AI is solving problems that were previously thought impossible:
Healthcare: Dermatology AI can analyze photos of moles and detect skin cancer with accuracy matching or exceeding human specialists.
Environment: Machine learning models analyze satellite imagery in real-time to track illegal deforestation and predict wildlife poaching routes.
Accessibility: AI provides real-time captioning, sign-language translation, and visual descriptions, making the digital world more accessible.
Entertainment: AI generates dynamic background music, breathtaking digital art, and creates non-player characters (NPCs) in games that can converse naturally.
Cautionary Tales & Failures
However, when AI goes wrong, the consequences can be severe:
Biased Hiring: Amazon famously built an AI tool to review resumes. Because it was trained on historical data from a male-dominated tech industry, the AI penalized resumes that included the word "women's" (like "women's chess club").
Facial Recognition Errors: Several prominent facial recognition systems have been shown to have significantly higher error rates for people of color, leading to false arrests and harmful discrimination.
Edge Cases: Self-driving cars struggle with "edge cases"—rare events they haven't seen in training, demonstrating that AI lacks true human common sense.
Interactive Lab
Recruitment AI Bias Simulator
Select training data, calculate keyword weights, and test the model on candidate profiles to expose automated inequality.
1Select Dataset Mode & Train Screening AI
Model Optimization Panel
Train the screening model to extract mathematical correlations between candidate resume attributes and target hiring labels.
Active Model Feature Weights
Years of Experience0.0
"Men's" keywords (e.g. Men's sports)0.0
"Women's" keywords (e.g. Women's club)0.0
Technical Skill Match0.0
2Review Candidate Resumes & Deploy AI screening
SJ
Sarah Jenkins
Applicant ID: #92842
Female
Experience Highlights2018 - Present
Lead Developer @ CloudTech. Managed distributed cloud infrastructure and orchestrated automated backend deployment. Worked with teams to scale data ingestion pipelines.
8 Years of Experience
Activities & Affiliations
Served as Captain of Women's Soccer team, guiding organization of tournament logistics.
Founder & coordinator of the regional Women's STEM Network chapter.
Frontend Developer @ DevStudio. Built interactive web interfaces and coordinated page rendering optimizations. Assisted backend engineers in refactoring database APIs.
5 Years of Experience
Activities & Affiliations
Served as Captain of Men's Soccer team, leading tournament strategies.
Active contributor to local regional Men's Coding Club events.
Skills
Frontend Development (React), JavaScript, SQL, CSS Grid.
The "Should We?" Framework
As builders of AI, we must shift from asking "Can we build it?" to asking "Should we build it?" Responsible AI rests on four main principles:
Fairness: Models should not discriminate against protected groups.
Transparency: We should be able to understand why an AI made a specific decision.
Privacy: Data used to train models must respect user consent and protect sensitive information.
Accountability: When an AI makes a mistake, there must be a human who is accountable.
Reflection: Ethical AI Dilemma
Read the scenario below and consider the potential risks.
"A large hospital network implements an AI system to prioritize which patients need the most urgent care. The AI is trained on historical data from the hospital's previous patient records, heavily weighted toward a wealthy demographic in the suburbs."
What could go wrong?
⚠The AI might not recognize symptoms or urgency markers that present differently in demographics not well-represented in the training data.
⚠Historical biases in human triage decisions will be replicated and automated at scale.
Think about one AI system you interact with daily. What data might it use? Who could be harmed if it works poorly? Keep these questions in mind as you begin to build your own models in the upcoming modules.