Georgia Institute of Technology, College of Computing
from
The studies were originally intended to test for gender bias, that is, if people thought a robot believed to be female may be less competent at some jobs than a robot believed to be male and vice versa. But researchers at the Georgia Institute of Technology discovered no significant sexism against the machines.
“This did surprise us. There was only a very slight difference in a couple of jobs but not significant. There was, for example, a small preference for a male robot over a female robot as a package deliverer,” said Ayanna Howard, the principal investigator on both studies. Howard is a professor in and the chair of Georgia Tech’s School of Interactive Computing.
Although robots are not sentient, as people increasingly interface with them, we begin to humanize the machines. Howard studies what goes right as we integrate robots into society and what goes wrong, and much of both has to do with how the humans feel around robots.
Remaking the University blog, Christopher Newfield
from
Should university officials be fatalistic about Covid-powered cuts to their core educational budgets? Or should they work 24/7 on their state governments to keep their current budgets whole?
What about state governments? Should they cut higher ed yet again, as various governors are doing (New Jersey, Ohio), and as Gavin Newsom proposes in California?
This post investigates the budget case for a zero-cuts policy. If your state’s public colleges and universities have an ample base budget, you can make some temporary cuts to their state funding. If they are already bare bones, further budget cuts will cut educational quality.
Tech giant Amazon has announced five new utility-scale solar projects in China, Australia and the US to support the company’s commitment to reach 80% renewable energy by 2024, 100% renewable energy by 2030, and decarbonised operations by 2040.
Alexei Efros, a professor in the Electrical Engineering and Computer Sciences department of the University of California, Berkeley, cites a problem with the current approach to handling images: they generally contain far more information than is implied by the relatively simple tags applied by humans for training purposes. “The problem I see now is that supervising with high-level concepts like ‘door’ or ‘airplane’ before the computer even knows what an object is simply invites disaster.”
What Efros wants is for the AI systems to capture the information that humans remember after they have seen a picture. “Do we have a photographic memory? No, we don’t. But we have a cool embedding that captures a lot of non-linguistic information,” Efros says. “We need to get away from semantics and force the computer to represent more of what is actually in the image.”
The artificial intelligence model showed great promise in predicting which patients treated in U.S. Veterans Affairs hospitals would experience a sudden decline in kidney function. But it also came with a crucial caveat: Women represented only about 6% of the patients whose data were used to train the algorithm, and it performed worse when tested on women.
The shortcomings of that high-profile algorithm, built by the Google sister company DeepMind, highlight a problem that machine learning researchers working in medicine are increasingly worried about. And it’s an issue that may be more pervasive — and more insidious — than experts previously realized, new research suggests.
The study, led by researchers in Argentina and published Monday in the journal PNAS, found that when female patients were excluded from or significantly underrepresented in the training data used to develop a machine learning model, the algorithm performed worse in diagnosing them when tested across across a wide range of medical conditions affecting the chest area. The same pattern was seen when men were left out or underrepresented.
Financial media company Bloomberg’s involvement with the OpenAPI Initiative and the open source community is built on aspirations wider than just about choosing the right tools to grow a business. The stakes are especially high in these turbulent times as the ravages of COVID-19 continue to take its toll, already wiping out large swaths of the economy. [audio, 37:17]
In this latest episode of The New Stack Makers podcast, we speak with two open source leaders from Bloomberg:
Richard Norton, Bloomberg’s head of the data license engineering group.
Kevin Fleming, head of open source community engagement and member of Bloomberg’s CTO office.
The National Science Foundation, National Institutes of Health and other funding sources have recognized the importance of COVID-19-related social science research with millions of dollars in grants already. The NIH, for example, is funding emergency research on topics including the effects and mechanisms of social connectedness, connection and isolation across the life span. Among other work, the NSF’s Division of Social and Economic Sciences is supporting a study on risk perception, media use, stress and protective behaviors within the U.S. population, via a $200,000 Rapid Response Research grant.
Social science journal editors are seeing a related bump in submissions. The American Psychological Association said that submissions for its publications were up 15 times more than the expected growth rate in April, though not all those papers were related to the coronavirus. The association also published a COVID-19-related stress survey, which found that stress levels right now are significantly higher for parents than nonparents (go figure).
Federal and state officials across the country have altered or hidden public health data crucial to tracking the coronavirus’ spread, hindering the ability to detect a surge of infections as President Donald Trump pushes the nation to reopen rapidly.
In at least a dozen states, health departments have inflated testing numbers or deflated death tallies by changing criteria for who counts as a coronavirus victim and what counts as a coronavirus test, according to reporting from POLITICO, other news outlets and the states’ own admissions. Some states have shifted the metrics for a “safe” reopening; Arizona sought to clamp down on bad news at one point by simply shuttering its pandemic modeling. About a third of the states aren’t even reporting hospital admission data — a big red flag for the resurgence of the virus.
“We’re looking for the right balance between cutting social connections and maintaining social connections,” said Sayama. “By integrating information from the individual, local, regional and state levels, we’re looking to find ways to maximize the benefits of both social contact and social isolation together.”
The research team is incorporating data from multiple sources, including from states and regions that have begun their reopening processes.
