Hello Inside readers! We recently relaunched our Inside Amazon newsletter – subscribe here to learn about all things Amazon/Bezos/AWS/Whole Foods/Twitch and more every Monday and Wednesday.
|
|
|
Welcome to Wednesday's Inside AI! In today's issue:
- Facebook is creating internal teams to study and search for racial bias in its algorithms.
- A deep learning model can predict the risk of a coronavirus patient developing severe illness.
- AI Masterclass: Questions that every executive should ask about their AI algorithms (premium only).
To enjoy this content and more, consider upgrading to premium! For a limited time, we are offering a 14-day free trial. If you like what you see, you can subscribe for less than $2 a week ($100/year), or $10 per month. For that, you'll receive Inside AI advert-free, along with content only our paid subscribers receive. Click here to give it a try.
Thanks again for your support,
|
|
|
Facebook is creating teams to search for racial bias in its algorithms, which comes after civil rights groups organized an advertising boycott against the company. The internal teams will specifically look into how machine learning systems impact Black and other minority groups on Facebook and Instagram. The algorithms automatically filter both the content and advertising that users see on the platforms.
More:
- Facebook and Instagram will both launch "equity teams" to study algorithms as well as analyze the company's enforcement of harassment policies, among other issues.
- MIT Tech Review reports Facebook is "well overdue" for an internal assessment of its machine learning algorithms, which have already been shown to discriminate against certain racial groups.
- Tuesday's announcement is widely viewed as an acknowledgment by the company that its AI-trained algorithms are unintentionally encoded with internal biases. This bias could come from inaccurate or non-representative data.
Related:
- The news comes after civil rights groups - such as the NAACP and Anti-Defamation League - organized an advertising boycott against Facebook, which prompted large corporations like Coca-Cola, Disney, and McDonald's to suspend their ad campaigns.
- Earlier this month, Facebook released the results of a civil-rights audit that said it should improve its AI tools for identifying racism, hate speech, and other dangerous content. The report said Facebook is not attuned enough to the way its algorithms inadvertently feed "extreme and polarizing content."
- As of spring 2019, 4% of Facebook employees are Black, according to an AI Now Institute survey. Facebook declined to provide diversity numbers about its AI Research unit. Experts have noted the potential for bias when employee groups are not diverse enough.
- Researchers recently trained a machine learning model to identify social media posts by foreign adversaries seeking to manipulate political campaigns.
THE WALL STREET JOURNAL
|
|
A deep learning model can predict the risk of a coronavirus patient developing severe illness. The newly published model comes from the Tencent AI Lab and a team of Chinese health scientists led by Zhong Nanshan, China's senior medical adviser on the virus. Medical professionals and others can now access the predictor online.
More:
- About 6.5% of coronavirus patients suddenly progress to severe illness. Among those patients, the mortality rate is 49%, the research published in Nature Communications shows.
- The model, called the Calculation Tool for Early Triage of Critically-ill COVID-19 Patients using Deep Learning, was created using data from 1,600 patients from various medical clinics.
- The risk level is calculated based on ten criteria, including age, cancer history, and X-ray abnormalities. It calculates the probability that a patient will develop severe illness within five, 10, and 30 days.
- A different Stanford algorithm predicts which coronavirus patients' conditions could deteriorate. Other research units have released similar AI models related to the virus.
SOUTH CHINA MORNING POST
|
|
SalesChoice CEO Dr. Cindy Gordon
AI Masterclass: Questions that every executive should ask about AI.
SalesChoice CEO Dr. Cindy Gordon explains why CEOs and boards of directors need to learn the ins and outs of AI and be able to answer specific questions about their models/algorithms.
- The questions include:
- What use case was the AI model/algorithm used for?
- What was the initial estimated value of the AI model and methods to the organization before designing, building, and implementing the use case?
- Who wrote the algorithm or built the AI model?...
|
|
Salesforce is retiring its Einstein Voice Assistant and Voice Skills features. The company will shut down the AI voice services and focus on its new product, the Salesforce Anywhere App, which also uses voice functionality. The larger Einstein AI platform for gleaning insights from customer data remains.
