Connect with us


When AI sees a man, it thinks “official.” A woman? “Smile”



Sam Whitney (illustration), Getty Images

Men often judge women by their appearance. Turns out, computers do too.

When US and European researchers fed pictures of members of Congress to Google’s cloud image recognition service, the service applied three times as many annotations related to physical appearance to photos of women as it did to men. The top labels applied to men were “official” and “businessperson”; for women they were “smile” and “chin.”

“It results in women receiving a lower status stereotype: that women are there to look pretty and men are business leaders,” says Carsten Schwemmer, a postdoctoral researcher at GESIS Leibniz Institute for the Social Sciences in Köln, Germany. He worked on the study, published last week, with researchers from New York University, American University, University College Dublin, University of Michigan, and nonprofit California YIMBY.

The researchers administered their machine vision test to Google’s artificial intelligenceimage service and those of rivals Amazon and Microsoft. Crowdworkers were paid to review the annotations those services applied to official photos of lawmakers and images those lawmakers tweeted.

Google's AI image recognition service tended to see men like senator Steve Daines as businesspeople, but tagged women lawmakers like Lucille Roybal-Allard with terms related to their appearance.
Enlarge / Google’s AI image recognition service tended to see men like senator Steve Daines as businesspeople, but tagged women lawmakers like Lucille Roybal-Allard with terms related to their appearance.

Carsten Schwemmer

The AI services generally saw things human reviewers could also see in the photos. But they tended to notice different things about women and men, with women much more likely to be characterized by their appearance. Women lawmakers were often tagged with “girl” and “beauty.” The services had a tendency not to see women at all, failing to detect them more often than they failed to see men.

The study adds to evidence that algorithms do not see the world with mathematical detachment but instead tend to replicate or even amplify historical cultural biases. It was inspired in part by a 2018 project called Gender Shades that showed that Microsoft’s and IBM’s AI cloud services were very accurate at identifying the gender of white men, but very inaccurate at identifying the gender of Black women.

The new study was published last week, but the researchers had gathered data from the AI services in 2018. Experiments by WIRED using the official photos of 10 men and 10 women from the California State Senate suggest the study’s findings still hold.

Amazon's image-processing service Rekognition tagged images of some women California state senators including Ling Ling Chang, a Republican, as "girl" or "kid" but didn't apply similar labels to men lawmakers.
Enlarge / Amazon’s image-processing service Rekognition tagged images of some women California state senators including Ling Ling Chang, a Republican, as “girl” or “kid” but didn’t apply similar labels to men lawmakers.

Wired Staff via Amazon

All 20 lawmakers are smiling in their official photos. Google’s top suggested labels noted a smile for only one of the men, but for seven of the women. The company’s AI vision service labeled all 10 of the men as “businessperson,” often also with “official” or “white collar worker.” Only five of the women senators received one or more of those terms. Women also received appearance-related tags, such as “skin,” “hairstyle,” and “neck,” that were not applied to men.

Amazon and Microsoft’s services appeared to show less obvious bias, although Amazon reported being more than 99 percent sure that two of the 10 women senators were either a “girl” or “kid.” It didn’t suggest any of the 10 men were minors. Microsoft’s service identified the gender of all the men, but only eight of the women, calling one a man and not tagging a gender for another.

Google switched off its AI vision service’s gender detection earlier this year, saying that gender cannot be inferred from a person’s appearance. Tracy Frey, managing director of responsible AI at Google’s cloud division, says the company continues to work on reducing bias and welcomes outside input. “We always strive to be better and continue to collaborate with outside stakeholders—like academic researchers—to further our work in this space,” she says. Amazon and Microsoft declined to comment; both companies’ services recognize gender only as binary.

The US-European study was inspired in part by what happened when the researchers fed Google’s vision service a striking, award-winning image from Texas showing a Honduran toddler in tears as a US Border Patrol officer detained her mother. Google’s AI suggested labels including “fun,” with a score of 77 percent, higher than the 52 percent score it assigned the label “child.” WIRED got the same suggestion after uploading the image to Google’s service Wednesday.

Schwemmer and his colleagues began playing with Google’s service in hopes it could help them measure patterns in how people use images to talk about politics online. What he subsequently helped uncover about gender bias in the image services has convinced him the technology isn’t ready to be used by researchers that way, and that companies using such services could suffer unsavory consequences. “You could get a completely false image of reality,” he says. A company that used a skewed AI service to organize a large photo collection might inadvertently end up obscuring women businesspeople, indexing them instead by their smiles.

When this image won World Press Photo of the Year in 2019 one judge remarked that it showed "violence that is psychological." Google's image algorithms detected "fun."
Enlarge / When this image won World Press Photo of the Year in 2019 one judge remarked that it showed “violence that is psychological.” Google’s image algorithms detected “fun.”

Wired staff via Google

Prior research has found that prominent datasets of labeled photos used to train vision algorithms showed significant gender biases, for example showing women cooking and men shooting. The skew appeared to come in part from researchers collecting their images online, where the available photos reflect societal biases, for example by providing many more examples of businessmen than businesswomen. Machine learning software trained on those datasets was found to amplify the bias in the underlying photo collections.

