AI Safety, Ethics & Best Practices

AI systems can perpetuate and amplify biases. Understanding this helps you use AI more critically.

Where Bias Comes From

AI learns from data created by humans - and humans have biases. If training data contains stereotypes or underrepresents certain groups, the AI will reflect this.

Examples of AI Bias

How to Recognize Bias

Watch for:

• Default assumptions about demographics
• Stereotypical associations
• Missing perspectives or representations
• Consistent patterns that seem unfair

What You Can Do

1. Be specific: Instead of "a doctor," specify "a female doctor" if that's what you want
2. Check outputs: Look for stereotypical patterns
3. Request diversity: Ask for multiple perspectives or representations
4. Report issues: Many AI companies want to know about bias problems

What you share with AI may be stored, analyzed, and used for training. Protect yourself.

Never Share with AI

Understanding Data Usage

Different AI services have different policies:

ChatGPT (OpenAI)

• Free tier: Conversations may train models
• Plus: Can opt out in settings
• API: Different terms

Claude (Anthropic)

• Does NOT train on conversations by default
• Conversations deleted after 90 days

Enterprise/Business Tiers

• Usually have stronger privacy protections
• Data not used for training

Best Practices

1. Read privacy policies (at least the summary)
2. Use anonymous examples instead of real data
3. Check privacy settings in your AI tools
4. Use work AI tools for work if your company provides them
5. Assume nothing is private when using free tiers

The Trust But Verify approach keeps you safe while leveraging AI's power.

The Verification Hierarchy

Verification Strategies

For Facts and Claims

1. Ask AI for its sources
2. Search for those sources independently
3. Cross-reference with authoritative sources
4. Check dates (AI knowledge has cutoffs)

For Citations

• Search the exact citation
• Check if authors/journals exist
• Use Google Scholar for academic papers
• Assume citations might be fabricated

For Code

• Test thoroughly
• Review for security issues
• Check against documentation
• Don't trust without testing

For Medical/Legal

• AI is for preliminary research ONLY
• Always consult professionals
• Never self-diagnose or self-treat based on AI

Red Flags

Be extra skeptical when AI provides:

• Very specific statistics
• Recent events (post-training cutoff)
• Information about obscure topics
• Medical or legal advice
• Financial recommendations

Using AI responsibly benefits everyone.

Give Credit

When AI substantially helps your work:

• Disclose AI assistance when appropriate
• Don't claim AI-generated content as purely your own
• Follow your organization's AI disclosure policies

Don't Deceive

Never use AI to:

• Create fake reviews or testimonials
• Impersonate real people
• Generate misleading content
• Spread misinformation

Academic Integrity

Always check your institution's AI policy.

Professional Context

• Follow your company's AI policy
• Don't input confidential information
• Verify before sending AI content externally
• Disclose AI assistance when required

AI is a powerful tool, but you're still in charge.

AI Amplifies, Doesn't Replace

Think of AI as:

• A brilliant but unreliable assistant
• A first-draft generator
• A brainstorming partner
• A research accelerator

NOT as:

• A replacement for expertise
• An infallible oracle
• A substitute for human judgment
• The final word on anything important

When to Trust AI vs. Seek Humans

Building Complementary Skills

As AI handles routine tasks, focus on developing:

• Critical thinking
• Emotional intelligence
• Creative vision
• Ethical judgment
• Leadership
• Complex problem-solving

These human skills become MORE valuable, not less, in an AI world.

Your AI Usage Guidelines

Create your own rules for how you'll use AI. Consider:

• What tasks will you use AI for?
• What will you never use AI for?
• How will you verify AI outputs?
• When will you disclose AI assistance?
• What privacy boundaries will you maintain?