The FDA has narrowly backed Merck’s covid pill—but it’s not that effective

The news: A US Food and Drug Administration panel has voted by 13 to 10 to recommend that the government authorize Merck’s antiviral pill for patients with early covid-19 who are at high risk for severe infection. The drug, called molnupiravir, has been shown to reduce the risk of hospitalization and death, although by less than previously thought. Initial results in October found it cut the risk of hospitalization or death by 50% when given to 755 unvaccinated volunteers who were mildly to moderately ill with covid-19 and had at least one risk factor for severe illness. Last week, with updated data from 1,433 patients, that figure was cut to 30%. “The efficacy of this product is not overwhelmingly good,” said panel member David Hardy.

Lacking consensus: The closeness of the vote was due to concerns over the change in the drug’s efficacy data, but also over its safety. James Hildreth, CEO of Meharry Medical College and one of the panel members, said he voted no because he worried that the use of molnupiravir could, theoretically, lead to new covid-19 variants. Other panelists, though, argued that the overall risk was small enough to vote it through. 

Who can take it: If it’s authorized, the drug will be prescribed to high-risk people who have begun experiencing symptoms to take twice a day at home for five days. Tens of millions of Americans who are older or have underlying medical conditions would qualify, and they’d need to begin taking it within five days of symptoms appearing. The committee recommended tight restrictions on molnupiravir’s use in pregnant women, given concerns about the potential side effects. 

What’s next: In the coming weeks, the FDA will assess a similar pill from Pfizer that seems to be significantly more effective than Merck’s, reducing the risk of hospitalization or death within the same patient groups by 89%. The hope is that these drugs could provide promising new weapons in our arsenal against covid, especially because they are easily stored and can be taken at home. The US government has already spent billions of dollars to secure large supplies of both new pills. 

The FDA has narrowly backed Merck’s covid pill—but it’s not that effective 2021/12/01 13:25

We still don’t know enough about the omicron variant to panic

The news: Just five days ago, South African scientists informed the World Health Organization that they’d identified a new covid-19 variant. The situation has escalated rapidly since then. The variant, known as B.1.1.529, has already been found in many countries across the world. On Friday it was designated a variant of “concern” by the WHO, which opted to name it “omicron,” the 15th letter of the Greek alphabet, following the organization’s naming system.

Governments are reimposing border restrictions and closures, as well as new measures to mitigate covid’s spread among their populations. Health ministers from G7 countries are set to meet today to discuss their response. 

What we know: Viruses mutate all the time, and that isn’t cause for alarm on its own. Part of the reason why the omicron variant is worrying people is that it has so many mutations in its spike protein—approximately 30, which is roughly double the number delta has. This protein is the part of the virus that helps it to enter human cells. Preliminary evidence suggests this variant brings a higher risk of reinfection, according to the WHO.

The omicron variant has been identified in at least 15 countries already, mostly in southern Africa but also in the UK, Europe, Hong Kong, Canada, Israel, and Australia.

What we don’t know: Amid all the panic, it’s important to remember that we still know very little about the new variant—and we’ve been worried before about variants that have come to nothing. The crucial questions are whether it increases transmissibility, whether it worsens health outcomes (thus pushing up deaths and hospitalizations), and, crucially, whether it erodes immunity afforded by vaccines or previous infections. We don’t have firm answers to any of these questions yet—although it seems likely, given the mutations, that it will affect the effectiveness of vaccines to some degree.

If that’s the case, then vaccine manufacturers will have to move quickly to come up with new versions. Luckily, with mRNA technology it is relatively easy to reformulate a vaccine. Moderna’s chief medical officer, Paul Burton, told the BBC on Sunday that his firm could have a new booster—one tweaked to handle omicron—ready to roll out as soon as early next year.

Researchers around the world are now racing to gather the data we need to know how worried we should be. We also don’t know exactly how omicron arose. Experts have long warned that uneven global vaccine access—South Africa, where omicron seems to have originated, has a vaccination rate of 35%—poses a global risk because it gives the virus more opportunities to mutate.

What you can do: As has been the case throughout the pandemic, the best thing you and your loved ones can do to protect yourselves is to get vaccinated. If you are offered a booster shot, take it. While it’s possible that omicron might degrade vaccine efficacy, it won’t eradicate it altogether.

We still don’t know enough about the omicron variant to panic 2021/11/29 12:24

Can Afghanistan’s underground “sneakernet” survive the Taliban?

