Underrepresented Groups: Building Inclusive Datasets

Small populations get hurt first when datasets are built carelessly. Fixing this requires intentional collection, not just better algorithms.

30 min · Reviewed 2026

The 1 Percent Problem

Imagine your training data is 99 percent able-bodied adults. Then a model trained on it is deployed to help with accessibility. It fails for people who use wheelchairs, screen readers, or sign language. Not because the team was malicious, but because the 1 percent was invisible in the data.

Common groups that get underrepresented

People with disabilities (blind, deaf, motor impairment)
Non-English speakers and speakers of creoles/dialects
Indigenous peoples and their languages
Rural populations (data skews urban)
Elderly adults (tech adoption skews younger)
Children (extra privacy protections limit data collection)
Trans and non-binary people (demographic forms often exclude them)

Why small groups stay small

Random sampling replicates population ratios
Self-selected online data over-samples tech-heavy demographics
Data collection platforms (Mechanical Turk) have demographic skews
Privacy rules (rightly) make minority data harder to collect
Researchers often do not have community connections

Intentional inclusion

import pandas as pd
import numpy as np

df = pd.read_csv('training_data.csv')

# Oversample minority groups to equal representation
def rebalance(df, group_col, target_size=None):
    groups = df[group_col].unique()
    if target_size is None:
        target_size = df[group_col].value_counts().max()
    
    balanced = []
    for g in groups:
        subset = df[df[group_col] == g]
        balanced.append(subset.sample(target_size, replace=True, random_state=42))
    
    return pd.concat(balanced).sample(frac=1, random_state=42)

balanced_df = rebalance(df, 'demographic_group')
print(balanced_df['demographic_group'].value_counts())Oversampling to rebalance minorities

Beyond oversampling

Partner with community organizations for targeted data collection
Pay data contributors from underrepresented groups fair wages
Report metrics per subgroup in every benchmark
Give affected communities a say in how the data is used
Allow data contributors to withdraw their data if the use changes

The big idea: equal representation in data is not automatic. It takes deliberate effort, community relationships, and willingness to reshape your sampling priorities. Inclusive data is always the result of intentional choices.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-data-underrepresented-groups

What is the core idea behind "Underrepresented Groups: Building Inclusive Datasets"?
1. Small populations get hurt first when datasets are built carelessly. Fixing this requires intentional collection, not just better algorithms.
2. Public records (court filings, property records, voter rolls)
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which term best describes a foundational idea in "Underrepresented Groups: Building Inclusive Datasets"?
1. oversampling
2. underrepresentation
3. Common Voice
4. inclusive sampling
A learner studying Underrepresented Groups: Building Inclusive Datasets would need to understand which concept?
1. underrepresentation
2. Common Voice
3. oversampling
4. inclusive sampling
Which of these is directly relevant to Underrepresented Groups: Building Inclusive Datasets?
1. underrepresentation
2. oversampling
3. inclusive sampling
4. Common Voice
Which of the following is a key point about Underrepresented Groups: Building Inclusive Datasets?
1. People with disabilities (blind, deaf, motor impairment)
2. Non-English speakers and speakers of creoles/dialects
3. Indigenous peoples and their languages
4. Rural populations (data skews urban)
Which of these does NOT belong in a discussion of Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. People with disabilities (blind, deaf, motor impairment)
3. Indigenous peoples and their languages
4. Non-English speakers and speakers of creoles/dialects
Which statement is accurate regarding Underrepresented Groups: Building Inclusive Datasets?
1. Self-selected online data over-samples tech-heavy demographics
2. Data collection platforms (Mechanical Turk) have demographic skews
3. Random sampling replicates population ratios
4. Privacy rules (rightly) make minority data harder to collect
Which of these does NOT belong in a discussion of Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Random sampling replicates population ratios
3. Self-selected online data over-samples tech-heavy demographics
4. Data collection platforms (Mechanical Turk) have demographic skews
Which statement accurately describes an aspect of Underrepresented Groups: Building Inclusive Datasets?
1. Imagine your training data is 99 percent able-bodied adults. Then a model trained on it is deployed to help with accessibility.
2. Public records (court filings, property records, voter rolls)
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
What does working with Underrepresented Groups: Building Inclusive Datasets typically involve?
1. Public records (court filings, property records, voter rolls)
2. The big idea: equal representation in data is not automatic. It takes deliberate effort, community relationships, and willingness to reshape…
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which best describes the scope of "Underrepresented Groups: Building Inclusive Datasets"?
1. It is unrelated to foundations workflows
2. It applies only to the opposite beginner tier
3. It focuses on Small populations get hurt first when datasets are built carelessly. Fixing this requires intentiona
4. It was deprecated in 2024 and no longer relevant
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Tiny dependencies: works without Excel, Python, or anything fancy
3. Write a labeling guideline document
4. Common groups that get underrepresented
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Why small groups stay small
2. Public records (court filings, property records, voter rolls)
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Intentional inclusion
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Tiny dependencies: works without Excel, Python, or anything fancy
3. Beyond oversampling
4. Write a labeling guideline document

← Back to interactive lesson

Tendril · Creators · AI Foundations

Underrepresented Groups: Building Inclusive Datasets

Small populations get hurt first when datasets are built carelessly. Fixing this requires intentional collection, not just better algorithms.

