Please enable JavaScript.
Coggle requires JavaScript to display documents.
AI & ML: Privacy, Security, and Ethics - Coggle Diagram
AI & ML: Privacy, Security, and Ethics
Branch 1: Privacy and Security Challenges
Key Issues
Data Harvesting & Permissive APIs (e.g., secondary profile scraping without consent)
Black-Box Model Opacity (difficulty auditing decision pathways)
Lack of User Control & Hidden Privacy Settings
Scenario 1 — Employee Monitoring & Data Collection
Situation: An employer deploys an AI tracking tool to monitor remote worker keystrokes and webcam feeds to evaluate productivity.
Options:
Option A: Silently run the tool in the background to prevent employees from altering behavior.
Option B (My Choice): Enforce full transparency and data minimization—notify staff, limit collection strictly to working hours, and allow opt-out for sensitive data.
Ethical Support: Kantian Deontology — Employees must be treated as autonomous individuals (ends in themselves), not merely as productivity metrics or passive surveillance targets without consent.
Scenario 2 — Healthcare Data Ingestion
Situation: A medical tech start-up uses patient health records to train an AI diagnostic model.
Options:
Option A: Ingest full patient records to maximize model accuracy, risking identity leaks.
Option B (My Choice): Anonymize datasets before training and mandate strict opt-in consent.
Ethical Support: Rights-Based Ethics — Individuals hold a fundamental right to medical privacy and informational autonomy over their personal health data.
Branch 2: Cybersecurity Advantages
Key Benefits
Real-time Anomaly Detection & Network Intrusion Prevention
Automated Threat Hunting & Rapid Incident Response
Predictive Malware Analysis & Pattern Recognition
Scenario 1 — Automated Threat Response vs. False Positives
Situation: An AI security system detects unusual traffic on a hospital network and can automatically block the IP, but it might accidentally lock out doctors accessing patient files.
Options:
Option A: Allow the AI to automatically shut down suspicious connections without human approval.
Option B (My Choice): Configure the AI to isolate threats temporarily while triggering an immediate alert for a human analyst to verify (Human-in-the-Loop).
Ethical Support: Rule Utilitarianism — Following the rule of mandatory human verification prevents catastrophic disruptions in critical environments, maximizing overall safety and societal utility long-term.
Scenario 2 — Defensive AI Vulnerability Scanning
Situation: An enterprise uses AI port scanning to find zero-day vulnerabilities in third-party vendor software connected to its network.
Options:
Option A: Disclose vulnerabilities publicly immediately to force the vendor to patch it.
Option B (My Choice): Practice Coordinated Vulnerability Disclosure (CVD)—privately report findings to the vendor with a remediation timeline before public release.
Ethical Support: Social Contract Theory — Maintains trust, cooperation, and implied obligations between cybersecurity researchers, software vendors, and end-users.
Branch 3: Cybersecurity Risks
Key Risks
Adversarial Machine Learning & Data Poisoning Attacks
AI-Powered Deepfakes & Hyper-Realistic Phishing Attacks
Model Extraction & IP Theft via API Exploitation
Scenario 1 — Deepfake Social Engineering Attack
Situation: An AI voice-cloning tool is used in an audio deepfake phone call requesting an urgent financial wire transfer from a company executive.
Options:
Option A: Rely solely on AI voice-detection software to automatically screen all incoming executive calls.
Option B (My Choice): Implement a defense-in-depth policy requiring out-of-band human confirmation (multi-factor authentication) for all financial transfers regardless of AI verification.
Ethical Support: Virtue Ethics — Practicing the virtues of prudence, vigilance, and sound judgment rather than relying on technical quick-fixes.
Scenario 2 — Poisoned Training Data in Spam/Phishing Filters
Situation: Attackers intentionally feed biased data to an organization's spam filter so malicious emails bypass detection.
Options:
Option A: Keep the filter running silently while attempting behind-the-scenes patches.
Option B (My Choice): Roll back the model to a validated baseline, alert users to heighten vigilance, and re-train with audited data.
Ethical Support: Ethics of Care — Prioritizes safeguarding the vulnerable user community from harm through active communication and protective action.
Branch 4: Ethical Challenges
Key Challenges
Algorithmic Bias & Discriminatory Proxy Variables (e.g., zip codes reintroducing bias)
Corporate Deflection ("Consumers can figure it out on their own")
Job Displacement & Loss of Human Oversight
Scenario 1 — AI Hiring & Resume Screening
Situation: A company uses an AI hiring tool that favors male applicants due to historical biased training data.
Options:
Option A: Continue using the tool because it speeds up hiring, while promising to tweak algorithms later.
Option B (My Choice): Immediately pause automated scoring, audit the training dataset for proxy variables, and mandate human review for all candidates.
Ethical Support: Rawlsian Justice (Justice as Fairness) — Ensures fairness and equal opportunity by removing systemic bias from selection processes.
Scenario 2 — Autonomous Systems & Crisis Decision Making
Situation: An autonomous vehicle AI must choose between crashing into a barrier (endangering the passenger) or swerving into a pedestrian in an unavoidable collision.
Options:
Option A: Program the vehicle to always prioritize the safety of the vehicle's owner.
Option B (My Choice): Program the vehicle using standardized, transparent safety rules focused on minimizing total harm and maintaining predictable braking behaviors without demographic profiling.
Ethical Support: Act Utilitarianism — Evaluates the specific action to minimize immediate net harm and preserve life across all stakeholders.