When Afghanistan fell to the Taliban in August, Mohammad Yasin had to make some difficult decisions very quickly. As the country reeled from the shock of the insurgent takeover, the 21-year-old—whose name has been changed to protect his safety—snuck into his small place of business and got to work. 

He began erasing some of the sensitive data on his computer and moving the rest onto two of his largest hard drives, which he then wrapped in a layer of plastic and buried underground at an undisclosed location.

Yasin didn’t take these precautions because he is part of Afghan intelligence, or linked to the government. He has no state secrets hidden on his computers. He is what is locally referred to as a “computer kar”: someone who sells digital content by hand in a country where a steady internet connection can be hard to come by. “I sell pretty much everything, from movies, music, mobile applications, to iOS updates. I also help create Apple IDs and social media accounts, and with backing up phones and recovering data,” he says, then adds, in a hushed voice, “I can also unlock [stolen] phones and provide other naughty videos.” 

When the Taliban captured the city of Herat on August 12, Yasin and his colleagues speculated that it wouldn’t be long before the Taliban’s invading forces took over their own city of Mazar-i-Sharif. 

“Things were more tense in Mazar, too, so me and other computer kars of Mazar who work together held a secret meeting to decide what to do to protect all our content,” he says. Among them, the informal union of computer kars had several hundred terabytes of data collected over several years, and much of it would be considered controversial—even criminal—by the Taliban. 

“We all agreed to not delete, but rather hide the more nefarious content,” he says. “We reasoned that in Afghanistan, these regimes come and go frequently, but our business should not be disrupted.” 

He isn’t too worried about being discovered.

“People are hiding guns, money, jewelry, and whatnot, so I am not scared of hiding my hard drives. They will never be able to find [them],” he says. “I am a 21st-century boy, and most Taliban are living in the past.”

Less than 20 years after former president Hamid Karzai made Afghanistan’s first mobile phone call, there are nearly 23 million mobile phone users in a country of fewer than 39 million people. But internet access is a different matter: by early 2021, there were fewer than 9 million internet users, a lag that has been largely attributed to widespread physical security problems, high costs, and a lack of infrastructural development across the country’s mountainous terrain. 

That’s why computer kars like Yasin can now be found all across Afghanistan. Although they sometimes download their information from the internet when they’re able to get a connection, they physically transport much of it on hard drives from neighboring countries—what is known as the “sneakernet.”

“I use the Wi-Fi at home to download some of the music and applications; I also have five SIM cards for internet,” says Mohibullah, another kar who asked not to be identified by his real name. “But the connection here is not reliable, so every month I send a 4 terabyte hard drive to Jalalabad, and they fill it with content and return it in a week’s time with the latest Indian movies or Turkish TV dramas, music, and applications,” for which he says he pays between 800 and 1,000 afghanis ($8.75 to $11).

“People are hiding guns, money, jewelry, and whatnot, so I am not scared of hiding my hard drives. I am a 21st-century boy, and most Taliban are living in the past.”

Mohammad Yasin, computer kar

Mohibullah says he can install more than 5 gigabytes of data on a phone—including movies, songs, music videos, and even course lessons—for just 100 afghanis, or $1.09. “I have the latest Hollywood and Bollywood movies dubbed in Dari and Pashto [Afghan national languages], music from across the globe, games, applications,” he told me in early August, days before the Taliban took over. 

For just a little more, Mohibullah helps customers create social media accounts, sets up their phones and laptops, and even writes emails for them. “I sell everything—A to Z of contents. Everything except ‘100% films,’” he said, referring to pornography. (Later he admitted that he did have some “free videos,” another nickname for porn, but that he only sells them to trusted customers.)

Most of his customers are men, but women also regularly buy music and movies from him. Much of it comes from Pakistan, which he says has better and cheaper internet connectivity.

As we were discussing the business in Mohibullah’s small store on a crowded street in west Kabul, two women walked in. They declined an interview request, but told us they were “wedding DJs” looking for latest music to play at their clients’ lavish wedding parties. Mohibullah offered them a selection of latest Indian music to browse through, and he transferred each of them a playlist of over 100 songs for 70 afghanis.

Unfortunately for the kars, such clients have entirely disappeared since the rise of the Taliban. The violent, extremist regime has banned music and restricted women’s freedoms.

