AI summary by TwinMind
What tech companies know about your kids | Veronica Barassi
TED
11m 1s
Introduction and Research Background
- Speaker: Anthropologist, mother of two daughters
- Research project launched 2015: "Child Data Citizen"
- Focus: Data collection about children from conception through life
- Problem: Systems-level issue, not individual behavior
Data Collection from Conception to Birth
- Parents tracked from conception decision through pregnancy
- Activities generating data:
- Online fertility research
- Ovulation tracking apps
- Ultrasound sharing on social media
- Pregnancy apps
- Medical searches (miscarriage risks, pregnancy symptoms)
- Baby tracking continues post-birth across multiple technologies
Mobile Health Apps Data Sharing Study
- 2019 British Medical Journal research findings:
- 24 mobile health apps studied
- 19 apps shared information with third parties
- Third parties shared with 216 other organizations
- Only 3 of 216 organizations belonged to health sector
- Recipients included: Google, Facebook, Oracle, digital advertising companies, consumer credit reporting agencies
Comprehensive Child Tracking Systems
- Multiple tracking environments:
- Home: Virtual assistants, smart home technologies
- School: Educational platforms and technologies
- Healthcare: Online records, patient portals
- Entertainment: Internet-connected toys, online games
- Many other technologies
Data Profiling and Decision-Making Systems
- Individual tracking combined with profiling using AI and predictive analytics
- Data sources aggregated: Family history, purchasing habits, social media comments
- Decision-making applications:
- Banks: Loan decisions
- Insurance: Premium calculations
- Employers/recruiters: Job fit assessments
- Police/courts: Criminal risk assessment, recidivism prediction
- No user knowledge or control over profiling processes
Educational Data Brokers Case Study
- 2018 New York Times report: Online college planning service data sold to educational data brokers
- Millions of high school students affected
- Profiling categories for children as young as 2:
- Religion, athletics, social awkwardness, other random categories
- Data sold includes: Names, home addresses, contact details
- Buyers: Trading companies, career institutions, student loan companies, credit card companies
- Research test: Data broker agreed to provide list of 14-15 year old girls interested in family planning services
Algorithmic Bias and Limitations
- AI/predictive analytics useful for disease prediction, climate change
- Cannot objectively profile humans
- Data traces ≠ human reality
- Algorithms inherently biased:
- Designed by humans within cultural contexts
- Shaped by specific cultural values
- Machine learning from biased algorithms and databases
Predictive Policing Bias Example
- AI Now Institute (New York) 2019 report
- AI technologies for predictive policing trained on "dirty data"
- Data from historical periods of racial bias and non-transparent police practices
- Outcomes: Amplifying and perpetuating police bias and error
Fundamental Societal Problem
- Society trusting biased, inaccurate profiling technologies
- Need for political solutions
- Data rights = human rights
- Generational difference: No public childhood records for current adults vs. comprehensive data collection for current children
Call to Action
- Demand greater data justice
- Collective action needed: Individuals, organizations, institutions
- Urgency: Prevent algorithmic discrimination affecting children's future opportunities
