LAWS10009_2021_SM2

For AI, Ethics and Law subject taught at UOM

11 flashcards · Law · by Rachaya Youngrod

1. AI Ethical Principles

   Human, societal and environmental wellbeing: benefit individuals, society and the environment Human-centred values: respect human rights, diversity, and individual autonomy Fairness: inclusive and accessible, and should not involve or result in unfair discrimination against individuals, communities or groups. Privacy protection and security: respect and uphold privacy rights and data protection, and ensure the security of data. Reliability and safety: reliably operate in accordance to the inten…

2. Technology behind FRT

   Neural Networks replicates the human brain's using layers of CPU to process a problem, converting 2d info to complicated 3d lapses. Trained by giving a vast library of images to be used as a basis of comparison to identify patterns It distributes given information across the network Each array makes a decision statistically on what may be correct. and passes on the probability to the next array Accumulation to the highest probability outcomes. Network repeats task and integrating previous error…

3. Current limitations of FRT

   Reliability Biased training data (epistemic) Discrimination Privacy FRT is already used by the police in numerous cities, though this technology also is notable for having extremely high error rates. eg. COMPAS is an AI used by law enforcement to estimate people’s risk of recidivism. Even without the additional facial recognition technology, COMPAS is unfortunately biased against Black people and it has led to instances of racial and intersectional discrimination against Black defendants. (1) W…

4. Case Study: LoopLearn

   Face identification and automatic attendance taking Is it beneficial to have the children be under a constant surveillance environment? Student's contestability

5. Case study: Clear View AI

   FRT scans the face of a person and the algorithm identifies the person's face from social media (scraping) Reliability: easy to be manipulated by criminals, people just cover their faces + can mislead people Accuracy poses risk; false positives (eg. reputational risk to individual when) Inferred consent to the scraping, Black box algorithm bc it's commercial: is the tool worth the privacy invasion?

6. FRT in law

   Public/Govt: exercise of power eg. R(Bridges) v CC South Wales Police Technology tries to identify sus individuals under watchlist Insufficient guidance on its use No reasonable steps to ensure it was not biased (on race & sex) Private/commercial: data protection/privacy law

7. Can FRT be used as a weapon? How can facial recognition technologies take advantage of the vulnerabilities of mob behaviour on social media?

   “Hated in the Nation”, the sixth episode of Black Mirror’s third season, imagines a dystopian society wherein online hate produces deadly consequences, namely mass murder enabled by artificial intelligence (AI). The ep. touches upon the weaponisation of online outrage and the possibility of FRTs taking advantage of the vulnerabilities of mob behaviour on social media. Mass surveillance of people is a central ethical concern countries are on a trajectory to becoming despotic surveillance states…

8. Case Study: COMPAS

   Predictive policing for reoffending used in the US (recidivism) Rate of false positive is debatable, around 60% accurate but that can be same accuracy as a judge PredPol doesn’t distinguish the level of crime — bias

9. Ethics of Predictive Technology

   Allows for disproportionate policing from data bias Even if the algorithm isn’t biased, the ethics may not be agreeable (is differetial treatment justified?) Not enough transparency, not knowing the chain of reasoning (factor weighting may be biased) In reasoning, system can be taken advantage of due to data errors

10. Case Study: State vs Loomis https://harvardlawreview.org/2017/03/state-v-loomis/

    Loomis was sentenced to 6 years due to COMPAS, not 1 Argues transparency and that it was biased Supreme court rejected arguement bc COMPAS score is accessed correctly similar to a judge’s Conclusion: court should exercise their own judgement, should be tool for judgement not sole basis

11. Case Study: Suspect Management Program in NSW

    Algorithm based program that predicts recidivism based on one’s behavior and environment Seems to have been used disproportionately against aboriginal young people result of criminalizing people, not pulling them out of crime might encourage unlawful police conduct