Yasin and Mohibullah have had to adapt their business quickly to the new regime. They replaced the raunchy Bollywood and Iranian music videos with the Taliban taranas (songs without music) and recitations from the Quran. Afghans love to carry pictures of celebrities on their phones; those have now been replaced with pictures of Taliban flags in different styles. And all the “free movies” kars offer are now hidden; only they know where. 

“If they ever find those, I will be punished very badly. They will execute me,” says Yasin, shuddering.

Content crackdowns

The Taliban takeover has been bad for business, they both admit. Their average earnings have fallen nearly 90%, from around 3,000 afghanis per day to less than 350—from $32 to $3.80. 

“From that, at least 100 afghanis goes for generator fuel and about 50 afghanis to the municipality for the space I use on the street,” says Yasin. “That isn’t enough to support my five siblings and [my] parents.”

In addition to policing their content, the Taliban have also been cracking down on kars like Yasin who have expanded their services to help Afghans fleeing persecution. 

“Those who are in hiding or who are waiting to be evacuated come to me to help them back up their phone data on flash drives, to avoid being caught by the Taliban fighters who are checking phones at the checkpoints,” he says. 

Sometimes he charges a nominal fee, he says, but he has also waived it in some cases. 

“It is usually personal data they want to take with them that the Taliban may not approve, and sometimes it’s information that can identify them as supporters of the previous government or foreign allies, that can get them arrested or even executed,” he says.

Mohibullah finds it ironic that the Taliban are cracking down on the content dealers now that they are in power, because they used the sneakernet themselves for radicalization and recruitment. 

“Every once in a while, some men would approach us to distribute the Taliban taranas praising their fighters, or graphic videos of the executions they’ve conducted,” he says. “They wanted to use our services to spread their ideology and propaganda among the youth.” 

He never shared such content with his clients before now, he says. 

“These days, however, the Taliban are among us, and they demand such content. They also ask for pictures of Taliban flags and fighters with their weapons. I oblige because I have to feed my family,” he says.

But the Afghan computer kars are nothing if not enterprising. Many of them continue to discreetly sell forbidden content. Others, searching for a silver lining, are hopeful that there may even be an uptick in business for certain entertainment content as many Afghans, particularly women, are forced to stay indoors. 

“During covid lockdowns there was an increase in demand for cartoon clips because children were locked at home,” says Mohibullah. “Now, with the Taliban and widespread unemployment, people are also stuck at home; they might watch more movies.”

Can Afghanistan’s underground “sneakernet” survive the Taliban? 2021/11/26 10:41

NASA wants to use the sun to power future deep space missions

In August 2022, a NASA probe called Psyche will set out to explore a giant metallic asteroid called Psyche 16, to help scientists learn more about how planets form. The way Psyche reaches its target, though, will be different from typical NASA missions.  

Building on technology used in previous missions, including Dawn and Deep Space 1, solar power will help propel Psyche into deep space. If that proves successful, it could be the start of a new era of using more fuel-efficient probes for both space exploration and commercial missions. 

Traditional spacecraft rely on chemical reactions between a combination of liquid fuels to get around rather than electricity. Psyche will use two giant solar arrays to convert solar energy into electricity that will power four ion thrusters. That electricity will turn tanks of xenon gas (the same kind used in car headlights) into xenon ions, which Psyche’s four thrusters will eject to gently propel the spacecraft toward the asteroid, which orbits between Mars and Jupiter, more than 1.5 billion miles from Earth.

While other spacecraft, like Lucy, have used solar energy to operate instruments, Psyche will be among the first of NASA’s deep-space missions to use solar energy for both onboard operations and propulsion. 

Paulo Lozano, director of MIT’s space propulsion laboratory, says Psyche could lay the groundwork for more solar-powered space exploration. Eventually, the technology could help us investigate multiple celestial objects for longer periods and potentially make human-crewed missions outside of Earth’s orbit more affordable and feasible.  

“It actually opens up the possibility to explore and to commercialize space in a way that we haven’t seen before,” Lozano says.  

Because a spacecraft that uses solar-electric propulsion requires less propellant than a chemically powered one, it has more space on board for cargo, scientific instruments, and, someday, astronauts. One company, Accion Systems, is developing more efficient ion thrusters for Cubesats as well as larger satellites and other spacecraft. 

Solar propulsion technology is already common in satellites that orbit Earth, but until now it has not been a powerful enough alternative to chemically powered engines to be used as often in spacecraft headed to deep space.  Advances in solar electric propulsion will change that.

