The House approved seven of eight telecom and privacy-related amendments to the “minibus” FY 2020 budget bill (HR-3055) that includes funding for NTIA, other Commerce Department agencies and the Agriculture Department (see 1906190061). Lawmakers voted 245-186 against language from Rep. John Rutherford, R-Fla., that would have decreased NTIA's FY 2020 salaries and expenses funding to FY 2019 levels as a way to increase funding for NOAA by $3.5 million. The chamber voted 408-22 for an amendment led by Rep. Abigail Spanberger, D-Va., that would increase funding for the Rural Utilities Service's $600 million ReConnect rural broadband funding program (see 1812130064) by $55 million, which would be paid for by decreasing funding for USDA administration, the department's chief information officer and the department's general counsel. Lawmakers voted 425-6 for a proposal led by Rep. Greg Pence, R-Ind., to increase funding for RUS' distance learning and telemedicine grant program by $25 million by cutting an equal amount from the CIO's office. The House cleared the other five amendments on voice votes, including two that target broadband mapping. One led by Rep. Antonio Delgado, D-N.Y., would bar NTIA from using its funding to update broadband coverage maps using data collected only via Form 477. The other, from Rep. Xochitl Torres Small, D-N.M., designates $1 million of NTIA’s funding be used for broadband mapping. Language led by Rep. David Trone, D-Md., would increase funding for community connect grants by $5 million for broadband deployments in underserved rural areas. An amendment from Rep. Angie Craig, D-Minn., would allocate $1 million of the distance learning program funding to “express the importance of broadband access to rural communities, schools and small businesses.” The other cleared proposal, from Rep. Don Beyer, D-Va., would direct $1 million of National Science Foundation funding to report to Congress on “its efforts to incorporate social impact assessments into the artificial intelligence research it funds.”
The National Institute of Standards and Technology shouldn’t rush developing artificial intelligence standards and should rely on existing international efforts, stakeholders commented through Monday (see 1905300048). A “rush to impose standards could hamper innovation or lead to standards that quickly become irrelevant as technology advances,” AT&T said. Microsoft said it’s “premature” to develop “sector‐specific vertical standards at this time,” given AI’s continued development. Shift focus to promoting development of “open frameworks, shared definitions, and related tools -- including evaluations, data sets, and metrics,” IBM recommended. “Premature standardization is even more important to avoid given the rapid rate of innovation,” the Information Technology Industry Council said. Rather than creating new standards, “look to existing data standards for acquisition, storage, access and use,” ITI said. The association emphasized existing international standards established by organizations like the International Organization of Standardization/IEC Joint Technical Committee. Microsoft also urged NIST to retain international principles like those from the Organisation for Economic Co‐operation and Development. The agency should address analysis gaps with government, academia and industry, but it needs to avoid “becoming a standards‐setting organization,” the company said. Like Microsoft and ITI, BSA|The Software Alliance backed “robust” U.S. participation in the development of international standards. Global standards “have the added benefit of mitigating the risks that can accompany country-specific standards,” BSA said. Strive for federal standards in the handling and securing AI data, the Center for Democracy & Technology said, and emphasize transparency for AI development. Establish a “uniform vocabulary for describing structures, elements, parameters, hyperparameters, and techniques for developing” machine learning systems, CDT said.
