Lesson 4: Ethics Basics - Bias, Copyright, and Responsible Use

In 2015, Amazon built an AI recruiting tool to sort through resumes and recommend candidates. They trained it on 10 years of their historical hiring data.

Within a year, they discovered the system was penalizing resumes that included the word "women's" (as in "women's chess club captain") and downgrading graduates of all-women's colleges.

Why? Amazon's tech workforce had been predominantly male for years. The AI learned that being male was correlated with being hired, so it started preferring male candidates. It had automated Amazon's existing bias and made it worse.

They scrapped the entire project in 2017.

This is what happens when you treat AI as neutral. It's not. It's a reflection of the data it was trained on, the choices its creators made, and the biases baked into both.

You can't use AI responsibly without understanding how it can go wrong, who gets hurt when it does, and what you're responsible for when you press "generate."

How Bias Gets Into AI

AI models learn patterns from massive amounts of data scraped from the internet, books, and other sources. If that data reflects historical inequalities, stereotypes, or skewed perspectives, the AI absorbs all of it.

Some examples:

Image generation: Early AI image generators would show doctors as men and nurses as women by default. Ask for "a CEO" and you'd get a middle-aged white man in a suit. This wasn't a bug in the code, it was a pattern in the training data.

Language processing: AI trained on internet text absorbed gender stereotypes. When researchers tested various models, they found AI was more likely to associate words like "programmer" and "engineer" with male pronouns, and "nurse" and "teacher" with female ones.

Criminal justice: An AI system called COMPAS, used to assess recidivism risk in criminal sentencing, was found to be biased against Black defendants. It falsely flagged Black defendants as high-risk at nearly twice the rate of white defendants.

The problem isn't just historical bias. It's amplification. When AI makes millions of decisions based on biased patterns, it can reinforce and spread those biases faster than any human system could.

How to Spot Bias in AI Outputs

You can't eliminate bias completely (it's baked into the training data), but you can learn to recognize it:

1. Look for Missing Perspectives

If you ask AI to brainstorm solutions to a problem and all the suggestions reflect one cultural perspective or economic situation, that's a red flag. Try asking: "What perspectives am I missing?" or "How would this look different from [specific viewpoint]?"

2. Check Default Assumptions

When AI generates examples, stories, or scenarios, notice who it centers by default. Are leadership roles automatically male? Are service workers automatically people of color? Are families automatically straight couples with kids?

These defaults reveal the patterns in its training data.

3. Test for Stereotype Replication

If you're using AI to generate content about people, try swapping demographic details and see if the output changes in stereotypical ways. Does changing a character from male to female suddenly shift their described personality traits? That's bias at work.

4. Question Overconfident Claims About People

Be skeptical when AI makes broad statements about groups of people. "Millennials prefer," "Studies show women are more," "Research indicates men typically" - these kinds of claims are often oversimplified or based on biased research.

The Copyright Gray Zone

In January 2023, Getty Images sued Stability AI for allegedly using millions of Getty's copyrighted photos to train their image generation AI without permission or compensation.

In September 2023, a group of authors including John Grisham and George R.R. Martin sued OpenAI for training ChatGPT on their books without permission.

The legal question at the center of both cases: is training an AI on copyrighted work fair use, or is it theft?

Courts haven't fully settled this yet, but here's what you need to know practically:

What's Probably Safe

Text output for your own use: Using AI to write emails, reports, or internal documents is almost certainly fine. You're the one using the tool to generate new text.

Code for your own projects: Using AI to generate code snippets or help with programming is generally accepted, though you should check your employer's policy on code ownership.

Ideas and concepts: If AI helps you brainstorm or understand concepts that you then use to create your own original work, that's no different from getting ideas from a book or conversation.

Properly licensed tools: Tools like Microsoft Copilot, Adobe Firefly, and Shutterstock AI are trained on licensed or owned content. If you're concerned about copyright, use these instead of open models.

What's Risky

Selling AI-generated images: If you use an AI image generator to create pictures and then sell them, you could face copyright challenges. Some AI companies claim copyright over generated images, and the training data's copyright status is contested.

Using AI output in commercial products: If you're building a product or service that incorporates AI-generated content, consult a lawyer. The copyright situation here is evolving fast.

