never known before