Ethics of AI Society — Briefing Notes

1. Opening Updates (Pre-Transcript)

Symposium – March 13

• Planning for the March symposium is nearly complete.
• Speaker slots are essentially full.
• The event will conclude with a panel discussion (participants TBD).

2. Report From David Hildebrand — International AI Conference in Paris

David summarized insights from his recent talk in Paris on AI, privacy, and surveillance.

Key Points from Hildebrand’s Report

EU Digital Wallet Initiative

• EU is legally required to roll out government-authorized digital identity wallets by end of 2026.
• Goal: allow citizens to share only the minimal information required for transactions.
• Could improve privacy by reducing oversharing of personal data.

Privacy Requires Transparency

• You can only know you have privacy if you know how and when you’re being watched.
• Modern “hypernudging” and subtle surveillance undermine this awareness.

Surveillance Concerns

• Hildebrand drew on Foucault’s concepts (panopticon, governmentality).
• Citizens begin to self-monitor when uncertain about surveillance.
• Palantir’s presence in Denver illustrates real-world risks of concentrated data power.

3. Discussion: AI, Privacy, and Data Governance

3.1 Privacy as a Philosophical and Practical Problem

• Privacy is complex, evolving with technology.
• AI systems raise questions about:
  • What counts as private information?
  • Whether inferences made without explicit data violate privacy.
  • How prediction and personalization blur public/private boundaries.

3.2 Malicious Use, Malfunction, and Systemic Risks

• Raised via the International AI Safety Report (2026).
• AI systems can perform complex tasks but still fail at basic details (e.g., citations).
• Digital wallets pose risks of:
  • Data breaches
  • Authentication errors
  • Misuse of central infrastructure

3.3 Human Behaviour: Over-Sharing

• Users frequently give AI more data than necessary (e.g., resumes, personal context).
• AI can reinforce confirmation bias by personalizing interactions.

3.4 Data Retention and Deletion

• Hildebrand noted systems like Claude claim short-term memory and non-retention of user data.
• But without public oversight, this relies on corporate trust.
• A public, democratically accountable AI system could allow meaningful audits.

3.5 Privacy as Property?

• Question raised: Should individuals legally own their data?
• Property framing could allow legal recourse if companies retain data beyond agreed terms.
• Example: Matthew McConaughey trademarking his face to prevent deepfake misuse.

3.6 Digital Inclusion Concerns

• Digital wallets assume citizens have smartphones.
• Raises equity issues: should governments provide devices to ensure fair access?

4. Privacy, Democracy, and Deweyan Theory

Privacy as a Democratic Good

• Democracy relies on citizens engaging in inquiry free from coercive oversight.
• Surveillance undermines:
  • Cognitive freedom
  • Experimental thinking
  • Equal footing in public discourse

Asymmetries of Power

• AI systems accumulate unknown quantities of data.
• Citizens cannot easily determine what is known about them.
• This erodes democratic agency.

5. Transition to Discussion on AI and Warfare

5.1 Overview of AI Use Cases in Warfare (Silk)

• Logistics: resource modelling, predictive maintenance
• Intelligence: satellite analysis, surveillance + facial recognition
• Cyber warfare: offense/defense, automated attacks
• Information warfare: propaganda, deepfakes, psychological operations
• Wargaming & simulation: modelling battlefield outcomes
• Autonomous weapons: drones, guided missiles
• Fire control systems: missile defense, high-speed interception

5.2 Psychological & Information Warfare

• AI-powered propaganda may outweigh traditional weapons in impact.
• Deepfakes can destabilize governments.
• “Psyops” may become the primary battlefield.

6. Ethical Issues in AI-Enabled Warfare

6.1 Proportionality & Discrimination

• AI may misclassify targets (e.g., tank vs. hospital with similar shapes).
• Raises classic Just War concerns:
  • Proportionality
  • Civilian protection
  • Combatant–noncombatant discrimination

6.2 Responsibility & Accountability Gaps

• Who is responsible when AI makes a lethal mistake?
• AI diffuses responsibility and removes clear decision chains.
• Danger of leaders hiding behind AI recommendations (“veneer of accountability”).

6.3 Algorithmic Opacity Enhances Existing Problems

• Similar to prior corporate fragmentation of responsibility, but amplified.
• Bias in training data directly affects battlefield decisions.

6.4 Lower Human Cost = Easier to Justify War

• If no soldiers die, domestic resistance to war diminishes.
• Risk of normalizing conflict as a default problem-solving tool.

7. Arms Races, Geopolitics, and the Prisoner’s Dilemma

• If one nation builds AI weapons, others feel compelled to follow.
• Similar to nuclear arms race dynamics.
• China’s technical leap (e.g., IPv6 infrastructure) discussed as a real-world example.

8. Collaboration Between Humans and AI

• Not necessarily AI replacing humans; rather hybrid systems where humans remain involved.
• Example: University of Waterloo admissions use joint human + AI evaluation.
• Multiple independent AI systems plus human oversight may increase reliability.

9. Organizational Matters

• Environmental impacts of AI postponed to next meeting.
• Future topics:
  • Transparent AI
  • Trustworthy AI
  • Environmental cost of AI systems

Next Meeting: February 27, 4:00 PM