Schwemmer believes biased training data may explain the bias the new study found in the tech giant’s AI services, but it’s impossible to know without full access to their systems.

Diagnosing and fixing shortcomings and biases in AI systems has become a hot research topic in recent years. The way humans can instantly absorb subtle context in an image while AI software is narrowly focused on patterns of pixels creates much potential for misunderstanding. The problem has become more pressing as algorithms get better at processing images. “Now they’re being deployed all over the place,” says Olga Russakovsky, an assistant professor at Princeton. “So we’d better make sure they’re doing the right things in the world and there are no unintended downstream consequences.”

One approach to the problem is to work on improving the training data that can be the root cause of biased machine learning systems. Russakovsky is part of a Princeton project working on a tool called REVISE that can automatically flag some biases baked into a collection of images, including along geographic and gender lines.

When the researchers applied the tool to the Open Images collection of 9 million photos maintained by Google, they found that men were more often tagged in outdoor scenes and sports fields than women. And men tagged with “sports uniform” were mostly outdoors playing sports like baseball, while women were indoors playing basketball or in a swimsuit. The Princeton team suggested adding more images showing women outdoors, including playing sports.

Google and its competitors in AI are themselves major contributors to research on fairness and bias in AI. That includes working on the idea of creating standardized ways to communicate the limitations and contents of AI software and datasets to developers—something like an AI nutrition label.

Google has developed a format called “model cards” and published cards for the face and object detection components of its cloud vision service. One claims Google’s face detector works more or less the same for different genders, but doesn’t mention other possible forms that AI gender bias might take.

This story originally appeared on

Continue Reading


The 2020 Atlantic hurricane season is finally over. What should we make of it?



Enlarge / All of 2020’s tropical storms and hurricanes in a single image.


Monday was the last “official” day of the Atlantic hurricane season, drawing down the curtain on what has been a frenetic year for storms forming in the Atlantic Ocean, Gulf of Mexico, and Caribbean Sea.

The top-line numbers are staggering: there were a total of 30 tropical storms and hurricanes, surpassing the previous record of 28 set in the year 2005. For only the second time, forecasters at the National Hurricane Center in Miami ran out of names and had to resort to using the Greek alphabet.

Of all those storms, 12 made landfall in the United States, obliterating the previous record of nine landfalling tropical storms or hurricanes set in 1916. The state of Louisiana alone experienced five landfalls. At least part of the state fell under coastal watches or warnings for tropical activity for a total of 474 hours this summer and fall. And Laura became the strongest hurricane to make landfall in the Pelican State since 1856.

Not all records broken

By some measures, however, this season was not all that extraordinary. Perhaps the best measurement of a season’s overall activity is not the number of named storms but rather its “accumulated cyclone energy,” or ACE, which sums up the intensity and duration of storms. So a weak, short-lived tropical storm counts for almost nothing, whereas a major, long-lived hurricane will quickly rack up dozens of points.

The ACE value for the 2020 Atlantic season to date is 179.8—and another weak tropical or subtropical storm could still form. This is notably higher than the climatological norm for ACE values (about 104), but it would not quite make the top 10 busiest Atlantic seasons on record, which is paced by the 1933 and 2005 seasons.

In terms of estimated damages, this season has been far from a record-breaker as well. So far, damages across the Atlantic basin are estimated at $37 billion. This is substantially less than the devastating 2017 season that included hurricanes Harvey and Irma and totaled more than $300 billion. It is also less than 2005, which featured Katrina, Rita, Wilma, and other storms topping $200 billion. One factor in 2020 was that most of the biggest storms missed heavily populated areas.

Also, the hyperactive Atlantic basin stands out amidst the other basins where tropical activity typically occurs, including the northeastern and northwestern Pacific Ocean, which were much quieter than normal this year. Overall, in 2020, the Northern Hemisphere is seeing an ACE value about 20 percent below normal levels for a calendar year.

Legacy of 2020

Perhaps the biggest legacy of this Atlantic hurricane season is the disturbing trend of tropical storms rapidly developing into strong hurricanes. This “rapid intensification” occurs when a storm’s maximum sustained winds increase by 35mph or more within the period of 24 hours, and it was observed in 10 storms this year.

Moreover, three late season storms—Delta, Eta, and Iota—increased their speeds by 100mph or more in 36 hours or less. Iota, which slammed into Nicaragua on November 17, was the latest Category 5 hurricane on record in the Atlantic.

Some recent studies, including a paper published by Nature Communications in 2019, have found that climate change has goosed intensification. The study observed, for the strongest storms, that rate of intensification over a 24-hour period increased by about 3 to 4 mph per decade from 1982 through 2009. Storms that strengthen more quickly, especially near landfall, leave coastal residents and emergency planners with less time and information to make vital preparations and calls for evacuation.

Continue Reading


Arecibo radio telescope’s massive instrument platform has collapsed



The immense instrument platform and the large collection of cables that supported it, all of which are now gone.

