Briefing Notes – Ethics of AI Society Meeting

Symposium Planning

• Funding raised through donations; additional funds welcomed via GoFundMe.
• Goal: Provide refreshments and potentially secure a keynote speaker.
• Estimated 7 presentation slots; ~5 confirmed presenters.
• Seeking participants from Laurier, especially philosophy faculty or students.
• Presentations:
  • 20-minute talk + 10-minute Q&A
  • Slides permitted
  • Recordings available on request

Teams Group Access

• New members encouraged to ensure access.
• Contact Matthew for troubleshooting.

A diverse group across Waterloo and Laurier attended, including:

• Philosophy instructors
• Undergraduate/graduate students
• Professionals in AI ethics, research, engineering, data science
• Returning members and new participants

Notable expertise:

• Ian MacDonald – specialist in Charles Sanders Peirce
• Patrick Matlock – built ML model to transcribe Peirce’s handwriting (81% accuracy)

3.1 Background: Peirce’s Theory of Signs

Presented by Ian MacDonald

Peirce distinguishes three sign types:

• Icons – resemble their object
• Indices – physically connected to their object
• Symbols – conventional representations (e.g., words)

AI primarily manipulates symbols, lacking robust handling of indices or icons, limiting grounding in real-world meaning.

3.2 Peirce’s View of Inquiry

Key elements:

• Inquiry moves from doubt to belief.
• Beliefs inform habits → influence practices.
• Meaning determined by the conceivable practical effects of a concept.
• Humans possess real-world contact; AI does not.

3.3 Issues Raised with Respect to AI

(1) Lack of Real-World Grounding

• Statistical understanding, not embodied experience.
• Cannot test ideas against reality.

(2) Logical Fragility / Hallucinations

• Produces contradictions or impossible answers.
• No internal grasp of logical laws.

(3) Coherence vs. Correspondence

• AI resembles coherence theory of truth.
• Humans rely on pragmatic/external correspondence.

(4) Inductive Scaling

• Mass statistical induction.
• Lacks real-world resistance to hypotheses.

4.1 Can AI Be Made More Logical or Grounded?

• Logic-checking modules on top of LLMs.
• Sensory-input robotic models.
• Reinforcement learning as weak external check.

4.2 Patrick’s Contributions

• Chain-of-thought workflows enhance reliability.
• World-modeling approaches (e.g., Yann LeCun).
• Classifying training data by epistemic status.

4.3 Limits of Current AI

• No qualitative experience (qualia).
• Understanding bounded by dataset.
• Humans form novel, non-statistical thoughts.
• True grounding requires continuous embodied sensory streams.

4.4 Hybrid Intelligence

• Combine human inquiry with AI pattern recognition.
• AI as concept-mapper, assistant, replication tool.

5.1 AI as a Tool in Academic Inquiry

• Unreliable for specific claims/testimony.
• Useful for:
  • Mapping conceptual connections
  • Processing large datasets
  • Routine or replication-heavy tasks
• Human oversight essential.

5.2 Concerns About Over-Automation

• Loss of human purpose in research.
• Homogenization of thought.
• Erosion of creativity.
• Potential increased creativity if routine work automated.

• AI, privacy, & medical data
• AI in warfare
• AI hallucinations and testimony
• Ethical issues in data scraping
• Bias and big-tech accountability
• Healthcare AI applications (e.g., mammography)

Matthew will circulate reading materials.

Proposed Date: Friday, February 13, 2026
Location: Same room (if available)
Time: Evening session
Matthew to confirm booking and share article beforehand.