This week’s list of data news highlights covers April 25-May 1, 2020, and includes articles about using AI to improve tax policy and using delivery robots for contactless delivery.
Researchers from Salesforce have developed the AI Economist, a system that uses reinforcement learning algorithms to model how tax frameworks can affect economic productivity and equality. The system’s AI agents attempt to maximize a goal, such as earning a high income with minimal effort, to simulate how people might react to different policies. The system identified policies that could both increase productivity and reduce inequality compared to U.S federal income tax rates.
Facebook has open-sourced Blender, a conversational chatbot it developed. Facebook trained Blender on 1.5 billion Reddit conversations as well as on additional datasets that provided it a sense of empathy, domain knowledge, and personality. Nearly half of human evaluators thought Blender’s conversation logs were more likely to be human conversations than actual conversations between real people.
Nuro, an autonomous vehicle startup based in Silicon Valley, is using its autonomous delivery robots to deliver linens, food, and protective equipment to two COVID-19 field hospitals in Sacramento. The robots, which unlock when a worker flashes a thumbs-up signal, allow workers to maintain social distance.
BrainBox AI, a startup based in Montreal, has developed an AI system that boosts the energy efficiency of commercial buildings. The system uses a variety of data, such as HVAC and weather data, to predict internal building temperatures up to 72 hours in advance. The system, which autonomously controls a building’s HVAC system, reduced the energy bill of a Holiday Inn in Montreal by 30 percent.
Researchers from Expert Systems, an Italian firm that creates AI systems, and Sociometrica, a Spanish analytics firm, have used natural language processing algorithms to assess how social media users feel about coronavirus. The researchers analyzed the sentiment of 44,000 English-language posts about coronavirus, finding that “fear” displaced “sadness” as the most common emotion on April 24. Nonetheless, the researchers also found that posts with negative feelings declined from 62 percent to 45 percent of all relevant posts over the previous ten days.
Chinese autonomous vehicle startup AutoX has partnered with AutoNav, a navigation company owned by Alibaba, to enable individuals in Shanghai to ride-hail autonomous vehicles. While other autonomous vehicle companies have offered rides from dedicated locations in a city, AutoX is allowing users to order a ride to and from any location. An algorithm in AutoNav’s app, Amap, will determine whether to send a human-driven vehicle or an autonomous vehicle based on estimates of which would arrive first.
The Washington, D.C. based developers of OurStreets, an app that crowdsources information on bad driving behavior, have redesigned the app to crowdsource data on what items are in stock in stores. The app collects and displays user-sourced data on the prevalence of items such as toilet paper, eggs, and hand sanitizer. Users can search the app by location and product.
Researchers from BenevolentAI, a startup based in London, used AI tools to identify an arthritis drug that could treat coronavirus. The drug, baricitinib, fights unwanted activity from the immune system. The researchers used AI language models to scan millions of scientific documents and generate a database of biological processes related to coronavirus. This work allowed the researchers to map connections between genes and biological processes and identify medications that targeted those genes. The U.S. National Institutes of Health is testing the drug in an accelerated clinical trial.
Manna Aero, a logistics and supply chain company based in Dublin, is using autonomous drones to deliver medicine to individuals’ homes in Ireland. The drones, which carry up to nine pounds, deliver medications that local general practitioners have prescribed to patients after a video consultation.
Researchers from the University of North Carolina have used commuter data from the U.S. Census Bureau to define metropolitan boundaries more accurately. Government assistance programs rely upon these boundaries for activities such as allocating housing subsidies and funding for infrastructure projects. The researchers measured the number of commuters who crossed over and commuted within county lines, finding that most traditional boundaries underestimate the actual size of a metropolitan region.