The technology behind Psyche had its first major test in Dawn, an exploration spacecraft that used solar power and ion thrusters. Dawn eventually went silent while orbiting the dwarf planet Ceres (where it will remain in orbit for decades) in 2018, three years after the mission was supposed to end. These thrusters can operate for years without running out of fuel, but they provide relatively low thrust compared with conventional propulsion.  

Psyche’s thrusters will be able to generate three times as much thrust as its predecessors, and about a year after launch, it will get some help from Mars’s gravitational pull to change its trajectory before eventually reaching its target in 2026.  

After that, Psyche will spend just under two years orbiting the asteroid. Its mission will be to examine the asteroid’s iron core to determine whether it has the same elements that have been discovered in Earth’s high-pressure core, which can help researchers learn more about how planets form. 

Although we can’t directly view Earth’s core, Psyche will use a multispectral imager, an instrument that uses filters and two cameras to obtain high-resolution geologic, compositional, and topographical data from the asteroid. If its core proves similar to those of small, rocky planets, scientists could determine whether the two have similar origins. Because Psyche 16 (the asteroid) is thought to be the core of a planet that failed to form, an up-close look could provide details about the formation of the inner solar system. 

Back in 2017, Psyche was chosen as one of two missions for NASA’s Discovery Program, a series of low-cost missions to targets around the solar system. Led in part by Arizona State University, Psyche had about a $450 million development cap to ensure it could make it all the way to deep space. But the farther a spacecraft goes from the sun, the more difficult it becomes for its solar arrays to capture sunlight and power its ion thrusters. That’s why once Psyche is past Mars, it will have to slow down. 

To go even further, spacecraft may need to rely on nuclear thermal propulsion, which NASA is also developing. 

NASA wants to use the sun to power future deep space missions 2021/11/25 13:00

Machine learning improves Arabic speech transcription capabilities

Thanks to advancements in speech and natural language processing, there is hope that one day you may be able to ask your virtual assistant what the best salad ingredients are. Currently, it is possible to ask your home gadget to play music, or open on voice command, which is a feature already found in some many devices.

If you speak Moroccan, Algerian, Egyptian, Sudanese, or any of the other dialects of the Arabic language, which are immensely varied from region to region, where some of them are mutually unintelligible, it is a different story. If your native tongue is Arabic, Finnish, Mongolian, Navajo, or any other language with high level of morphological complexity, you may feel left out.

These complex constructs intrigued Ahmed Ali to find a solution. He is a principal engineer at the Arabic Language Technologies group at the Qatar Computing Research Institute (QCRI)—a part of Qatar Foundation’s Hamad Bin Khalifa University and founder of ArabicSpeech, a “community that exists for the benefit of Arabic speech science and speech technologies.”

Qatar Foundation Headquarters

Ali became captivated by the idea of talking to cars, appliances, and gadgets many years ago while at IBM. “Can we build a machine capable of understanding different dialects—an Egyptian pediatrician to automate a prescription, a Syrian teacher to help children getting the core parts from their lesson, or a Moroccan chef describing the best couscous recipe?” he states. However, the algorithms that power those machines cannot sift through the approximately 30 varieties of Arabic, let alone make sense of them. Today, most speech recognition tools function only in English and a handful of other languages.

The coronavirus pandemic has further fueled an already intensifying reliance on voice technologies, where the way natural language processing technologies have helped people comply with stay-at-home guidelines and physical distancing measures. However, while we have been using voice commands to aid in e-commerce purchases and manage our households, the future holds yet more applications.

Millions of people worldwide use massive open online courses (MOOC) for  its open access and unlimited participation. Speech recognition is one of the main features in MOOC, where students can search within specific areas in the spoken contents of the courses and enable translations via subtitles. Speech technology enables digitizing lectures to display spoken words as text in university classrooms.

Ahmed Ali, Hamad Bin Kahlifa University

According to a recent article in Speech Technology magazine, the voice and speech recognition market is forecast to reach $26.8 billion by 2025, as millions of consumers and companies around the globe come to rely on voice bots not only to interact with their appliances or cars but also to improve customer service, drive health-care innovations, and improve accessibility and inclusivity for those with hearing, speech, or motor impediments.

In a 2019 survey, Capgemini forecast that by 2022, more than two out of three consumers would opt for voice assistants rather than visits to stores or bank branches; a share that could justifiably spike, given the home-based, physically distanced life and commerce that the epidemic has forced upon the world for more than a year and a half.

