AI

Inside a Hot-Button Research Paper

On Dec. 2, Dr. Timnit Gebru, the co-lead of Google’s Ethical AI team, tweeted that she had been fired after a dispute over a research paper. The next week, search engine traffic for “large language models”—the focus of the contested paper—went from zero to 100 (literally). 

Recap: Large language models (e.g., GPT-3) are key in generating human-like text, and use cases vary widely—from analyzing tomes of medical research to auto-generating email responses. Generally, the more parameters they're trained on, the smarter they appear. 

Now that the paper has been accepted into a global computer science conference, we chatted with Dr. Emily M. Bender, its co-author and a professor of linguistics at the University of Washington, about top-level takeaways—and the future of AI ethics. 

This interview has been edited for length and clarity. 

Key finding

Bender: The system learns to recognize patterns from the training data, and it reproduces those patterns when it’s deployed. That’s really important in understanding what language models do and don't do. 

When they spit out this well-formed looking English, we take it in and apply meaning. It's easy to be fooled into thinking that the computer “said” something, when in fact it just reproduced some patterns. 

It’s also about putting some thought into the patterns that it's matching: Do we want those models out in the world taking action—by putting text into the world or being part of classification systems—given that the training data has lots of patterns that we don't want reproduced? These are patterns of systems of oppression, racism, sexism, transphobia, and so on.

The future of corporate AI ethics 

Bender: It’s easy to say, “Oh, corporations could never do this [ethics work]—they shouldn't even be pretending to.” But that does a discredit to the researchers who are employed by the corporations doing really fantastic work. 

If we want to get to a place where the technology we build serves humanity broadly—and not just the interests of people who are making the most money off of it—we need an all-hands-on-deck situation. 

For the rest of our conversation, click here to read the full Q&A with Bender.

BIG TECH

Earnings Recap

Big Tech and Big Tesla reported earnings last week. Here are the highlights you missed. 

Apple: Posted its strongest quarter ever, with mobile sales up 17% year over year. 5G should continue to give Apple a boost through 2021. Apple’s accessories division pulled in $10 billion in Q4, meaning it’s a bigger moneymaker than Mac. CEO Tim Cook said that wearables alone are the size of a Fortune 150 company. 

Microsoft: Blew away analysts’ expectations. The company’s core businesses grew double digits YoY and most notably, Azure revenues are up 50%. In Q3, Azure grew 48%. AWS is definitely watching its back...  

Facebook: Strong Q4, but its daily active user base shrunk in the US, its most lucrative market. “Other” revenue—from Oculus and Portal—leaped 156% YoY to $885 million for Q4. Total revenue was $28.1 billion.

• FB warned that iOS privacy changes will clip the wings of its ad business.

Tesla: Missed on earnings but beat on revenue. Tesla will launch a “Full Self-Driving” subscription service and start making the Model Y in Austin, TX, and Brandenburg, Germany, with in-house battery cells this year.

SPONSORED BY OPENREEL

Produce Professional Video From the Couch

Or from just about anywhere in the world, actually.

OpenReel is the remote video creation platform that can help your distributed global teams produce HD/4K videos at scale.

Put on a fancy director’s beret and your favorite sweats, because you can now virtually control the camera on a subject’s smartphone, tablet, or computer, eliminating the need for travel, expensive equipment, or set-up and breakdown.

Now, let’s bring this thing further into focus: OpenReel’s flexible remote video capture technology allows your brand’s entire crew—from tech to comms to creative—to collaborate on all elements of production in a way that’s more affordable, more sustainable, and less time-consuming.

Brands like Verizon Media, ViacomCBS, and Boston Scientific use OpenReel to tell their stories—see how it can be used to tell yours.

Put your remote video creation into motion with OpenReel today.

SMART HOME

Ring Ring

Ring, Amazon’s smart doorbell division, is partnered with over 2,000 local police and fire departments around the US, according to a tally from the FT. Under the partnership, authorities can request timestamped footage, which is captured by the motion-sensor-equipped smart cameras.

