Correctly labelling training data for AI models is vital to avoid serious problems, as is using sufficiently large datasets. However, manually labelling massive amounts of data is time-consuming and laborious. Using pre-labelled datasets can be problematic, as evidenced by MIT having to pull its 80 Million Tiny Images datasets. For those unaware, the popular dataset was found to contain thousands of racist and misogynistic labels that could have been used to train AI models. AI News caught up with Devang Sachdev, VP of Marketing at Snorkel AI, to find out how the company is easing the laborious process of labelling data in a safe and effective way. AI News: How is Snorkel helping to ease the laborious process of labelling data?

Devaki Raj leaned over a gilded railing inside Times Square's Marriott Marquis last Thursday, racking her brain to find the words that won't get her in trouble. Minutes before, she had stood confidently in front of a conference room full of investors, academics, military contractors, and Air Force acquisitions officers to deliver a slick pitch as part of the Air Force's first ever startup demo day. In her pitch, Raj explained how her company, CrowdAI, has mixed machine learning with mapping technology to identify flooded Texas roadways in the aftermath of Hurricane Harvey, or decimated buildings after bombings in Aleppo. The pitch, delivered to a closed-door crowd the day before, had already earned CrowdAI a small grant from the Air Force, which she hopes will soon blossom into a formal military contract. But now, Raj was finding it trickier to answer my questions about what this technology might be used for in practice by the military--or in war. When I asked if there are some applications of the tool that she would consider off limits, she said yes, but declined to name any in particular.