30 min · Reviewed 2026

The 1 Percent Problem

Common groups that get underrepresented

People with disabilities (blind, deaf, motor impairment)
Non-English speakers and speakers of creoles/dialects
Indigenous peoples and their languages
Rural populations (data skews urban)
Elderly adults (tech adoption skews younger)
Children (extra privacy protections limit data collection)
Trans and non-binary people (demographic forms often exclude them)

Why small groups stay small

Random sampling replicates population ratios
Self-selected online data over-samples tech-heavy demographics
Data collection platforms (Mechanical Turk) have demographic skews
Privacy rules (rightly) make minority data harder to collect
Researchers often do not have community connections

Intentional inclusion

import pandas as pd
import numpy as np

df = pd.read_csv('training_data.csv')

# Oversample minority groups to equal representation
def rebalance(df, group_col, target_size=None):
    groups = df[group_col].unique()
    if target_size is None:
        target_size = df[group_col].value_counts().max()
    
    balanced = []
    for g in groups:
        subset = df[df[group_col] == g]
        balanced.append(subset.sample(target_size, replace=True, random_state=42))
    
    return pd.concat(balanced).sample(frac=1, random_state=42)

balanced_df = rebalance(df, 'demographic_group')
print(balanced_df['demographic_group'].value_counts())Oversampling to rebalance minorities

Beyond oversampling

Partner with community organizations for targeted data collection
Pay data contributors from underrepresented groups fair wages
Report metrics per subgroup in every benchmark
Give affected communities a say in how the data is used
Allow data contributors to withdraw their data if the use changes

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-data-underrepresented-groups

What is the core idea behind "Underrepresented Groups: Building Inclusive Datasets"?
1. Small populations get hurt first when datasets are built carelessly. Fixing this requires intentional collection, not just better algorithms.
2. Public records (court filings, property records, voter rolls)
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which term best describes a foundational idea in "Underrepresented Groups: Building Inclusive Datasets"?
1. oversampling
2. underrepresentation
3. Common Voice
4. inclusive sampling
A learner studying Underrepresented Groups: Building Inclusive Datasets would need to understand which concept?
1. underrepresentation
2. Common Voice
3. oversampling
4. inclusive sampling
Which of these is directly relevant to Underrepresented Groups: Building Inclusive Datasets?
1. underrepresentation
2. oversampling
3. inclusive sampling
4. Common Voice
Which of the following is a key point about Underrepresented Groups: Building Inclusive Datasets?
1. People with disabilities (blind, deaf, motor impairment)
2. Non-English speakers and speakers of creoles/dialects
3. Indigenous peoples and their languages
4. Rural populations (data skews urban)
Which of these does NOT belong in a discussion of Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. People with disabilities (blind, deaf, motor impairment)
3. Indigenous peoples and their languages
4. Non-English speakers and speakers of creoles/dialects
Which statement is accurate regarding Underrepresented Groups: Building Inclusive Datasets?
1. Self-selected online data over-samples tech-heavy demographics
2. Data collection platforms (Mechanical Turk) have demographic skews
3. Random sampling replicates population ratios
4. Privacy rules (rightly) make minority data harder to collect
Which of these does NOT belong in a discussion of Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Random sampling replicates population ratios
3. Self-selected online data over-samples tech-heavy demographics
4. Data collection platforms (Mechanical Turk) have demographic skews
Which statement accurately describes an aspect of Underrepresented Groups: Building Inclusive Datasets?
1. Imagine your training data is 99 percent able-bodied adults. Then a model trained on it is deployed to help with accessibility.
2. Public records (court filings, property records, voter rolls)
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
What does working with Underrepresented Groups: Building Inclusive Datasets typically involve?
1. Public records (court filings, property records, voter rolls)
2. The big idea: equal representation in data is not automatic. It takes deliberate effort, community relationships, and willingness to reshape…
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which best describes the scope of "Underrepresented Groups: Building Inclusive Datasets"?
1. It is unrelated to foundations workflows
2. It applies only to the opposite beginner tier
3. It focuses on Small populations get hurt first when datasets are built carelessly. Fixing this requires intentiona
4. It was deprecated in 2024 and no longer relevant
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Tiny dependencies: works without Excel, Python, or anything fancy
3. Write a labeling guideline document
4. Common groups that get underrepresented
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Why small groups stay small
2. Public records (court filings, property records, voter rolls)
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Intentional inclusion
3. Tiny dependencies: works without Excel, Python, or anything fancy
4. Write a labeling guideline document
Which section heading best belongs in a lesson about Underrepresented Groups: Building Inclusive Datasets?
1. Public records (court filings, property records, voter rolls)
2. Tiny dependencies: works without Excel, Python, or anything fancy
3. Beyond oversampling
4. Write a labeling guideline document

← Back to interactive lesson