“Everything is data right now,” he said. “We aren’t just looking at the immediate short-term; we’re looking at this upcoming fall and next year. Humanity needs to cope with this terrible thing for quite a substantial length of time, so that means we need to develop some more sustainable strategies on how to cope.”
Artificial intelligence (AI) just seems to get smarter and smarter. Each iPhone learns your face, voice, and habits better than the last, and the threats AI poses to privacy and jobs continue to grow. The surge reflects faster chips, more data, and better algorithms. But some of the improvement comes from tweaks rather than the core innovations their inventors claim—and some of the gains may not exist at all, says Davis Blalock, a computer science graduate student at the Massachusetts Institute of Technology (MIT). Blalock and his colleagues compared dozens of approaches to improving neural networks—software architectures that loosely mimic the brain. “Fifty papers in,” he says, “it became clear that it wasn’t obvious what the state of the art even was.”
The researchers evaluated 81 pruning algorithms, programs that make neural networks more efficient by trimming unneeded connections. All claimed superiority in slightly different ways. But they were rarely compared properly—and when the researchers tried to evaluate them side by side, there was no clear evidence of performance improvements over a 10-year period. The result, presented in March at the Machine Learning and Systems conference, surprised Blalock’s Ph.D. adviser, MIT computer scientist John Guttag, who says the uneven comparisons themselves may explain the stagnation. “It’s the old saw, right?” Guttag said. “If you can’t measure something, it’s hard to make it better.”
The majority of Americans (87 percent) are comfortable sharing personal and lifestyle-focused data for the benefit of lower insurance premiums, according to a new insurance study conducted by DXC Technology
As part of the study, detailed in the 2020 DXC Insurance Survey Report: The Voice of the U.S. Consumer, more than 2,000 U.S. consumers were surveyed on their views about insurance and how they interact with those providing coverage. The study highlights how shifting consumer expectations are presenting opportunities for insurers to adopt and strategically deploy digital technologies — not only to gain market share and boost lifetime customer value, but also to assert their brands as exemplars of a new, more collaborative concept of the insurer/policyholder relationship.
National security journalist Barton Gellman talks about “the surveillance-industrial state,” the possibility of a Biden presidency or a second Trump term, and his gripping new book.
Consider dude and its many formulations. “That can convey basically anything, like ‘Duuuuude, that’s awful,’” says University of Vermont applied mathematician Peter Sheridan Dodds, one of the study’s coauthors. On the other hand, “Dude!” is different. “It could be excitement; it could be joy,” says Dodds. … To quantify this, Dodds, Danforth, and the lead author of the paper, University of Vermont computational linguist Tyler Gray, randomly selected 10 percent of all tweets sent out between 2008 and 2016, around 100 billion in all. (They have an arrangement with Twitter to obtain this data.) Gray wrote a program that searched the data for stretched words, specifically looking for repeated letters.
Plans for Stanford’s new Innovative Medicines Accelerator arose before the COVID-19 pandemic struck, but now its programs are focused entirely on helping faculty generate and test new medicines to slow the spread of the disease.
Online June 10-11. “The workshop is collaboration of ETH Zurich and NYU Courant. It is also a part of Data Science in Techno-Socio-Economic Systems ETH course and supported by EU SoBigData++ project.” [registration required]
We frequently find ourselves giving the same advice to different students on how to write rebuttals. So we thought we’d write it up. Our experience is with AI conferences (e.g., CVPR, ECCV, ICCV, NeurIPS, ICLR, EMNLP).The core guiding principle is that the rebuttal should be thorough, direct, and easy for the Reviewers and Area Chair (RACs) to follow.
Google is introducing a new “quick settings” menu in Gmail aimed at helping users browse, discover and use different themes and settings to customize their Gmail experience. These options include the ability to change the density of text, select from different inbox types and add reading panes and options to theme your inbox. They are not new features, but before had been buried in Gmail’s settings. Many users may have not even known the options existed, unless they went digging.
On April 8, 2020, the Federal Trade Commission (FTC) published a blog post titled, “Using Artificial Intelligence and Algorithms,” that offers important lessons about the use of AI and algorithms in automated decision-making. The post begins by noting that headlines today tout rapid improvements in AI technology, and the use of more advanced AI has enormous potential to improve welfare and productivity. But more sophisticated AI also presents risks, such as the potential for unfair or discriminatory outcomes. This tension between benefits and risks is a particular concern in “Health AI,” and the tension will continue as AI technologies are deployed to tackle the current COVID-19 crisis.
The FTC post reminds companies that, while the sophistication of AI is new, automated decision-making is not, and the FTC has a long history of dealing with the challenges presented by the use of data and algorithms to make decisions about consumers.
The Raspberry Pi Foundation is always looking at ways to make its models bigger and better, and today it’s realized its ambition of an 8GB Raspberry Pi 4. It joins the 2GB and 4GB line up, and will set you back $75.
It’s essentially the same deck-of-cards-sized single board computer as its predecessors, just with more RAM, which gives DIY projects more scope and means tinkerers can start exploring more memory-hungry applications, such as streaming. A couple of components have moved on the board to help supply the slightly higher peak currents needed by the new memory package, but everything else essentially remains the same: lots of ports, ARM-based CPU, WiFi, Bluetooth…