Machine perception and learning depend on large, unbiased training data¶

Claim · All claims
Evidence · 1 study · theoretical r2 · q1
Currency · newest study 2019 · not yet checked against its sources · not reviewed by a person · overdue for a recheck since 2022: tool and model accuracy changes quickly

Subclaims¶

q1 i? Machine perception of language or imagery requires extensive domain knowledge, and machine learning requires large amounts of data whose collection must avoid introducing biases. → Touretzky 2019

Evidence¶

Touretzky 2019¶

Touretzky, D.; Gardner-McCune, C.; Martin, F.; and Seehorn, D. (2019). Envisioning AI for K-12: What Should Every Child Know about AI? The Thirty-Third AAAI Conference on Artificial Intelligence (AAAI-19). https://doi.org/10.1609/aaai.v33i01.33019795

q1 · i? · theoretical · r2

Expository claim from the Big Idea #3 section: machine learning is statistical inference that finds patterns in data, illustrated by the Open Image Dataset V4 with "9 million training images and 30 million labels". The article offers this as background explanation, not a tested result.

"the Open Image Dataset V4 contains 9 million training images and 30 million labels. Processing data at this scale requires computing power that was unavailable a few years ago. Care must be taken in the collection of this data to avoid introducing biases into the training set."

Discussion¶

Related Claims¶

• Algorithmic bias and data privacy are central ethical risks of AI in adult education — related
• Prior studies finding nonexperimental estimators biased examined weaker evaluation designs than RCT-benchmark studies — related
• Anonymization algorithms preserving result patterns may require sample sizes far larger than typical experiments — related
• Teachers know far more about their students than an investor knows about a stock, including data not typically in statistical models — related