Briefing Notes – Ethics of AI Society Meeting

1.1 New Website Launch

The group now has an official website hosting:

• Meeting archives
• Media contributions
• Space for potential short articles on AI ethics

Possibility discussed of future newsletters or publications if volunteers join for editing and writing.

1.2 Meeting Milestone

This session marks the 20th official meeting of the Ethics of AI Society.

1.3 Introductions

Several new attendees introduced themselves, including graduate students, faculty, engineers, and returning members.

Topics of interest among attendees included:

• Medical AI
• Pragmatism and philosophy of language
• Digital communications
• Racial profiling and AI
• System design and engineering
• Logic frameworks and Peircean graphs

2.1 Logistics & Needs

Potential need for:

• A door volunteer to manage refreshments if they must remain outside the room.
• A tech volunteer to assist with microphones and setup.
• An additional microphone for audience audio capture during Q&A.

2.2 Keynote Panel

Will include:

• Ian McDonald
• Patricia Marino (Philosophy)
• Ontario MPP Rob Cerjanec

Members are invited to submit potential questions for the panel via a dedicated Teams chat.

Questions may relate to ethics, policy, or public impacts of AI.

3.1 Opening Framing

AI systems require:

• Large electrical power for training and inference
• Significant water usage for datacenter cooling
• Rare minerals for hardware manufacturing

Key overarching question: Is the environmental cost of AI “worth it”?

3.2 Main Themes Raised

A. Urgency of Climate Crisis

Some participants argued that accelerating AI development worsens environmental degradation during an already critical climate period.

Concerns about society’s ability to adapt regulations and policies quickly enough.

B. Individual vs Corporate Responsibility

Debate on whether responsibility should lie more with:

• Tech companies (for creating energy-intensive systems)
• Users (for unnecessary usage)

Several members argued that blaming individuals is oversimplified; structural and corporate choices have far more impact.

C. Question of “Necessary” vs “Excessive” AI Use

No clear societal standards exist for what qualifies as necessary usage.

Examples:

• Medical decision support may justify resource expenditure.
• Homework help and chatbots may not.

D. Efficiency Improvements Over Time

Anticipated advancements:

• More efficient training
• Better cooling methods
• Smaller, more efficient chip architectures

But: Greater efficiency may lead to more total usage, offsetting gains (rebound effect).

E. Geographic and Ethical Implications

Data centers often placed in:

• Areas with scarce water
• Developing countries with lax environmental regulations

Raises justice concerns about disproportionate harms.

F. Competition and Innovation Pressures

Limiting the number of AI models / supercomputers could reduce short-term environmental cost.

But competition drives companies to develop more efficient systems over time.

G. Possible Alternative Computing Paradigms

Mention of human brain organoid computing as a potential ultra-low-power alternative (25 watts), though extremely speculative and ethically complex.

4.1 Case Prompting Discussion

Recent news:
BC government criticized OpenAI for not notifying authorities after concerning user conversations from an individual later involved in the Tumbler Ridge shooting.

Raises core question:
When should AI developers report harmful content to authorities?

4.2 Concerns & Tradeoffs

A. Privacy vs Public Safety

Reporting too frequently risks:

• Mass privacy violations
• Wasted police resources
• Chilling effects on legitimate use

Reporting too rarely risks:

• Failure to prevent violence
• Public distrust of AI companies

B. Difficulty Setting a “Reporting Threshold”

Who decides when content is dangerous?

AI-detected red flags often generate false positives.

Even humans reviewing flagged data may disagree on risk severity.

C. The Problem of Evasion

Users can circumvent safeguards:

• Rephrasing questions
• Asking in code (e.g., Morse)
• Using reverse-engineering prompts
• Running offline open-source models

Therefore, even perfect reporting policies cannot catch all cases.

D. Mental Health–Related Prompts

AI may be used as a “sanctuary” for distressed users.

Overreporting suicide-related prompts could:

• Flag far too many cases
• Reduce public trust
• Push vulnerable people away from seeking help

E. Anthropomorphizing AI

Chatbots often appear friendly or supportive.

Members raised concerns that:

• Politeness consumes resources
• It reinforces emotional dependence
• It encourages confirmation bias

Suggestions:

• Stronger disclaimers reminding users the AI is not human
• More neutral tone to reduce emotional bonding
• Reduced sycophantic behavior

F. Government vs Corporate Pressure

Canadian and U.S. governments pushing in opposite directions:

• Canada wants more safety & reporting.
• U.S. agencies want fewer restrictions (e.g., Pentagon pressuring Anthropic; Grok being used in secure networks).