Welcome to your Thursday edition of Inside AI!
Before you read on, we'd like to offer a quick reminder of all the benefits you'll get for signing up for our Premium edition. This includes access to our weekly job listings, Masterclass summaries, AI startup funding, and more, such as this week's roundup of the latest AI research and our podcast notes on the three AI federal white papers.
Right now, we're offering a 14-day free trial, which you can unsubscribe to at any time. To gain access to our complete AI newsletter, click here to upgrade to premium!
Thanks for your support,
|
|
|
A judge gave preliminary approval for Facebook's $650m settlement of a facial recognition lawsuit, one of the largest payouts ever related to data privacy. The suit argued Facebook violated an Illinois privacy law through the use of its "tag suggestions" feature.
More:
- The lawsuit - which the U.S. Supreme Court declined to hear - dates back nearly five years ago, when several law firms sued Facebook on behalf of users, arguing that the company's use of facial recognition in "tag suggestions" is illegal under the Illinois Biometric Information Privacy Act.
- The state law requires organizations to display a written policy about how they collect data and why, and explain if that data will be erased eventually. It also requires that people provide written consent for their biometric markers to be gathered and stored.
- The settlement is $100m more than Facebook's initial offer of $550m, which a U.S. district judge rejected in June.
- Judge James Donato, of the U.S. District Court for the Northern District of California, set a final approval hearing for Jan. 7, 2021.
- A newer class-action lawsuit accuses Facebook of illegally gathering biometric data through Instagram's photo-tagging tool. Facebook faces up to $500b in fines from that lawsuit.
- Last year, Facebook agreed to pay $5b as part of a settlement with the Federal Trade Commission, the largest penalty ever in a consumer privacy rights case.
REUTERS
|
|
A Michigan State University machine learning model found that mutations to the SARS-CoV-2 genome may have made the coronavirus more infectious. In an AI analysis of the virus' infection-causing spike protein, researchers found that five out of the six subtypes of the virus have become more infectious. The research was led by professor Guowei Wei from MSU's Mathematics and Biochemistry and Molecular Biology departments, whose research interests include AI, machine learning, and big data.
More:
- Wei and his team developed the ML model, an advanced neural network, to analyze SARS-CoV-2 genotyping from 20,000+ viral genome samples.
- The team has been tracking mutations for months in the official viral genome sample, which dates back to January.
- The paper, "Mutations strengthened SARS-CoV-2 infectivity, along with much of the published research concerning COVID-19 and the SARS-CoV-2 virus," was published on ArXiv.
Related:
PHYS.ORG
|
|
AI Masterclass: Trust in AI is key
Ajay Bhalla, president of cyber and intelligence solutions for Mastercard, stresses the importance of making sure AI is used ethically and transparently.
According to Bhalla, the pandemic has prompted more businesses to embrace AI technology. It's accelerated the volume and pace of data creation as more people work remotely and use digital transactions. This means AI, in particular, will play a bigger role in our work and personal lives.
This growth, however, should be balanced by the need to build consumer trust. Bhalla cites a report, commissioned by Mastercard, that delves into the importance of the issue, with interviews from senior thought leaders, practitioners, and ethicists. He summarizes the report further:
Trust must be earned
- There's been too much focus so far on what AI can do, and not how it can do it. Organizations now need to demonstrate that their systems and algorithms are responsible, fair, ethical, and explainable (aka, trustworthy).
- Well-known cases of AI misuse have put a dent in this consumer trust, and people are more aware of data privacy and other issues these days...
|
|
Aaron Goldfeder and Yue Ning
Panda AI, a new B2B startup, has raised its first round of funding. The company, which recently formed at Seattle's Allen Institute for Artificial Intelligence (AI2), spun out of AI2 and raised $3.3m from PSL Ventures, AI2, Ascend VC, DocuSign co-founder Court Lorenzini, and Smartsheet co-founder Eric Brown.
More:
- The startup says it uses natural language processing AI in "ridiculously practical ways" to work less but get more done.
- CEO Aaron Goldfeder, the co-founder of enterprise analytics company EnergySavvy, founded Panda with former Amazon software engineer Yue Ning, who started the Seattle NLP startup Civet AI. In 2019, Goldfeder and Ning joined AI2 as entrepreneur-in-residences. AI2 researchers Matthew Peters and Noah Smith also work with Panda.
- In a blog post, Goldfeder wrote that the way in which teams remain "on the same page about stuff is badly broken," adding that "we can do better" using today's technology.
- In January, AI2 raised $10M for its incubator for AI startups in deep learning, computer vision, and natural language processing. AI2 was created by Microsoft co-founder Paul Allen in 2014.
GEEKWIRE
|
|
YouTuber Denis Shiryaev used AI techniques to remaster the oldest film footage in existence. The three-second film, called Roundhay Garden Scene, was shot in the British city of Leeds on October 14, 1888, and features four people walking around a garden.
More:
- Shiryaev posted an AI-enhanced version of the video with 4K quality at 250 frames per second. The original only had 20 frames.
- He used neural networks to generate additional frames and add artificial coloring and face restoration.
- Shiryaev said he chose to apply "the full force of modern machine learning algorithms" to enhance the video, which was originally recorded by French inventor Louis Le Prince.
- Another AI model taps temporal convolutional neural networks to remaster old films.

YOUTUBE
|
|
Over 150 pre-recorded paper presentations for next week's European Conference on Computer Vision are now online. No ECCV registration is required to view the videos, which were shared by authors to the public for free. A Reddit thread cited some of the top papers chosen for the oral sessions:
CROSSMINDS.AI
|
|
QUICK HITS
*This is a sponsored post.
|
|
|
|
|
Beth is a tech writer and former investigative reporter for The Arizona Republic. A graduate of the Walter Cronkite School of Journalism, she won a First Amendment Award and a Pulitzer Prize nomination for reporting on the rising costs of public pensions.
|
|
Editor
|
|
|