TED Talks Daily How bad data keeps us from good AI | Mainak Mazumdar
Jan 15, 2021
Mainak Mazumdar, a data scientist, delves into the hidden perils of biased data in AI. He illustrates how flawed data from places like Shanghai and New York leads to poor decisions in job and loan allocations. Mazumdar emphasizes the undercounting of minority communities in the 2010 census, showing its dire impact on AI models. He advocates for enhanced data quality and inclusivity to ensure that AI serves everyone fairly. His insights call for a fundamental shift in how we approach data collection to build ethical AI systems.
AI Snips
Chapters
Transcript
Episode notes
AI's Societal Impact
- AI is becoming a gatekeeper to the economy, influencing access to jobs and loans.
- This power is accelerating biases at speed and scale, raising societal concerns.
Bias in Image Recognition
- Duke University's AI model, Pulse, incorrectly enhanced a non-white image into a Caucasian image.
- This highlights how underrepresentation in training data leads to biased outcomes.
Census Undercounting
- The 2020 US Census potentially undercounted minority groups, impacting data infrastructure.
- Undercounting minorities introduces bias and affects policy decisions.