All modules
Module 02 · 12 minBeginner

How AI learns from data

Training data, pattern recognition, and why "garbage in, garbage out" is a clinical safety concept — not just a slogan.

The training process

Data → Training → Model → Prediction. Every clinical AI tool you encounter went through some version of this loop.

1
Data

Labeled examples are collected — e.g. pediatric chest X-rays marked 'normal' or 'pneumonia'.

2
Training

The model is shown examples over and over, adjusting itself to match the labels.

3
Model

A statistical pattern that has 'learned' from the examples.

4
Prediction

Given a new X-ray, the model estimates the likely label.

A pediatric case study

A children's hospital trains a model to flag possible pneumonia on chest X-rays. They include images from newborns, toddlers, school-age, and adolescents, across multiple body sizes and X-ray machines.

Works well on

Pediatric X-rays similar to the training set — the population the model was built for.

Watch out for

Adult X-rays, unusual portable views, or rare congenital findings the model rarely saw during training.

Interactive simulation

The Garbage In, Garbage Out (GIGO) Lab

Three hands-on activities. Tune the data, hunt for problems, then clean a clinical note — and earn a Data Steward badge.

Pediatric scenarios
Goal: Tune the data dials until your AI report card shows a strong overall rating. Try at least one preset and one custom configuration before locking in your score.
Training data dials
Dataset size
50
100 charts1M+ charts
Data quality
50
messy free-textstructured & validated
Missing information
50
fully populatedlots of gaps
Duplicate records
50
dedupedmany dupes
Patient diversity
50
single clinicmulti-site, all ages
Label accuracy
50
guessed labelsspecialist-confirmed
Documentation quality
50
sparse notescomplete H&P + follow-up
AI strength: Fair
Live model gauges
70%
Accuracy
Good
70%
Reliability
Good
60%
Fairness
Good
55%
Hallucination risk
Elevated
80%
Bias risk
High
54%
Clinical safety
Fair
AI Report Card
Needs work
Accuracy
70%
Bias
High
Safety
Fair
Recommendations
  • •Have specialists re-review labels — unverified labels drive hallucinations.
Vague / low-confidence output

Pediatric chest X-ray: possible infiltrate, etiology unclear. Confidence 0.62. Recommend clinical correlation.

Try this: drop diversity to 10 — watch fairness and bias collapse even when overall accuracy looks fine. That's how a model can be "accurate on average" while failing specific patient groups.

Every interaction reinforces a core safety lesson: data quality, label accuracy, and documentation directly shape AI behavior at the bedside.

Important: A polished, confident AI output tells you nothing about whether the underlying data was any good. Always ask about training data before trusting the result.

What happens when data is biased?

Bias in healthcare AI is rarely intentional — it's usually a reflection of who is and isn't in the training data. The consequences land on patients.

Dermatology models trained on lighter skin

Some early skin-cancer models performed dramatically worse on darker skin tones, because the training images underrepresented that population.

Risk scores calibrated on one demographic

A widely-used risk algorithm relied on healthcare spending as a proxy for need — systematically underestimating risk for Black patients.

Adult-trained models applied to children

Vital-sign and imaging models trained on adults can miss or misclassify pediatric findings the model never saw during training.

Why this matters at the bedside

If a model performs unevenly across subgroups, the patients who happen to look least like the training data get the worst care. Diversity in healthcare data is a patient-safety issue, not a checkbox.

Myth vs. fact

Myth

More data always makes a model better.

Fact

Only if it's the right data. Adding more of the same biased data can entrench errors.

Myth

If a model worked at one hospital, it'll work at ours.

Fact

Models often degrade when patient mix, equipment, or workflows change. Local validation matters.

Myth

AI is objective because it's math.

Fact

Models inherit the biases of the data they were trained on — and of the people who chose what to label.

Key takeaways
  • AI learns by adjusting itself to match many labeled examples — it does not memorize a rulebook.
  • Data quality and diversity matter more than data quantity.
  • Garbage in → garbage out. Poor data produces confident, polished, wrong answers.
  • Models perform best on patients who resemble the training set; performance can collapse outside it.
  • Bias in healthcare AI is a patient-safety issue — always ask who was (and wasn't) in the data.

How models actually learn

Flip each card to see the mechanism and a clinical analogy.

Quick check

Your hospital wants to deploy a readmission model trained at a different health system. What's the first thing to check?

Check your understanding

Five quick questions on data, bias, and quality.

Knowledge check
Question 1 of 5

You're not signed in — this score won't be saved or counted toward your level average. Sign in first to record it.

What does 'garbage in, garbage out' mean for clinical AI?

Score: 0
Previous: What is AI?