Non-partnered departments can also subpoena Amazon for video, which is becoming more common (but the company’s rate of compliance is falling).  

By the numbers

• The vast majority of departments partnered with Ring are local police.
• Montana and Wyoming are the only two US states where Ring doesn’t have a partnership with the police. 
• In 2020, the departments collectively requested videos for 22,335+ incidents.

Ring is controversial. Critics say it’s building a privatized, networked Eye of Sauron, camera by camera, in US neighborhoods. Ring’s service may provide solace to individual homeowners, but it may also trample on the privacy of passersby on the streets. 

• Ring’s app has also had its fair share of security issues. Ring enabled opt-in, end-to-end encryption in January. 

Zoom out: Surveillance cameras can be found in nearly 33% of smart homes in the US, the UK, Germany, and France, according to Strategy Analytics research released last summer. 

BITS & BYTES

Stat: Smartphone shipments in China fell 4% year over year to 84 million units in Q4 2020, according to Canalysis. Shipments for 2020 fell 11% YoY to 330 million. 

Quote: “One hell of an invention.”—Ray Dalio, on Bitcoin. 

Read: Our first product review. 

SPONSORED BY ASANA

Work about work isn’t working. In fact, searching for info, bouncing between apps, and other busywork costs knowledge workers in the US 61% of their time—time that could be spent doing actual work. And enjoying more than 23 seconds for lunch. In their Anatomy of Work Index report, the work management wizards of Asana dig into what it will take to overcome disruption and thrive in a distributed world. Read the full report right here.

WHAT ELSE IS BREWING

• Neuralink has an Austin office, per an Elon Musk tweet from yesterday. We checked Neuralink’s job board and, sure enough, the brain-machine interface company is hiring for a head of construction in Austin. 
• Myanmar’s military seized power in a coup against the democratically elected government and appears to have throttled the internet at a national level. 
• Developers are finding creative ways to skirt the App Store’s new privacy nutrition labels listing requirements. 
• Some Hyundai execs are worried about the prospect of becoming a contract manufacturer for Apple, Reuters reports. 
• Huawei officials retweeted fake accounts to help spread a pro-5G influence campaign in Belgium. 
• TSMC is fast-tracking auto chip production to alleviate the global shortage of semiconductors for vehicles. 

THREE THINGS WE’RE WATCHING

Tuesday: Alibaba reports earnings. How is the company handling Chinese authorities’ antimonopoly investigation into the e-commerce conglomerate? And where is Alibaba founder Jack Ma? Also on Tuesday: Google, Amazon, Match Group, Electronic Arts, and NXP Semiconductors report earnings. 

Hump Day and the day after: The earnings don’t stop. On Wednesday, we’ll be watching Qualcomm for 5G news. On Thursday, we’ll be wearing our metaverse analyst hats and listening to Unity’s earnings for updates...and checking on Snap for AR updates. 

Friday: SMIC, China’s biggest semiconductor manufacturer, reports earnings. We’ll be interested to see how the company fared in Q4, when the US placed more restrictions on SMIC, and how it’s 2021 outlook is taking shape. 

PATENTS ARE A VIRTUE

After two years in limbo, Spotify has been granted a patent for using speech recognition to analyze a user’s “emotional state, gender, age, or accent”—then potentially using the results to recommend music or artists.

According to the patent, Spotify might then use the data to categorize a user’s emotional state as happy, angry, sad, or neutral. And it’s not just speech recognition—there’s also the potential to analyze background noise, reports Music Business Worldwide. 

Spotify, we’ll make things easier for you: When Hayden and Ryan are sad, you’ll find them both listening to LCD Soundsystem.

ICYMI

Catch up on the top Emerging Tech Brew stories from the past few editions: 

• Review: 2021 Ford Mustang Mach-E 
• Nine experts on the biggest obstacle facing AI in the next 5 years
• Spin + Tortoise = teleoperated e-scooters