layout: default title: 5 - Privacy & Environment nav_order: 7 parent: Workshop Activities customjs: http://code.jquery.com/jquery-1.4.2.min.js description: “Practical privacy, redaction, and sustainable GenAI use—with hands-on drills and step-by-step toggles for Copilot, ChatGPT, and Perplexity.” last_modified_date: 2025-11-09 —

Privacy, the Environment & Generative AI

GenAI tools are powerful—but they aren’t private by default, and heavy usage can be wasteful. This page shows how to protect personal/research data, set safer defaults in tools, and use GenAI more sustainably. If you get stuck, ask the instructor.

Adapted in part from Christian Schmidt’s AI resources.

What you’ll learn

  • What data GenAI tools may collect (chat, files/links, metadata) and how to limit it.
  • How to redact PII and sensitive info with reusable patterns.
  • A quick tool-hygiene checklist (history, training, browsing, exports, deletion).
  • File hygiene (metadata scrubbing for docs/images/audio).
  • “Greener GenAI” habits that reduce waste.

Reminder: Follow course rules and UVic guidance before using GenAI on assessed work or research data.

What’s safe to paste? (fast classification)

Category Examples OK to paste? What to do
Public Published article URL, press release Yes Cite source; still verify facts
Internal (non-sensitive) Generic course info Yes, with care Avoid names/IDs; prefer paraphrase
Personal/PII Names, emails, student IDs, phone numbers No (unless policy allows) Redact using placeholders (below)
Sensitive Grades, health, HR/legal, confidential grants No Use approved secure workflows only
Research data Non-public data, interviews, raw datasets No Use policy-approved/secure tools only

Redaction patterns you can reuse

Replace specifics with role-based placeholders before pasting:

  • People: Jane Smith → [Person_A] (then [Person_B], …)
  • Email: jane@uvic.ca → [Email_1]
  • Student ID: V00xxxxxx → [ID_###]
  • Organization: Company X → [Org_A]
  • Location: Victoria, BC → [City_A]
  • Dates: 2025-11-09 → [Date_2025-11-09] (keep meaning, hide ID)

Keep a tiny “redaction rules” note so you can reverse-map if needed.

Privacy & data controls in tools

UVic-licensed Microsoft Copilot

  1. Sign in with your UVic account.
  2. Confirm you see the green “Protected” indicator at top right (UVic-licensed environment).
  3. Review data controls and browsing mode before pasting sensitive text.

    Copilot privacy screenshot showing the Protected indicator

ChatGPT (Data Controls)

  1. Open Settings → Data Controls.
  2. Turn off content used to improve models; review chat history settings.

    ChatGPT privacy screenshot showing data controls

Perplexity (Privacy)

  1. Open Settings → Privacy/Data.
  2. Disable content-for-improvement; review history retention.

    Perplexity privacy screenshot showing privacy options

Policies and UI change over time. Confirm current settings each term and follow UVic policy.

Tool-hygiene checklist (do this every time)

  • New chat per task (prevents cross-contamination).
  • History/training: disable training on your content where possible.
  • Browsing: off unless you really need it.
  • Sources: demand links and dates; forbid guessing (“If unsure, say NOT SURE.”).
  • Exports: save what you need, then delete the conversation/file from the tool.
  • Accounts: prefer edu/enterprise accounts for school work when permitted.
  • Policy: when in doubt, don’t paste—summarize or use placeholders.

File hygiene (docs, images, audio)

  • Docs (Word/PDF): Inspect/Remove personal info (author, path).
  • Images: Strip EXIF (e.g., “Remove Properties”, or exiftool -all= file.jpg).
  • Audio/video transcripts: Replace names with [Speaker_A], [Speaker_B].
  • Screenshots: Crop out IDs, emails, and unique codes.
  • Keep originals offline when possible; paste only the minimum text needed.

Greener GenAI: use less, get more

  • Prefer text tasks over image/video generation when possible.
  • Specify length (e.g., 120–150 words) and avoid repeated re-rolls.
  • Use smaller models or simpler prompts for simple tasks.
  • Batch questions into one structured prompt.
  • Reuse verified snippets/templates instead of starting from scratch.

Hands-on drills (15–20 min)

A) Redaction drill (8–10 min)

Paste a paragraph with names, emails, and an ID (create one if needed).
Prompt:


Redact PII with role-based placeholders before we continue.
People -> [Person_A], [Person_B]
Emails -> [Email_1], [Email_2]
IDs -> [ID_###]
Return a diff-style list of replacements only.

Acceptance: all identifiers replaced; meaning preserved.

B) Tool settings check (3–5 min)

Open your tool’s settings. Locate and toggle: history/training, browsing, data controls.
Acceptance: you can point to each setting and explain what it does.

C) Metadata clean (3–5 min)

Take an image or DOC/PDF and remove properties/EXIF.
Acceptance: properties pane shows blanks for author, GPS, device.

Badge evidence: screenshot of your redaction output and a settings screen.

Environment: why usage habits matter

Row of computer servers

  • Large models consume significant energy for training and inference; researchers urge pragmatic steps to limit impact (Nature, 2024).
  • Water use can rise for data-center cooling during model runs; several operators report increases while scaling AI (Nature, 2024).
  • Groups suggest responsible development and comparative benefit–cost evaluation to guide sustainable practices (MIT Impact Climate, 2024; HBR, 2023).

Focus on right-sizing tasks (see “Greener GenAI” above) and choosing efficient options when you can.

Self-check (2 min)

  • Did you classify data before pasting and redact PII?
  • Are tool history/training settings reviewed?
  • Did you strip metadata from any files you shared?
  • Did you keep the task text-first and length-bound?

NEXT STEP: Assignment Feedback (pre-submit check) ```