DataDownload: How Bellingcat predicted the Capitol siege

8 min readJan 9, 2021

DataDownload: How Bellingcat predicted the Capitol siege A weekly summary of all things Media, Data, Emerging Tech View this email in your browser

This week, we’re looking in the mirror.

The insurgent attack on our government didn’t just happen. It was conceived, socialized, shared, amplified, and ratified by technologies that were built and unleashed under our watch. So, now what?

David Kaye, who spoke at our SXSW Section 230 panel, has strong words about the de-platforming of the President. MIT Professor Sinan Aral looks at how big tech mobilized Trump’s DC army.
The Markup is getting some very interesting inside info on Facebook with their new project. And Cory Doctorow — a NYCML Summit speaker — is provocative and relevant as always. And Manoush Zomorodi on the TED Radio Hour explores the future of cities. She’s hosting our Wikipedia Birthday next week, it’s a Must Watch event.

In a rare bit of good news, publishers are gaining subscribers. Yea! And finally — Avocado armchairs. Erica did a deep dive on OpenAI’s new DALL·E and CLIP algorithms in her weekly Innovation Monitor Newsletter (what? You don’t subscribe? fix that right away!)

Pace yourself.

Steve

Steven Rosenbaum
Executive Director
The NYC Media Lab
Steve@NYCMediaLab.org Must-Read How the Insurgent and MAGA Right Are Being Welded Together on the Streets of Washington D.C.

A day before the Capitol was briefly overtaken in a failed insurrection attempt, Bellingcat posted an excellent investigative dive that pointed to clear signs that this rally was going to be anything but peaceful… “If it follows the same pattern as the previous gatherings, the day will be filled with mostly peaceful speeches and marches while the night will bear witness to horrific street violence.”

In bellingcat.com Conflict journalist Robert Evans walks through extremist social media chatter filled with ominous hints of what came the day after, from suggestions to “storm the Capitol”, to discussions of burning it down, to protest-goers posting that they are ready to die. It’s a disturbing read, but offers some insight on how to spot signs of upcoming violence on extremist platforms.

15 min read

After years of giving Trump a platform to disseminate false beliefs and propaganda, social media giants have started to cut off the President’s access to tens of millions in light of the Capitol siege. The alternatives are popular echo chamber — Parler, Spreely, Gab, and other relatively small, highly-insulated communities.

“So he goes to Parler. Who cares?” said David Kaye, a law professor and former UN special rapporteur on free expression in the NY Times. “The major platforms have already demonstrated the undeniable value of network and reach beyond the narrow confines of an affinity group.” Kaye pointed to Alex Jones and Milo Yiannopoulos, who both struggled to stay relevant after their accounts were banned on major platforms.

5 min read Read More Tech+Media How We Built a Facebook Inspector

In 2018, Facebook committed to sharing a petabyte of its data for researchers. Instead, after two years of delays, “researchers were given access to an extremely limited dataset and CrowdTangle, a social analytics firm owned by Facebook.”

Unfortunately, CrowdTangle doesn’t provide data about the number of times users see content. To get a clearer picture of Facebook’s algorithms at work, The Markup paid over 1k participants to periodically submit their Facebook data to the Citizen Browser project (which was actually announced in October). We’ll keep you posted when the results come in.

5 min read Read More Cory Doctorow: Neofeudalism and the Digital Manor “Apple has now arrogated to itself the power to know, with a reasonable degree of granularity, which programs its customers are using, and to decide whether customers should be permitted to do so.”

In November hacker and security researcher Jeffrey Paul published Your Computer Isn’t Yours, a condemning piece on macOS’s invasive practices. The knowledge that Apple was keeping track of their users — and also open to giving up that data to federal agencies… and the Chinese government — caused an uproar. In Cory Doctorow’s essay in sci-fi mag Locus, he borrows a phrase from security researcher Bruce Schneier to describe the double-edged behavior — feudal security.

Here’s the scenario. Nefarious actors — bandits — are out to get you, and they are awfully good at breaching security measures. “To be safe, then, you have to ally yourself with a warlord. Apple, Google, Facebook, Microsoft, and a few others have built massive fortresses bristling with defenses.”

