Good Monday morning. It’s November 2nd. We’ve got information for you below about a great new free tool from First Draft that shows disinformation in social media posts, ads, and elsewhere online. This week promises to be unlike any we’ve ever faced so please practice self-care and don’t believe everything that you read or hear.
Today’s Spotlight is 1,121 words — about a 4 minute read.
1. News to Know Now
a. Amazon, Facebook, Google, Apple, and Twitter all announced earnings Thursday following another round of testimony on Capitol Hill. No one in that group is hurting. Amazon’s profit soared to over $6 billion for the quarter and Facebook added more advertisers despite an advertising boycott that included dozens of brands. (AdAge)
Wow moment: Amazon has saved one billion dollars in travel expenses so far this year.
b. Twitter continues labeling disinformation. The company started last week by flagging a dishonest tweet by President Donald Trump that claimed that there were problems and discrepancies with mailed-in ballots throughout the country. That is not true. Twitter announced last week that it will use headers and images on its site this week to show accurate voting information. (New York Times)
c. Facebook has told New York University researchers that they may not use information downloaded by the team using software that it built to access its political ads library. Facebook says its rules prohibit third party software from downloading the contents. (CNN)
2. COVID-19 Online Resources and News
COVID-19 Tech News
Eight In-Store Innovations for the COVID-19 Era – Econsultancy
MIT: AI Identifies Asymptomatic COVID-19 Carriers – Venture Beat
Older People, Republicans Share Inaccurate COVID News – Nieman Lab
Post election, Vaccine is Biggest Disinformation Threat – CNBC
SF Stops Google-affiliated Testing After Results Take 10 Days – SF Gate
The Challenges of Contact Tracing as U.S. Battles COVID-19 – Pew
3. Search Engine News
Google continues remaking Google My Business listings into a profit center by selling upgraded proafiles and the “Google Guaranteed Badge” for a $50 monthly fee. The process includes background checks for customer-facing employees, insurance verification, and appropriate license checks. Working directly with small businesses allows Google to build direct relationships with organizations that typically use a third party when interacting with the company.
What we think: The $600 annual fee is inexpensive, but still a budget-sized item that was previously free. We’ll be advising customers that we’ll monitor performance on Google My Business since it is now effectively paid advertising.
4. In the Spotlight — Machine Learning
Let’s stop using the term AI as a synonym for machine learning. Machine learning uses a lot of data to create software that is capable of determining whether new input fits that pattern. Image recognition and malware scanning are two common applications of machine learning.
We’ve told you in the past about GPT-3, which is a deep learning model that can create human-sounding text when prompted. Above, in the COVID resources section, we link to a fascinating article about an MIT research team that used 200,000 audio samples of people coughing, including some who were infected with COVID-19, to train their model. That is a level of audio analysis that is beyond human limits.
The world is racing to train machine learning algorithms to handle all sorts of analysis that was previously thought impossible. Being human, there are good and bad applications for this technology.
A machine learning algorithm that was trained on nude imagery is being used as an automated chatbot to create deepfake nudes of ordinary people. The user uploads an image of a clothed woman, and the algorithm removes the clothing while building a credible-looking nude image of the woman.
More than 104,000 women had their images faked in this way by midsummer. Research by technology firm Sensity found that over 60% of those images were subjects known to the individual while another 15% were celebrities. Sensity also reported that a limited number of images appeared to feature children. The images carry a watermark from the software that can be removed by purchasing higher levels of access.
This level of technology is commonplace. Microsoft has a new program for software developers called Lobe that automates machine learning of images so that anyone can create a training model by uploading labeled images to the software. I’ve reviewed the initial videos and anyone can easily be taught to train the model.
On Thursday, Google announced URL2Video, a software tool that converts a website page into a 12 second video. Google says that it is now working on generating audio for the video created from a web URL as well as voice-overs.
Our take: For more than forty years technologists spent much of their time making information digitally accessible. The next phase — telling stories about that information — is here. When consumers first began using automobiles, they needed to understand how to repair them and spent much of their time maintaining them. Many of today’s sophisticated automobiles can’t be repaired without a mechanic’s specialized equipment.
5. Debunked: The First Draft Dashboard
First Draft has published a must-use online dashboard dealing with election misinformation. They’re a trusted source funded by Google, Facebook, and multiple tech companies who are heavily invested in cleaning up misinformation and disinformation appearing online.
At this link you’ll find ads, identified misinformation, Twitter feeds, and reliable news and information. I can’t stress enough how much you need to bookmark this website for this week.
6. Following Up: Self-Driving Data
We’ve told you that Alphabet’s Waymo unit has introduced self-driving taxis in Phoenix. This week Waymo released public road testing data from January 2019 through September 2020.
7. Protip: Changing Messenger Themes
Facebook’s new Messenger interface comes with some nifty themes including Pride, Love, and Tie-Dye, as well as different emojis and colors.
Screening Room: Vipps
9. Coffee Break: MIT’s Nightmare Machine
A little more machine learning for you before you go to face this first week in November. MIT has created a website that shows scary images one of its algorithms creates. You get to click through some of them and help train the model on whether you think it’s a scary picture.
I was going to show you this last week, but I was distracted when the U.S. government decided to sue Google.