Lesson #5
Trust, Risk & Running AI Responsibly
- 1
Bias
Category: Data“BYE-uss”
When an AI's answers unfairly favor one group or view because the data it learned from leaned that way.
Think of it like… A referee who only ever watched one team play — the calls tilt without them meaning to.
Construction: A bidding AI that lowballs jobs in one neighborhood because old records did.
Business: A hiring screener that ranks résumés lower for people who took a career break.
Why this matters: Biased output can quietly cost you customers, good hires, and legal standing.
Open full entry → - 2
Explainability
Category: Fundamentals“ex-PLAIN-uh-BIL-ih-tee”
How well an AI can show you the reasons behind an answer, in words a person can follow.
Think of it like… A doctor who tells you why, not just what pill to take.
Construction: An estimator AI listing the labor and material lines behind its number.
Business: A fraud tool naming the three signals that flagged a transaction.
Why this matters: If you can't explain an AI decision, you can't defend it to a customer, a boss, or a regulator.
Open full entry → - 3
AI Governance
Category: Infrastructure“A-I GUV-ern-ance”
The written rules for who may use AI at your organization, for what, and who is accountable when it goes wrong.
Think of it like… The safety manual and sign-off sheet on a job site.
Construction: A policy that AI drafts every permit letter but a licensed person signs it.
Business: A one-page standard listing approved AI tools, banned data types, and a review owner.
Why this matters: Governance is what turns AI from a risky experiment into something you can safely scale.
Open full entry → - 4
Data Privacy
Category: Data“DAY-tuh PRY-vuh-see”
Keeping personal or confidential information out of places it shouldn't go — including AI tools.
Think of it like… A locked filing cabinet: not everything belongs on the open counter.
Construction: Stripping client addresses out of job notes before an AI summarizes them.
Business: Using a business AI account so customer records aren't used to train outside models.
Why this matters: One careless paste can turn private data into a permanent leak you can't undo.
Open full entry → - 5
Prompt Injection
Category: Infrastructure“PROMPT in-JEK-shun”
A trick where hidden text in a document or web page tells the AI to ignore your instructions and do something else.
Think of it like… Someone slipping a fake note into your assistant's inbox signed with your name.
Construction: A spec PDF with hidden text telling the AI to approve a change order.
Business: An email that tricks a support AI into revealing internal pricing.
Why this matters: Any AI that reads outside content can be hijacked — treat what it reads as untrusted.
Open full entry → - 6
Open-Source Model
Category: Models“OH-pen SORSE MOD-ul”
An AI model whose inner workings are published, so anyone can download, inspect, and run it themselves.
Think of it like… A recipe printed in full versus a secret sauce you can only buy by the bottle.
Construction: Hosting a private model in the office so job data never leaves the building.
Business: A clinic running an open model on its own server to meet privacy rules.
Why this matters: Open models give you control and predictable cost; closed ones often give you more raw power.
Open full entry → - 7
Inference Cost
Category: Infrastructure“IN-fur-ence COST”
What you pay each time an AI produces an answer, usually measured by how many tokens go in and out.
Think of it like… A taxi meter: the ride is cheap, but it runs every single trip.
Construction: Summarizing a 300-page spec costs far more than answering one short question.
Business: Cutting support costs by using a small model first and a big one only when needed.
Why this matters: AI pilots look cheap; the bill arrives when you multiply by every customer, every day.
Open full entry → - 8
Workflow
Category: Agents“WORK-flow”
A fixed set of steps an AI follows in order, every time — unlike an agent that decides its own path.
Think of it like… A recipe card: same steps, same order, repeatable results.
Construction: Site photo → damage summary → draft change order → email to the client.
Business: Invoice arrives → data pulled out → matched to a PO → queued for approval.
Why this matters: Most business value comes from boring, reliable workflows — not clever improvised agents.
Open full entry → - 9
Evaluation
Category: Fundamentals“ee-val-yoo-AY-shun”
Testing an AI against known-good examples to measure how often it gets things right before you trust it.
Think of it like… A road test before you hand someone the keys to the truck.
Construction: Running an estimator AI on 20 past jobs and comparing to the real invoices.
Business: Scoring a support bot on 100 real tickets before letting it reply to customers.
Why this matters: Without evaluation you're guessing, and you won't notice when an update makes things worse.
Open full entry → - 10
Responsible AI
Category: Fundamentals“ree-SPON-suh-bul A-I”
Building and using AI in ways that are fair, honest, safe, and clearly accountable to real people.
Think of it like… Building to code, not just building fast.
Construction: Having a human sign off on any AI-generated safety instruction.
Business: Publishing where your company uses AI and who to contact about a bad outcome.
Why this matters: Trust is the hardest thing to rebuild — responsible practice protects it up front.
Open full entry →