Amazon tapped into artificial intelligence again, announcing Wednesday the StyleSnap feature to help shoppers find styles and products that match their taste. Consumers click the camera icon in the upper right-hand corner of the Amazon app, select Style Snap, upload a photograph or screenshot of a fashion look they like, and the feature will present recommendations for similar items on Amazon.com, blogged Amazon spokesperson Arun Krishnan. StyleSnap uses computer vision and deep learning to identify apparel items in a photo. Deep learning is a class of machine learning techniques based on artificial neural networks, “inspired by the working of the human brain,” comprising millions of artificial neurons connected to each other, which can be “trained” to detect images of outfits by seeing a series of images, Krishnan said. Deep learning can eventually be able to tell the difference between subtleties such as maxi and accordion skirts, Krishnan said, but learning is required: “If we … present it with one Scottish kilt, it may be confused and predict an incorrect class until enough examples are provided to train it otherwise.” To have neural networks identify a greater number of classes, they have to go through stacks of layers to learn concepts such as edges and colors, then identify patterns such as 'floral' or 'denim,' he said. After passing through all of the layers, the algorithm can accurately identify concepts like fit and outfit style in an image. “Feed-forward neural networks will stall and eventually degrade after a certain number of layers have been added,” he said, referencing “the vanishing gradient problem, where the signal from the training data is so spread out between layers that it is lost entirely.” Amazon uses residual networks to overcome the problem, using shortcuts to allow the training signal to skip over some of the layers in the network. This helps the network learn basic features like “edges” and “patterns” first, and then focus on complex concepts, he said. Amazon researchers developed a way for the network to learn new concepts while also remembering things it has learned in the past, critical for enabling StyleSnap to work through large volumes of data effectively, he said. While customers can discover fashion finds by taking screenshots of the looks they like, the technology can also help fashion influencers expand their communities, Amazon maintains; fashion influencers who participate in the Amazon Influencer Program are eligible to receive commissions for purchases they inspire, he said.
The National Institute of Standards and Technology extended to June 10 the comment period for a request for information (see 1905010151) on developing technical standards for artificial intelligence, the agency said Thursday.
About 29 percent of companies report “making regular use” of artificial intelligence, CompTIA reported Wednesday. That compares with 24 percent of 2017 respondents. The association polled 500 U.S. business and tech professionals in March and April. The lack of AI use could be linked to a lack of knowledge, the association said. About 19 percent reported having “expert knowledge” of the technology, “while another 29 percent classify their knowledge as moderately high.”
Sens. Rob Portman, R-Ohio, and Martin Heinrich, D-N.M., introduced legislation Tuesday to set a five-year national artificial intelligence strategy authorizing $2.2 billion in federal AI spending. The Artificial Intelligence Initiative Act would establish a National AI Coordination Office, an AI Interagency Committee and a nongovernment expert committee for research and development. AI workforce development is a focus.
AT&T's artificial intelligence "guiding principles" are human oversight; open source "communities whenever appropriate"; and "ethics, safety, and values" including "our privacy principles and security safeguards." The ISP/MVPD uses "varied, validated datasets and diverse human input," it said Wednesday. "We use a transparent approach to algorithms that includes safeguards." The company monitors "outcomes to ensure accuracy and help minimize biases." When "outcomes are owned by people, no one should be able to claim, 'The machine did it,'" blogged Chief Privacy Officer Tom Moore. "No organization will be perfect, but that’s what humans must try to anticipate, catch and repair." Even as many organizations have advanced their own AI and privacy principles, some widely endorsed ones exist, noted Electronic Privacy Information Center President Marc Rotenberg. The "benchmark for AI policy" are the EPIC-established Public Voice coalition's universal guidelines for artificial intelligence, or UGAI, he said in an interview. Possibly later this month, the Organisation for Economic Co-operation and Development may announce its 38 member countries endorsed OECD guidelines, "which reflect many of the principles contained in the UGAI," said Rotenberg, who has worked on the issue. The U.S. would be among the signers. The White House didn't comment. All OECD members and Argentina, Brazil, Colombia and Costa Rica "are due to formally endorse a new set of AI Principles designed by the OECD, next Wednesday" at the group’s annual ministerial meeting, a spokesperson emailed. AT&T meanwhile actively participates "in discussions with industry organizations, such as Linux Foundation, IEEE and Future of Privacy Forum, on a variety of AI topics," a company spokesperson emailed. That includes "AI and ethics, responsible development and deployment of AI" and machine learning, he added.
Google and Walmart representatives will testify about machine learning before the Senate Communications Subcommittee during a hearing at 2:30 p.m. Tuesday in 106 Dirksen, announced Chairman John Thune, R-S.D. The hearing is to focus on “how algorithmic decision-making and machine learning on internet platforms influence the public.” Witnesses are Google User Experience Director Maggie Stanphill, Walmart Behavioral Science Head Jason Hreha, Center for Humane Technology Executive Director Tristan Harris and MIT Sloan School of Management assistant marketing professor Dean Eckles.