Alphabet’s Loon internet balloons are now being piloted by a machine-designed deep reinforcement learning system. The AI-powered flight control system is now navigating the stratospheric helium balloons over Kenya, where Loon launched its initial commercial internet service this past summer.

More:

• The machine-devised navigation software can calculate the optimal navigation path for Loon balloons more quickly and efficiently than a human-made system.
• The AI, developed by Loon and Google AI, can maintain balloons in target areas, where they stay aloft in optimal conditions. The balloons use noticeably less power than human-made algorithms.
• After pitting the AI against the human-created system in a 39-day test over the Pacific Ocean, the AI model could keep balloons aloft for longer periods and even devised its own novel navigational moves, prompting Loon to deploy it across its entire fleet.
• According to Loon, the system marks the first use of a reinforcement learning model in "a production aerospace system." Reinforcement learning refers to the technique whereby software teaches itself through trial and error.

Salesforce, which uses AI to turn customer data into business insight, has acquired Slack for $27.7B. The sale marks Salesforce's largest acquisition to date and the six largest tech acquisitions overall. 

More:

• In addition to the acqusition, Salesforce announced the debut of Einstein Automate, which is a set of AI-powered workflow solutions for customers.
• Key elements include Flow Orchestrator, an AI-based low-code workflow development tool, and MuleSoft Composer to connect apps and data flows and automate workflows.
• Einstein refers to the CRM company's commercial AI platform, which automates many of its processes.
• Salesforce acquired Tableau in 2019 to compete with Microsoft's Power BI. It could be acquiring Slack (12.5 million DAUs as of March 2020) to compete with Microsoft Teams, which has 115 million users. You can read more about Salesforce vs. Microsoft here. 
• Salesforce launched its enterprise social network tool Chatter in 2009 and bought the collaborative word processing app Quip in 2016. 
• The company also acquired professional services firm Acumen Solutions, which provides digital solutions using Salesforce to customers across various industries. The value of the deal was undisclosed.

Some academics have expressed skepticism over DeepMind's AlphaFold program, which purports to have solved what's known as “the protein folding problem." While many people have lauded the AI system as a way to better understand diseases and develop new drugs, some experts are saying it's overhyped and is more of a step forward rather than a complete solution, Business Insider reports.

Background:

• Alphabet-owned Deepmind announced AlphaFold's progress this week, saying that the system can quickly and accurately predict the 3D shape of a protein, based on its string of amino acids.
• A protein's shape defines what they can and cannot do. But identifying the shape has been a difficult problem for scientists for decades.
• Alphafold can predict what structure proteins will fold into in only days; knowing their shape could help researchers determine how effective certain drugs could be on the human body.
• A group called CASP, or Critical Assessment for Structure Prediction, said AlphaFold has achieved "unparalleled levels of accuracy" in protein structure prediction. Nobel Prize-winning Professor Venki Ramakrishnan called the results a "stunning" achievement.

Criticisms:

• Computer science associate professor Max Little, from the University of Birmingham, argued that AlphaFold only has potential within the CASP database challenge. Right now, "We can't really be sure how well AlphaFold will work when faced with the far more rich and varied array of proteins found in the real world of living organisms," he said.
• Professor Michael Thompson, a structural biology expert at the University of California, Merced, called it "laughable" that AlphaFold has actually solved the protein folding problem. Until DeepMind shares the code, "nobody in the field cares and it's just them patting themselves on the back," he said.
• Lior Pachter, a California Institute of Technology professor of computational biology, said he doesn't mind that Google hyped the AI, but he does care "that many (computational) biologists who ought to know better are going around screaming: 'Protein folding is solved!'" 
• DeepMind’s chief executive and co-founder Demis Hassabis said the research lab plans to publish more details about its work in 2021 at the earliest.

Facial recognition algorithms have improved their ability to recognize people's faces when they're wearing masks, according to the latest research by the National Institute for Standards and Technology (NIST). Error rates are still higher with masks (about 5%) than without (0.3%), the study shows.

More:

• The research shows improvements in facial recognition compared to a July NIST study, which showed that facial recognition algorithms from Panasonic, Tencent, and elsewhere failed up to 50% of the time when tested on faces wearing digitally-applied masks.
• Starting in May, NIST began analyzing independent testing of different facial recognition algorithms to find out that they were better at mask identification than initially thought.
• The latest study examined 65 algorithms that were submitted after face masks became a requirement in many countries. They were tested on 6.2 million photos.
• Error rates were higher when masks covered more of a person's face. Round masks had fewer errors than lengthier ones, while black and red masks had more errors than light blue and white masks.

In related news:

• Massachusetts lawmakers just approved a bill that would bar law enforcement statewide from using facial recognition, which has come under scrutiny for its racial, gender, and age-related biases.

AWS launched Amazon SageMaker Data Wrangler, which speeds up data preparation for machine learning training. The service was one of several AI/ML features announced during AWS re:Invent 2020, a three-week virtual conference that kicked off on Nov. 30.

More:

• AWS said it's launching SageMaker Feature Store, which makes it easier to name, organize, find, and share inputs in machine learning systems in SageMaker Studio
• Sagemaker Pipelines, another new service, provides a CI/CD service for AI/ML to develop and automate workflows and an audit trail for model components.
• In addition, the company announced five new machine learning services—Amazon Monitron, Amazon Lookout for Equipment, the AWS Panorama Appliance, the AWS Panorama SDK, and Amazon Lookout for Vision—to help the industrial and manufacturing embed intelligence in their production.
• During the conference, AWS CEO Andy Jassy noted that data preparation is still a big challenge in machine learning. Data scientists often focus more on the infrastructure to build the models rather than actually building them, he noted, though Amazon's improvements should help with this.

New job postings in AI/ML:

AI Success Director
DataRobot | Washington, D.C.

Technical Project Manager - AI Research and Development - ML Platform
TuSimple | San Francisco...

QUICK HITS:

• During this holiday season, find out how SimpliSafe is protecting over 3 million homes with less markup and more security.*
• Qualcomm says its new Snapdragon 888 flagship smartphone processor has more powerful AI.
• Google used AI to sync all YouTube covers of Billie Eilish’s "Bad Guy" into an infinite loop.
• MIT engineers developed an AI algorithm named RoboGrammer that can design its own robot bodies.
• A research grant will fund a Penn State project that seeks to test for AI risk and vulnerability at the development stage.
• The U.S. Defense Department is working to scale and adopt AI across all of its operations, according to Michael Groen, director of the department's Joint Artificial Intelligence Center, which is leading the effort.
• Revtown’s changing the denim game by applying the best parts of workout clothes—comfort, flexibility & durability—to jeans.*

*This is a sponsored post.

Tweet of the Day: Mathematician and broadcaster Hannah Fry hosts "DeepMind: the Podcast" about the London AI lab's rseaecrh and discoveries.