Slashdot is powered by your submissions, so send in your scoop

 



Forgot your password?
typodupeerror
×

Submission + - Google Engineers Advocate for Geoengineering (ieee.org)

the_newsbeagle writes: Two engineers from Google's renewable energy department wrote a feature article for IEEE Spectrum arguing that engineers can play a vital role in addressing climate change through both scaling up fairly mature technologies and seeking breakthroughs in nascent fields. They argue for carbon removal and sequestration to reverse warming trends, estimating that 2000 gigatonnes of CO2 must be removed from the atmosphere over the next century, and also for solar radiation management as a stopgap measure while humanity figures out how to do carbon removal on that scale.

Submission + - OpenAI's GPT-3 Speaks! (Kindly Disregard Toxic Language) (ieee.org)

the_newsbeagle writes: The massive language model GPT-3 from OpenAI has been hailed as a breakthrough in naturalistic text generation. It learned its 175 billion parameters by ingesting much of the English-language content on the Internet, and it can generate text with astounding fluency in a variety of styles and formats. But its power comes at a cost: While learning about language, it also learned all the biases, stereotypes and insults that can be found on the web. As a result, it sometimes sounds like the worst kind of Internet troll.

That might be of only passing concern if it was simply an academic project, but in fact OpenAI has created a private beta where hundreds of companies and developers are trying out GPT-3 to power a host of commercial products and services. GPT-3 is going to work in customer service, video games, tutoring services, mental health apps, and a variety of other applications.

OpenAI's position is that the only way to figure out GPT-3's weaknesses is to deploy it and see what happens. Says an OpenAI policy researcher: “We have to do this closed beta with a few people, otherwise we won’t even know what the model is capable of, and we won’t know which issues we need to make headway on,” she says. “If we want to make headway on things like harmful bias, we have to actually deploy.”

This long article in IEEE Spectrum tells the whole story.

Submission + - Can Computer Models Prescribe Public Health Interventions for COVID-19? (ieee.org)

the_newsbeagle writes: Since last winter and spring, the world has relied on computer models to predict the spread of the coronavirus (though many models have a spotty record of accuracy). Now, a new XPrize competition seeks models that can not just predict case numbers and mortality, but can also prescribe the best public health interventions for specific regions.

Submission + - Deep dive into COVID-19 vaccine manufacturing & distribution (ieee.org)

the_newsbeagle writes: Making safe and effective COVID-19 vaccines was just the first step. Now we need to manufacture and distribute billions of doses. This IEEE Spectrum article is bullish about humanity's ability to pull it off. The reporter talked to Corning about glass vials, to the president of the compressed gas association about dry ice for shipping, and to the head of Operation Warp Speed about the different manufacturing challenges of the four types of vaccine (RNA, viral vector, inactivated virus, and protein).

Submission + - What UV Light Can Really Do Against the Coronavirus (ieee.org)

the_newsbeagle writes: Remember back in April when Trump was promoting ultraviolet light as a cure for COVID-19? Here's a quote: "Supposing you brought the light inside the body... the whole concept of the light, the way it kills it in one minute — that's pretty powerful." Since then, often-dangerous misinformation has circulated about whether people can use UV to protect against the coronavirus. To combat the bad info, this article from IEEE Spectrum explains what UV light can actually do. It starts by clarifying what kind of UV light we're talking about:

Ultraviolet light lies in a region of the electromagnetic spectrum beyond indigo and violet. Anyone who’s read the label on a bottle of sunscreen knows the UV wavelengths that give you a suntan or a sunburn are called UV-A (with wavelengths between 400 and 315 nanometers) and UV-B (315 to 280 nm). Germicidal UV tech focuses on shorter, more energetic UV wavelengths, known as UV-C, which lie between 280 and 100 nm. The Earth’s ozone layer prevents virtually all UV-C light from reaching us. So microbes and viruses (and everything else, really) evolved for millions and billions of years without ever being exposed to these wavelengths.

Since UV-C can cause skin caner and cataracts in humans, the mercury-vapor lamps and LEDs that emit the light have to be used carefully. They've been used to sterilize the air near the ceilings of rooms, to blast operating rooms and airplanes when there are no people present, and to clean personal protective equipment. But researchers are now trying out a more versatile type of UV light.

Given the harmful effects of 254-nm UV-C, scientists are exploring the higher-energy wavelength of 222 nm, in the far-UV region. This wavelength has been found to kill viruses and bacteria, and initial studies show that it’s substantially safer than photons in the 254-nm range. In fact, far-UV may be able to safely bathe an entire room in sterilizing light, even with people present.


Submission + - Photo Essay: Robots in the COVID-19 Response (ieee.org)

the_newsbeagle writes: To fight a disease that thrives on human contact, robots have increasingly taken the place of vulnerable humans. Sentry robots have performed screenings and patrolled streets, looking for lockdown violators. Avatars have allowed family members to visit loved ones in senior homes and enabled graduating students to walk across the stage. In hospitals, germ zappers have blasted UV-C light through hospital rooms, while doctor assistant bots have checked on patients. This photo essay takes a tour of essential robot workers during the time of COVID.

Submission + - Why Modeling the Spread of COVID-19 Is So Damn Hard (ieee.org)

the_newsbeagle writes: At the beginning of the pandemic, modelers pulled out everything they had to predict the spread of the virus. This article explains the three main types of models used: 1) compartmental models that sort people into categories of exposure and recovery, 2) data-driven models that often use neural networks to make predictions, and 3) agent-based models that are something like a Sim Pandemic. Sometimes they got things very wrong, but they say they've learned valuable lessons from the experience.