On Monday night, the enormous instrument platform that hung over the Arecibo radio telescope’s big dish collapsed due to the failure of the remaining cables supporting it. The risk of this sort of failure was the key motivation behind the National Science Foundation’s recent decision to shut down the observatory, as the potential for collapse made any attempt to repair the battered scope too dangerous for the people who would do the repairs.

Right now, details are sparse. The NSF has confirmed the collapse and says it will provide more information once it’s confirmed. A Twitter account from a user from Puerto Rico shared an image that shows the support towers that used to hold the cables that suspended the instrument platform over the dish, now with nothing but empty space between them.

The immense weight of the platform undoubtedly caused significant damage to the disk below. The huge metal cables that had supported it would likely have spread the damage well beyond where the platform landed. It’s safe to say that there is very little left of the instrument that’s in any shape to repair.

It’s precisely this sort of catastrophic event that motivated the NSF to shut down the instrument, a decision made less than two weeks ago. The separate failures of two cables earlier in the year suggested that the support system was in a fragile state, and the risks of another cable snapping in the vicinity of any human inspectors made even evaluating the strength of the remaining cables unacceptably risky. It’s difficult to describe the danger posed by the sudden release of tension in a metal cable that’s well over a hundred meters long and several centimeters thick.

With inspection considered too risky, repair and refurbishment were completely out of the question. The NSF took a lot of criticism from fans of the telescope in response to its decision, but the collapse both justifies the original decision and obviates the possibility of any alternatives, as more recent images indicate that portions of the support towers came down as well.

The resistance the NSF faced was understandable. The instrument played an important role in scientific history and was still being used when funding was available, as it provided some capabilities that were difficult to replicate elsewhere. It also played a role as the most important scientific facility in Puerto Rico, drawing scientists from elsewhere who engaged with the local research community and helped inspire students on the island to go into science. And beyond all that, it was iconic—until recently, there was nothing else like it, which made it a feature in popular culture and extended its draw well beyond the island where it was located.

Lots of its fans were sad to contemplate its end and held out hope that some other future could be possible for it. With yesterday’s collapse, the focus will have to shift to whether there’s a way to use its site for something that appropriately honors its legacy.

Continue Reading


Russian spaceport officials are being sacked left and right



Vladimir Putin, center, and Dmitry Rogozin, far right, tour Russia’s new Vostochny Cosmodrome in October 2015.


The controversial leader of Russia’s space enterprises, Dmitry Rogozin, has continued a spree of firings that have seen many of the nation’s top spaceport officials fired, arrested, or both.

Most recently, on November 27, Russian media reported that Rogozin fired the leader of the Center for Exploitation of Ground-Based Space Infrastructure, which administers all of Russia’s spaceports. Andrei Okhlopkov, the leader of this Roscosmos subsidiary, had previously faced a reprimand from Rogozin for “repeated shortcomings in his work.” The spaceport organization has more than 12,000 employees.

Earlier this month, Rogozin also fired Vladimir Zhuk, chief engineer of the center that administers Russian spaceports. According to Russian media reports, Zhuk was then arrested for abusing his authority in signing off on water supply contracts.

Both of these officials were working to bring Russia’s newest spaceport, Vostochny, in the far eastern region of the country, up to its full capacity. In an article titled “At Vostochny A Day Never Goes By Without Someone Going to Jail,” The Kommersant newspaper reported that Zhuk knew that water supply networks for the Vostochny spaceport were not completed when he authorized their payment. (This article was translated for Ars by Rob Mitchell).

Construction project drags on

Several other key officials connected with the Vostochny Cosmodrome—under development since 2011 and intended to reduce Russia’s reliance on the Baikonur Cosmodrome in Kazakhstan—have also been recently let go. These include Vostochny head Evgeny Rogoz (fired and under house arrest), Vostochny Director Roman Bobkov (fired and arrested), and Defense Ministry Inspector General Dmitriy Fomintsev (arrested).

Construction of the spaceport has been riven with corruption, often through embezzlement, and overall cost estimates of the facility have increased to more than $7.5 billion. Of the planned seven launch pads, just one is operational. A Soyuz-2 rocket first launched from this “Site 1S” in April 2016. A second pad, “Site 1A,” may see the launch of an Angara rocket next year.

Russian President Vladimir Putin has been critical of delays at Vostochny, most recently in 2019, citing concerns about corruption. It is not clear whether the latest round of firings is related to a recent meeting Putin had with Rogozin to go over the country’s space affairs. It seems that by firing and arresting his subordinates, Rogozin has so far been able to shirk the blame for the Vostochny troubles onto other officials.

Nevertheless, his time may be coming. Rogozin is no stranger to corruption concerns, and Roscosmos is facing serious financial challenges. Not only is Russia no longer receiving large payments from NASA for Soyuz seats to carry its astronauts to the International Space Station, but funding from United Launch Alliance for the RD-180 rocket engine will also be ending within a few years. And there are serious questions about whether Russia’s next-generation Angara rocket will be able to compete with SpaceX’s Falcon 9 rocket for commercial launches.

Continue Reading