Inside AI - December 5th, 2019

Inside AI (Dec 5th, 2019)

Subscribe | View in browser

1. After taking a sudden leave of absence earlier this year, DeepMind co-founder Mustafa Suleyman announced on Wednesday that he's moving to Google to work on the company's AI policy. On Twitter, Suleyman said he will begin working with Google's head of AI Jeff Dean and its chief legal officer Kent Walker, ending speculation that he would quit both Alphabet-owned DeepMind and Google due to reported tensions. Suleyman, who co-founded the DeepMind AI lab with Demis Hassabis and Shane Legg before Google acquired it in 2014, will join Google's own internal AI team, according to the company. - THE VERGE

2. U.S. citizens can choose to opt out of its proposed facial recognition checks, which would screen people at airports, Customs and Border Protection (CBP) said on Wednesday. The Department of Homeland Security wants to expand the checks to include U.S. citizens, who have so far been exempt from the screenings. (It's only been used on non-U.S. citizens traveling into the country). As part of its new proposal, the agency says that all travelers "may be required to be photographed upon entry and/or departure" in the future. As a result, Democratic Senator Ed Markey says he will propose legislation to block the proposed facial recognition checks, which have drawn criticism from groups like the ACLU. On Wednesday, CBP said U.S. citizens will be able to opt out of the checks by notifying an airline rep or CBP officer. People who choose to do so can "simply present their passport for visual inspection, as is standard practice at ports of entry today," it noted. - TRAVEL + LEISURE

3. China appears to be spending less on AI research than estimates have predicted, particularly when it comes to military-related research, according to a new report from the Center for Security and Emerging Technology (CSET). The D.C.-based think tank released preliminary findings that estimate the upper and lower bounds of China’s AI investments during 2018. For non-defense AI spending, China's upper-bound estimate was $5.7 billion in 2018, compared to the U.S.'s $1 billion in fiscal 2020. For defense AI spending, China's upper estimate was $2.7 billion, compared to $4.9 billion for the U.S. According to MIT, the report emphasizes how much "exaggeration and fear-mongering" dominate current conversations in D.C. involving AI strategy, with many saying the U.S. isn't investing enough compared to China. - MIT TECH REVIEW

4. Some reviewers of Amazon Web Services' DeepComposer are not sold on the new keyboard. The $99 keyboard is geared toward developers and links to a software interface that uses machine learning and cloud computing to create music based on what people play. But according to The Verge, many developers "don’t understand what to do with it," in part because it requires some understanding of music theory terms (and how to play a keyboard). In addition, the DeepComposer software already has a virtual keyboard. Amazon's version is actually a MIDI controller made by the Taiwanese company Midiplus, which has existed for years. Developers "could learn about the same basics of machine learning and music through any number of other interfaces, like Google Magenta," without paying as much, The Verge notes. Meanwhile, an expert on musical AI at Durham University said the recent audio demo provided by Amazon sounded "terrible," according to the BBC. - THE VERGE

5. MIT's ADEPT system could help inject infant-like cognition into future AI agents, according to the researchers who built it. ADEPT is capable of understanding the basic laws of physics and can predict how an object would behave next in video footage. The MIT team "wanted to capture and formalize that knowledge to build infant cognition into artificial-intelligence agents," said researcher Kevin A. Smith, noting that "We're now getting near human-like in the way models can pick apart basic implausible or plausible scenes." ADEPT works by examining an object to determine things like its shape and velocity, and simulating how that object should act based on the laws of physics. Once it predicts several outcomes, it compares those to what actually happens next in the video and sends out signals when discrepancies occur. In tests, its level of "surprise" matched that of humans. - ENGADGET

6. An AI expert at Nasa’s Jet Propulsion Laboratory says robots and other intelligent software are much more suited than humans to search for life beyond Earth. In an interview with The Irish Times, Dr. Steve Chien - the technical group supervisor of the lab's AI group - said "humans are very sensitive, very fragile" and aren't equipped to search our solar system for liquid water. Because of the distances involved and difficulties with communication, AI and robots "have to be smart enough to look on their own," he said, noting that when it comes to space exploration, we need more specialized intelligence rather than artificial general intelligence, when machines are considered as smart (generally) as humans. "So by no means are we saying, when we send an AI spacecraft to another world, that it has to be fully intelligent," Chien noted. - THE IRISH TIMES

7. Amazon is partnering with Novartis, one of the world's biggest pharmaceutical companies, to make both the manufacturing and delivery of new medicines more efficient. Using its artificial intelligence and machine learning expertise, Amazon Web Services will work with the Swiss drugmaker to help it forecast and track production, detect potential bottlenecks and monitor its sprawling operations in real-time. The partnership was announced Wednesday and signifies Amazon's latest push into the health care industry. The tech giant is already working with Berkshire Hathaway and JPMorgan Chase on an employee-health venture called Haven and is rebranding PillPack, the online pharmacy it bought last year. - BLOOMBERG

A version of this story first appeared in Inside Amazon.

8. A new AI that can read lips in videos leads the industry in two benchmarks. Researchers at Alibaba, Zhejiang University, and the Stevens Institute of Technology developed Lip by Speech (LIBS), which has speech recognizer and lip reader components based on an attention-based sequence-to-sequence architecture. While experts have already developed similar lip reading models (such as this one from Google and the University of Oxford that was introduced in 2016), LIBS beat two benchmarks in the character error rate baseline by a margin of 7.66 percent and 2.75 percent. LIBS could help people who have hearing loss follow videos without subtitles. - VENTURE BEAT

9. Drexel University officially opened its AI lab on Wednesday. Right now, the 5,000-square-foot space will support 100 students who are pursuing master's degrees while working with the IT company D.X.C. Technology. Researchers at the university's Multimedia and Information Security Lab have already developed a deep neural network that can spot deepfakes, after training it on a large database of videos from 46 different cameras. - WPVI

10. Apple employees will make presentations at next week's Conference and Workshop on Neural Information Processing Systems (NeurIPS) in Vancouver, the company announced Wednesday. In its Machine Learning Journal, Apple laid out the schedule for speakers, noting that its product teams conduct research in areas like machine hearing, speech recognition, and natural language processing. The conference begins December 8 and runs until December 14. - MAC RUMORS

Written and curated by Beth Duckett in Scottsdale, Arizona. Beth is a former reporter for The Arizona Republic who has published a book about the solar industry and frequently writes about hobby and commercial drones. You can follow her tweets about breaking news in artificial intelligence here.

Editor: Kim Lyons (Pittsburgh-based journalist and managing editor at Inside).

Copyright © 2020, All rights reserved.

Our mailing address is:
767 Bryant St. #203
San Francisco, CA 94107

Did someone forward this email to you? Head over to to get your very own free subscription!

You received this email because you subscribed to Inside AI. Click here to unsubscribe from Inside AI list or manage your subscriptions.

Subscribe to Inside AI