Submission + - Blogger Behind "AI Weirdness" Says Today's AI Is Dumb and Dangerous (ieee.org) 1

the_newsbeagle writes: Janelle Shane has trained neural nets to generate names for new paint colors (such as Ronching Blue and Turdly), to come up with ideas for Halloween costumes (sexy barnacle sounds like a winner), and has generally tried to show people the absurdity and the serious limitations of today's artificial intelligence/machine learning systems. Her humor blog, AI Weirdness, led to her new book, You Look Like a Thing and I Love You, whose title is taken from a neural-net generated pickup line.

In a Q&A with IEEE Spectrum, she explains that you can learn a lot about an AI system by giving it a task and "watching it flail," and describes the perils see foresees if people overestimate the abilities of AI systems. She also talks about giraffing.

Submission + - Microsoft's Chief Environmental Officer Wants to Optimize the Earth (ieee.org)

the_newsbeagle writes: Lucas Joppa runs Microsoft's AI for Earth program, a $5-million and 5-year effort that seeks AI-powered solutions to global problems such as climate change and biodiversity loss. He spoke with IEEE Spectrum about the "ultimate optimization problem" of determining how best to use every square meter of the Earth's surface — and his long term goal of creating an AI-powered optimization engine that can help humans with those decisions.

Submission + - NASA's Lunar Space Station Might Be a Boondoggle (ieee.org)

the_newsbeagle writes: NASA is under pressure to put humans back on the moon by 2024 — that's the target date that VP Mike Pence boldly announced in a speech earlier this year. NASA's plan for meeting that ambitious target relies on building a space station in lunar orbit, called the Gateway. NASA says it will use its (over budget and behind schedule) SLS rocket and Orion crew capsule to dock at this (yet unbuilt) Gateway, then send down a lunar lander. Critics say this is a stupid and over-complicated plan.

This article by veteran space reporter Jeff Foust explains how NASA got itself into this situation.

Submission + - Rwanda Is Way Ahead on Delivery Drones (ieee.org)

the_newsbeagle writes: While the Google spinoff Wing is just launching its first commercial service in Australia (scones and coffee are on the menu) and getting FAA clearance to operate in the United States, a company called Zipline is already offering a nationwide service in Rwanda. Zipline currently uses its fixed-wing drones to deliver blood products to hospitals across the country, and will soon begin delivering other medical supplies as well.

Medical supplies seem like they could be a killer app for delivery drones, since they're often lightweight and urgently needed. But Zipline hasn't yet proven that its business model is sustainable. So two technology reporters from IEEE Spectrum traveled to Rwanda to embed themselves in a Zipline operations center, and produced an in-depth report and a 360 video about the company's logistics and technology.

Submission + - When Charles Babbage played chess with the original Mechanical Turk (ieee.org)

the_newsbeagle writes: The 19th century British engineer Charles Babbage is sometimes called the father of the computer. But his first design for a massive computing machine, a contraption called the Difference Engine that had some 25,000 parts, was just a giant calculator intended to handle logarithmic tables. It wasn’t until he began designing his first Analytical Engine that he began to dream of a smart machine that could handle more general-purpose computations.

This short essay argues that Babbage’s creative leap was inspired by an early example of AI hype: A supposed chess-playing machine called The Turk that had astounded onlookers throughout the courts of Europe. Babbage played two games against the Turk, and lost both.

Submission + - First Medical Device to Treat Alzheimer's Is Up for Approval by the FDA (ieee.org)

the_newsbeagle writes: An FDA advisory committee met today to consider approving the NeuroAD device, which is supposed to help with the cognitive symptoms of Alzheimer's disease. The device uses a combination of brain stimulation and cognitive training tasks to strengthen the neural circuits involved in language, memory, and other components of cognition. The treatment requires patients to come to the clinic daily for 1-hour sessions. Regulators in Israel and Europe have already approved the device.

The CEO of the company behind the device, Neuronix, says that they're not attempting to cure the underlying biological causes of Alzheimer's. "We’re attempting to modify the course of the disease,” he says. The cognitive improvements last for up to a year, after which they fade away.

Submission + - Wireless Sensors for NICU Babies Enable Cuddling (ieee.org)

the_newsbeagle writes: In newborn intensive care units (NICUs) today, tiny fragile babies lie in incubators, wired to a variety of monitors that track their vital signs. This mess of wires makes it complicated for nurses to pick up the babies for routine tasks like diaper changes, and makes it hard for new parents to pick up their infants for cuddling. Skin-to-skin contact between parents and infants has been proven not only to help with bonding, but also to have a host of medical benefits for the infants, so the wires that tether babies to their beds are a real problem.

At Northwestern University, an electrical engineer who works on flexible, stretchable electronics teamed up with a pediatric dermatologist to invent a solution. They devised a system of stick-on wireless biosensors (with a gentle adhesive that's safe even for thin preemie skin) that actually provide more information than today's standard setup.

Submission + - Pictionary-Playing AI Collaborates With Human Player (ieee.org)

the_newsbeagle writes: You know that feeling when you're playing Pictionary and you can't guess the answer, and your partner keeps drawing the same thing over and over? If you play the game with the new AI from the Allen Institute for Artificial Intelligence, you may encounter that same frustrating scenario.

The AI plays a simplified version of Pictionary (called Iconary), and can play as either the artist or the guesser. It's abilities in the game are limited, but Allen Institute researchers say it will get better as it plays with more humans.

The researchers' goal isn't to create a Pictionary champion, though. Ali Farhadi says the goal is to create an AI that's good at collaborating with humans: “To play Pictionary, the AI has to do some common sense reasoning, it has to know about abstraction, and it needs a little bit of theory of mind... By learning to play Pictionary, the AI can acquire skills and knowledge that transfer to real world applications.”

Slashdot Top Deals

Trying to be happy is like trying to build a machine for which the only specification is that it should run noiselessly.

Working...