Asking AI to "write like [specific author]": Deliberately trying to mimic a specific person's copyrighted style is legally risky. Courts might see this differently than general AI use.

Publishing AI-generated content without disclosure: Some publishers now require disclosure if AI was used substantially. Check submission guidelines.

The Practical Rule

For now, treat AI output as raw material that needs your creative input to become truly yours. Don't just generate and publish. Generate, edit, add your own ideas, and create something that's transformatively different.

The more of your own thought and effort you add, the safer you are.

Who's Responsible When AI Screws Up?

In February 2024, Air Canada's chatbot told a customer they could apply for a bereavement discount retroactively after buying a full-price ticket. The customer bought the ticket, then requested the discount.

Air Canada refused, saying the chatbot was wrong and they weren't responsible for what it said.

The customer sued. A Canadian tribunal ruled that Air Canada was responsible for the chatbot's promises. The company had to honor the discount.

The lesson: you can't outsource accountability to AI.

If you use AI to interact with customers, create content, make recommendations, or assist with decisions, you're responsible for what it produces. "The AI made a mistake" is not a legal defense.

This applies to you personally too. If you use AI to help with work and it generates something wrong, illegal, or harmful, you're the one who'll face consequences. The AI can't be fired or sued. You can.

Before using AI for anything that affects other people, ask:

• What could go wrong?
• Who could get hurt?
• Am I prepared to take responsibility if this output is wrong?
• Can I verify this is accurate before using it?

The Environmental Cost

One fact that doesn't get talked about enough: AI uses massive amounts of energy.

A single ChatGPT query uses roughly as much electricity as keeping a light bulb on for an hour. Training large AI models can emit as much carbon as five cars over their entire lifetimes, including manufacturing.

This doesn't mean you shouldn't use AI. But it does mean you should use it thoughtfully. Some practical guidelines:

Be efficient with queries: Instead of asking AI the same question five different ways, think through what you need and ask once.

Use it for things that matter: AI is great for saving you hours of work. Using it to generate a joke? That's a lot of energy for low value.

Consider the alternative: If you'd otherwise be driving somewhere, printing reams of paper, or shipping physical products, AI might actually be the more efficient choice.

This is an emerging area. As AI becomes more efficient and renewable energy grows, the environmental math will change. But for now, being mindful of energy use is part of responsible AI use.

6 Simple Rules for Responsible AI Use

Here's how to use AI in ways that won't keep you up at night:

1. Human in the Loop

Never let AI make a decision without human review, especially if it affects people. You review the output, you check the facts, you make the call.

2. Disclose When It Matters

If someone's making a decision based on AI output and they deserve to know that (a client, a student, a judge), tell them.

3. Verify Before You Trust

If accuracy matters, fact-check the AI's output. If bias could be present, look for it. If legal issues could arise, consult someone who knows.

4. Don't Automate Discrimination

If you're using AI to make decisions about people (hiring, lending, admissions), actively check for bias. Test it with diverse inputs. Get diverse perspectives on the outputs.

5. Give Credit Where It's Due

If you use AI substantially in creating something, be honest about that. You don't need to write "generated by ChatGPT" on every email. But if you're publishing work or submitting something for evaluation, appropriate disclosure protects you and respects others.

6. Consider Who Bears the Risk

Before using AI for something, ask who gets hurt if it goes wrong. If the answer is "someone more vulnerable than me," be extra careful.

The Bottom Line

AI isn't good or evil. It's a tool that amplifies human intent and reflects human data, including all our flaws.

Using it responsibly means staying aware of those flaws, not assuming the technology is neutral, and taking responsibility for how you deploy it.

You don't need to become an AI ethicist. You just need to think critically about what you're asking AI to do, check its work, and remember that you're the human in charge.

The goal isn't to avoid AI because it's imperfect. The goal is to use it in ways that make things better, not worse, and to catch the problems before they become disasters.

Bias, copyright, accountability, these aren't abstract concepts. They're practical considerations that affect real people. Including you.

Use AI as a tool to augment your judgment, not replace it. That's the difference between being strengthened by AI and being made irrelevant by it.

You're still the one making decisions. Own them.