Please enable JavaScript.
Coggle requires JavaScript to display documents.
Central Idea: Artificial Intelligence (AI) and Machine Learning (ML) -…
Central Idea: Artificial Intelligence (AI) and Machine Learning (ML)
Privacy and Security Challenges
Scenario 1: Massive Data Scraping for Generative Model Training
Context: A tech company scrapes publicly available personal blogs, images, and posts to train a commercial generative model without explicit user consent.
Option A: Allow unrestricted scraping of public internet data to maximize model performance and innovation.
Option B: Require explicit opt in consent and provide financial compensation or opt out mechanisms for original content creators
Justification: By Utilitarianism and basic property rights. While Option A speeds up development, Option B prevents mass violation of user expectations and long term break of trust. Facts show that web scraping frequently captures sensitive personal identifiable information and copyrighted intellectual property without pathways.
Scenario 2: Employee Monitoring via Facial Recognition & Behavioral Tracking. Context: A corporation installs AI driven webcam monitoring to track employee eye movement, keypresses,and emotional states during remote work hours.
Option A: Haave full AI surveillance to optimize workforce productivity and deter data theft.
Option B: Limit tracking strictly to project deliveries and secure system access logs
Justification: Based on Kantian Deontology monitoring everything treats employees purely as productivity units, violating human dignity.
Cybersecurity Advantages
Scenario 2: Predictive Policing and Algorithmic Cyber Threat Profiling Context: A federal agency uses an ML model to analyze historical network data and flag individuals high risk for committing insider cyber crimes.
Option A: Revoke network privileges and launch surveillance on high risk flagged individuals.
Justification: Supported by Rawlsian Justice and Social Contract Theory, System design must protect the least advantaged and prevent systemic bias.
Option B: Use risk scores solely to trigger enhanced general security audit logging without restricting individual rights or targeted surveillance
Option A: Allow full quarantine of suspicious servers immediately without human oversight
Option B: Force all AI alerts into a manual queue for human analysts to review before taking action.
Justification: Supported by Consequentialism. The consequences of a delay in ransomware outweigh the business disruption of a falsempositive quarantine. Modern automated attacks execute in milliseconds.
Cybersecurity Risks
Scenario 1: Automated Deepfake Audio in Financial Authorization. Context: A finance executive receives a voice call from their CEO requesting an urgent wire transfer to complete a task.
Option A: Rely on traditional voice verification and speed up approval to secure the business deal.
Option B: Mandate multifactor verification for all financial transfers, regardless of sender authority
Justification: Supported by Virtue Ethics, relying on visual cues is no longer reasonable. Implementing strict, non negotiable verification protocols is a safeguarding asset.
Scenario 2: Adversarial Data Poisoning of Open Source ML Models. Context: A cybersecurity startup uses an open ource ML training dataset.
Option A: Integrate open source datasets without verification to ship features faster than competitors.
Option B: Invest resources into zero trust dataset auditing.
Justification: Supported by Risk Ethics & Care Ethics. Organizations have a moral duty of care to their downstream users to maintain software supply chain integrity.
Ethical Challenges
Scenario 1: Algorithmic Bias in Automated Hiring and Resume Screening Context: HR uses an ML algorithm trained on 10 years of past hiring data. The model consistently ranks male candidates higher for software engineering roles due to historic hiring demographics.
Option A: Continue using the algorithm unchanged because it reflects real historical company success metrics.
Option B: Audit the algorithm, retrain the model with new data, and require human oversight on all rejections
Justification: Based on Justice as Fairness. Past discrimination in training data has inequality if left unchecked. Fairness requires actively removing systemic bias from automated gatekeepers to ensure equal opportunity.
Scenario 2: Algorithmic Transparency in AI Driven Medical Diagnoses. Context: A deep learning model diagnoses rare cancers with 98% accuracy but it cannot explain how it reached its diagnosis.
Option A: Deploy the black box model immediately because its raw accuracy rate outperforms human doctors.
Option B: Deploy the model only as a secondary advisory tool alongside AI features, requiring final diagnostic by a qualified physician
Justification: Supported by Principalism patients and doctors have a right to informed consent and understandable medical rationale. Relying entirely on an unexplainable model removes medical accountability and deprives patients of meaningful participation in their own healthcare decisions.