Nonetheless, these devices fail to deliver to vast swaths of the globe. For those 30 types of Arabic and millions of people, that is a substantially missed opportunity.

Arabic for machines

English- or French-speaking voice bots are far from perfect. Yet, teaching machines to understand Arabic is particularly tricky for several reasons. These are three commonly recognised challenges:

  1. Lack of diacritics. Arabic dialects are vernacular, as in primarily spoken. Most of the available text is nondiacritized, meaning it lacks accents such as the such as the acute (´) or grave (`) that indicate the sound values of letters. Therefore, it is difficult to determine where the vowels go.
  2. Lack of resources. There is a dearth of labeled data for the different Arabic dialects. Collectively, they lack standardized orthographic rules that dictate how to write a language, including norms or spelling, hyphenation, word breaks, and emphasis. These resources are crucial to train computer models, and the fact that there are too few of them has hobbled the development of Arabic speech recognition.
  3. Morphological complexity. Arabic speakers engage in a lot of code switching. For example, in areas colonized by the French—North Africa, Morocco, Algeria, and Tunisia—the dialects include many borrowed French words. Consequently, there is a high number of what are called out-of-vocabulary words, which speech recognition technologies cannot fathom because these words are not Arabic.

“But the field is moving at lightning speed,” Ali says. It is a collaborative effort between many researchers to make it move even faster. Ali’s Arabic Language Technology lab is leading the ArabicSpeech project to bring together Arabic translations with the dialects that are native to each region. For example, Arabic dialects can be divided into four regional dialects: North African, Egyptian, Gulf, and Levantine. However, given that dialects do not comply with boundaries, this can go as fine-grained as one dialect per city; for example, an Egyptian native speaker can differentiate between one’s Alexandrian dialect from their fellow citizen from Aswan (a 1,000 kilometer distance on the map).

Building a tech-savvy future for all

At this point, machines are about as accurate as human transcribers, thanks in great part to advances in deep neural networks, a subfield of machine learning in artificial intelligence that relies on algorithms inspired by how the human brain works, biologically and functionally. However, until recently, speech recognition has been a bit hacked together. The technology has a history of relying on different modules for acoustic modeling, building pronunciation lexicons, and language modeling; all modules that need to be trained separately. More recently, researchers have been training models that convert acoustic features directly to text transcriptions, potentially optimizing all parts for the end task.

Even with these advancements, Ali still cannot give a voice command to most devices in his native Arabic. “It’s 2021, and I still cannot speak to many machines in my dialect,” he comments. “I mean, now I have a device that can understand my English, but machine recognition of multi-dialect Arabic speech hasn’t happened yet.”

Making this happen is the focus of Ali’s work, which has culminated in the first transformer for Arabic speech recognition and its dialects; one that has achieved hitherto unmatched performance. Dubbed QCRI Advanced Transcription System, the technology is currently being used by the broadcasters Al-Jazeera, DW, and BBC to transcribe online content.

There are a few reasons Ali and his team have been successful at building these speech engines right now. Primarily, he says, “There is a need to have resources across all of the dialects. We need to build up the resources to then be able to train the model.” Advances in computer processing means that computationally intensive machine learning now happens on a graphics processing unit, which can rapidly process and display complex graphics. As Ali says, “We have a great architecture, good modules, and we have data that represents reality.” 

Researchers from QCRI and Kanari AI recently built models that can achieve human parity in Arabic broadcast news. The system demonstrates the impact of subtitling Aljazeera daily reports. While English human error rate (HER) is about 5.6%, the research revealed that Arabic HER is significantly higher and can reach 10% owing to morphological complexity in the language and the lack of standard orthographic rules in dialectal Arabic. Thanks to the recent advances in deep learning and end-to-end architecture, the Arabic speech recognition engine manages to outperform native speakers in broadcast news.

While Modern Standard Arabic speech recognition seems to work well, researchers from QCRI and Kanari AI are engrossed in testing the boundaries of dialectal processing and achieving great results. Since no one speaks Modern Standard Arabic at home, attention to dialect is what we need to enable our voice assistants to understand us.

This content was written by Qatar Computing Research Institute, Hamad Bin Khalifa University, a member of Qatar Foundation. It was not written by MIT Technology Review’s editorial staff.

Machine learning improves Arabic speech transcription capabilities 2021/11/24 15:20

    2 / 2