DataDownload: The tiny company supplying facial recognition to 600 law enforcement agencies

7 min readJan 25, 2020

DataDownload: The tiny company supplying facial recognition to 600 law enforcement agencies A weekly summary of all things Media, Data, Emerging Tech View this email in your browser

When you throw a party, it all comes down to the guests. So, when the NYC Media Lab and the Knight Foundation invited 150 people working in Journalism and AI to our new home at 370 Jay Street in Brooklyn, we worked hard to invite people who had something to say. It was — to say the least — aerobic. We’ll share videos from the event in the weeks ahead, but suffice it to say… AI is going to be a critical part of how local journalism evolves. Here’s a picture from the event — more on the way.

(Photo: Meredith Broussard, NYU / John Keefe, Quartz / Mark Hanson, Columbia / Jennifer Choi, CUNY / Justin Hendrix, NYC Media Lab)

Please read, comment, and send along your thoughts about how we can work together. Always excited to hear from you.

Cheers,

Steven Rosenbaum
Managing Director
The NYC Media Lab Must-Read The Secretive Company That Might End Privacy as We Know It

“Absent a very strong federal privacy law, we’re all screwed.” — Al Gidari, privacy professor at Stanford

Over 600 law enforcement agencies have tapped into Clearview.ai’s 3B photo database and facial recognition system. Thanks to the large number of photos — scraped from major public networks — Clearview’s software was an instant hit. The company has helped solve serious cases, but its potential to do harm is enormous: the code base even includes features that allow you to connect the software to AR glasses and identify strangers’ home addresses and sensitive details with a glance.

In other major surveillance news, London’s police department announced Friday that they will be employing facial recognition to spot suspects. The system goes beyond photo matching, using “software that can immediately identify people on a police watch list as soon as they are filmed on a video camera.” The announcement, of course, sparked outrage and debate.

Meanwhile, in a rare move, Chinese officials apologized for locating people wearing pajamas on the street via facial recognition and publicly shaming them on WeChat.

18 min read

Read More Emil’s Story as a Self-Taught AI Researcher Self-taught AI researcher and Google Arts & Culture resident Emil Wallner demonstrates why a diverse background is a necessity in a field dominated by academia. A teacher in Ghana, truck driver, and musician, Wallner’s background exposed him to many social issues, leading him to co-found an investment firm to fund education initiatives. It also influenced the way he works with AI.

Waller gives some stellar advice here on how to go about self-teaching yourself AI. Definitely check out some of his projects on Twitter.

15 min read Read More For the Media People Can Be Identified by the Way They Dance

What’s your unique dance ID? Finnish researchers used mocap to log subject movements while they danced, in order to determine specific traits: mood, empathy levels, extroversion, neuroticism, and the style of music they were dancing to (the latter worked 30% of the time).

Unintentionally, the researcher found that each person had their own dance signature, and could be identified by the way they moved. Yes, we thought of dance club surveillance tactics too. But the authors note that they’re “less interested in applications like surveillance than in what these results tell us about human musicality.” You can read the paper here.

2 min read

Read More Amazon Details the AI Behind Alexa’s Whisper Mode Alexa’s whisper mode — released fully last November — responds to your whispering by whispering back. Initial technical details were minimal, but last week Amazon released a blog and a paper in IEEE Signal Processing Letters that expanded on the mode’s inner-workings.

The paper presents a comprehensive overview of how Amazon applied scientist Marius Cotescu and his team used gaussian mixture models and deep neural networks to get Alexa understanding our whispers.

2 min read Read More Google’s Ads Just Look Like Search Results Now

Unless you have an ad blocker enabled, you probably noticed Google’s new search ad format, or lack thereof. Last week Google began rolling out ads that are indistinguishable from organic search results… except for a small “Ad” icon to the left of the page title.

Verge reporter Jon Porter calls it Google’s “dark pattern,” and it shows: here’s the visual history of Google’s ad shading and labeling.

3 min read

Read More What We’re Watching An Insight, An Idea with Sundar Pichai

Alphabet CEO Sundar Pichai discussed quantum computers, AI, and the future of tech governance at Davos 2020. On the latter: it’s worth mentioning FAMGA’s keen interest in the controversial defense and law enforcement marketspace (ex. Amazon Ring partnered with over 400 US law enforcement agencies, Microsoft’s JEDI contract). But big tech is also well aware that looming regulation might curb AI use (and revenue); the preference is to be on rule-making side of things.

This vested interest was on display at Davos, where we saw major AIaaS players like IBM, Alphabet, and Microsoft suggest ways to regulate the technology… but neither Microsoft nor Alphabet are on board with EU’s proposed five-year ban on facial recognition.

29 min watch Watch Now Events & Announcements Event: AI Trends Shaping 2020 w/ the NBA, Verizon and Etsy
Date: January 27, 6:30PM-8:30PM
Join General Assembly in their first meetup of the year as they bring in data experts across industries to discuss AI Trends that will shape 2020 including. Panelists include Jordy Kieto (Data Scientist at the NBA), Moumita Bhattacharya (Senior Data Scientist at Etsy), and Sanjit Paliwal (Senior Data Scientist at Verizon). Register Here.

Event: Data Science Tech Talks
Date: January 28, 6:15PM-7:45PM
Orson Adams introduces you to MLFlow Tracking — open-source tooling to track machine learning experiments; and Daulet Nurmanbetov covers how data programming and weak supervision can create a labeled dataset oftentimes cheaper and faster than labeling data by hand. Register Here.

Event: MetaProp 2019/2020 New York Demo Day
Date: February 5, 6PM-9PM
The MetaProp Accelerator @ Columbia University class of 2019/2020 is ramping up for Demo Day, the culmination of months of intense growth focus and is a rare chance for members of the public to see/hear from the latest crop of high potential PropTech startups. Register Here. A Deeper Look Worried About Privacy at Home? There’s an AI for That

AI’s data-hungry nature seems to permit companies to send your conversations with voice assistants to contract workers, or put your data at risk for breach in the cloud. But do we need all household appliances to quote Wikipedia and find local movie times? Wired author Clive Thompson thinks there’s a workaround: edge AI devices can enable functional, less-smart AI to process on-chip. Not only is edge AI fast, it’s more secure and more energy efficient.

Google’s little know project, Coral, is one example. The company’s accelerators and dev boards are meant to quickly prototype ideas offline with the help of Google’s Edge TPU (“a (very) little brother to the water-cooled TPU used in Google’s cloud servers,” according to this Verge feature). Xnor.ai “made an image-recognition neural net so lightweight, it can be fueled by a small solar cell,” and was recently acquired by Apple.

5 min read

Read More The Machines Are Whispering: We Tested AI Dungeon 2 and Cannot Stop Laughing “First of all, this isn’t a game. It’s interesting, but it’s not a game. More like computer-assisted literary masturbation.”

We know this section isn’t called “A Vapid Look.” We know we spoke about AI Dungeon 2 already. But just like talktotransformer.com, we can’t get enough of AI generated lore, especially when partially evokes early text adventures. And Ars Technica’s overview is informative and hilarious.

AI Dungeon mashes GPT-2 with 30MB of stories from ChooseYourStory.com, and offers players four settings: fantasy, mystery, apocalyptic, or zombie. The game that follows… isn’t much of a game. But it is an enormous well of amusement…