More:
- Voicebot.AI reports most of the voice assistant team is now working on the Salesforce Anywhere App, which uses software to localize an organization's chat, alerts, comments, and videos.
- Einstein Voice Assistant was a part of its AI-based Einstein technology for sales and marketing departments to gain more insights about customers and enhance its cloud products.
- Voice Assistant could transcribe, create tasks, provide metrics, and learn the slang and acronyms of specific organizations.
- Einstein Voice Skills, previously Einstein Voice Builder, helped users build voice-powered corporate apps.
- A Salesforce release note said the Voice Assistant was scheduled to be retired on July 10 for both the iOS and Android Salesforce mobile apps as well as the online Salesforce system.
- Voice capabilities are also built into other Salesforce products, like its Einstein Call Coaching and Service Cloud Voice.
- Earlier this month, Salesforce chief scientist Richard Socher announced he was leaving the company to launch his own startup. Socher, who worked on Einstein's AI capabilities, joined Salesforce when it acquired his AI startup MetaMind in 2016.
VOICEBOT.AI
|
|
ByteDance AILab and Shanghai Jiao Tong University researchers presented a paper about their news reporting and writing AI bot. Named Xiaomingbot, the bot can generate original news stories and also present as an animated avatar that reads the news in multiple languages.
More:
- The multilingual and multi-modal software bot summarizes Chinese news that it generates automatically from data tables, according to the research paper. It then translates those summaries into multiple languages to produce more articles.
- Its voice cloning technology synthesizes speech trained from a human's voice. Its animated avatar can generate, translate, and read the news in multiple languages.
- Researchers note that Xiaomingbot has already written more than 600k articles and gained 150k+ followers on social media.
- A Reddit poster wrote that the bot appears "kind of oversexualized."
SYNCED
|
|
Amazon's autonomous Scout robot has started making deliveries in Atlanta, Georgia, and Franklin, Tennessee. Human assistants will accompany the bots as they deliver packages during daylight hours Monday-Friday to meet increased customer demand during the pandemic.
More:
- The bots, which cut down on human-to-human contact, travel at a walking pace and can navigate around objects on the sidewalk.
- Amazon started testing the battery-powered bots in Snohomish County, Washington, in Jan. 2019, and expanded the service to Irvine, California, in Aug.
- Currently, human handlers opened the bots' storage hatch and unload packages if customers aren't at the doorstep. Amazon hasn't explained how that will happen once the bots are fully autonomous.
- The company said it also plans to partner with Atlanta and Franklin schools to support robotics and STEM activities.
TECHCRUNCH
|
|
Autonomous driving developer Aurora is expanding its self-driving truck and van tests to Dallas-Forth Worth in Texas. The company plans to test a variety of different commercial routes using a mix of Class 8 trucks and Chrysler Pacifica minivans.
More:
- Initially, Aurora will deploy a small fleet of Pacifica minivans, eventually expanding to a fleet of trucks by the end of 2020.
- In Oct. 2019, Aurora said it would apply its self-driving system for commercial purposes, including Class 8 trucks and commercial vans.
- Aurora's backed by Amazon, Hyundai, Shell, and Sequoia, among others.
- The company plans to open an office in the Dallas area that will employ about two dozen people.
A version of this story first appeared in Inside Transportation
|
|
Quick Hits
- Fiat Chrysler signed an exclusive agreement with Waymo to build a fleet of autonomous vans for Waymo’s delivery service, Via. Fiat also ended its 18-month relationship with Waymo competitor, Aurora.
- Facial recognition technology will be installed on 1,500 subway cars on Moscow's metro system by the end of 2020.
- A Redditor collected GPT3 demos explanation articles from Twitter and has shared them on GitHub.
- Most security companies send a technician to your home. SimpliSafe doesn’t. Order online. Set it up in 30 minutes. After that you’re protected 24/7.*
*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
|
|
|