But who will help you when the warlords turn on you? Google, for example, is blocking commercial surveillance on Chrome — except Google’s: “if a marketer pays Google, and convinces Google’s gatekeepers that it is not a scumbag, Google will allow them to spy on you.” Facebook… well, just see The Markup piece above on why the company can’t be trusted to keep to its policies.

15 min read Read More Publishers Expect to Pick Up the Pace of Innovation and Rely on Readers More Than Advertisers in 2021

The Reuters Institute for the Study of Journalism released a new report surveying 234 media leaders about their thoughts on the next 12 months in journalism. Here are a few highlights:

“More than 75 percent of publishers said the coronavirus has accelerated their transition to an all-digital future.”
“With little money available for big new investments, companies are likely to focus on improvements to existing products and brands (70%) rather than developing ‘moonshots’ or creating entirely new products (28%).”
“Of the publishers surveyed, 36 percent said policy interventions could help journalism, compared to just 18 percent this time last year.”

5 min read

The Hype Machine author and MIT David Austin Professor of Management Sinan Aral discusses the role of Big Tech in violence and conspiracy theories with Yahoo Finance.

4 min watch

Watch Now What We’re Listening To Podcast: The Life Cycles Of Cities

“Cities are never static; they can transform in months, years, or centuries.” In this podcast, TED speakers “explore how today’s cities are informed by the past, and how they’ll need to evolve for the future.”

52 min listen

Listen Now Virtual Events

Free Event: NYC Media Lab Celebrates Wikipedia’s 20th Birthday
Date: January 15, 1PM EST
January 15, 2021 marks Wikipedia’s 20th birthday, and NYC Media Lab is celebrating with some of the brightest minds in media. Join us for a virtual chat with Wikimedia Foundation CEO Katherine Maher and host of NPR’s TED Radio Hour, Manoush Zomorodi. Manoush and Katherine will talk about Wikipedia’s role in the ongoing fight against mis and disinformation, address some of the biggest misconceptions you hear about Wikipedia, and discuss how research and knowledge sharing may change over the next 20 years. Free RSVP.

Free/Paid Event: The Future of Health & Moonshot Thinking w/ Vinod Khosla
Date: January 12, 3PM-4PM EST
A virtual, interactive Fireside Chat with Vinod Khosla, founder of Khosla Ventures, to discuss the future of health in a post-pandemic world and why moonshot thinking is more important than ever. Register Here. A Deeper Look This Avocado Armchair Could Be the Future of AI

“In the long run, you’re going to have models which understand both text and images. AI will be able to understand language better because it can see what words and sentences mean.” — Ilya Sutskever, chief scientist at OpenAI

GPT-3 has proved its versatility beyond generating English sentences — developers granted beta API access have generated SQL from text input, created UIs in Figma, and developed plenty of other projects outside typical NLP use cases. Around the same time GPT-3 was announced, OpenAI published some fascinating research on image generation: “When we train GPT-2 on images unrolled into long sequences of pixels, which we call iGPT, we find that the model appears to understand 2-D image characteristics such as object appearance and category.” As Sutskever notes in the above quote, the ability to understand language through images (and vice versa) grounds AI in reality more than GPT-3 could do with text alone.

This week OpenAI released two new models — DALL·E and CLIP. For DALL·E (“a portmanteau of the artist Salvador Dalí and Pixar’s WALL·E,” according to the recent blog), the company trained a smaller GPT-3 model on text-image pairs; the resulting model generates convincing images from descriptions alone, and when it works, the output is quite impressive.

For CLIP, OpenAI also used a text-image dataset, but for a different goal — matching a caption to an image. According to the blog published on the same day as DALL·E’s, “this data is used to create the following proxy training task for CLIP: given an image, predict which out of a set of 32,768 randomly sampled text snippets, was actually paired with it in our dataset.” The CLIP model was actually used to rank the examples in the DALL·E blog. (Aside: CLIP is currently on GitHub. There’s also an unofficial repo for DALL·E.)