This week’s list of data news highlights covers May 18-25, 2018, and includes articles about researchers using AI to build microchips on the atomic scale and a project to use brain scanning to identify good bomb-sniffing dogs.
1. AI Could Free the Streets From Taxis
Researchers at the Massachusetts Institute of Technology have developed a dispatching algorithm that could potentially reduce a city’s taxi fleet by 30 percent. Most efforts to reduce traffic congestion through the use of algorithms are either not scalable. For example, solving the “traveling salesman” problem, in which a driver attempts to take the shortest route possible between all possible destinations, is not feasible at the city-level. Others require significant changes in behavior, such as expecting people to be willing to wait longer and carpool. Instead, the researchers attempted to identify the minimum number of taxis needed to maintain the same level of service. In a test on a dataset of 150 million taxi trips taken in New York City over a year, the researcher’s algorithm could shrink the fleet by 30 percent while still retaining optimal service levels.
2. Building Microchips on the Atomic Scale with AI
Researchers at the University of Alberta have developed a system for assembling circuitry for computer chips atom by atom through the use of machine learning. Transistors on a traditional chip use electrons to represent binary information and move these electrons around as a computer writes or stores data. By contrast, the researchers’ approach relies on individual atoms in a circuit to represent binary information based on whether they hold electrons, and these atoms simply exchange electrons with each other to record data, which requires substantially less energy. The researchers used machine learning to develop manufacturing equipment fine enough to place individual atoms, as well as automate a scanning technique to correct errors.
3. Machine Learning Could Stop People from Arguing on the Internet
Researchers at Cornell University, Wikimedia, and Google Jigsaw, which develops tools to promote freedom of expression online, have developed a machine learning system that can analyze conversations and predict whether it will become angry and turn into an argument. The researchers trained their system to identify phrases that indicate a conversation’s mood, such as the use of “please” and statements of gratitude, which indicate a friendly mood, or the use of direct questioning and sentences that start with second person pronouns in a conversation’s first message, which indicate hostility. In a test, the system could correctly predict when an argument would start with 65 percent accuracy, which is less accurate than humans but could eventually allow for automated moderating tools to quickly identify toxic arguments online.
4. Making Cape Town Transit Less Chaotic
A company Where Is My Transport has successfully mapped Cape Town’s minibus taxi system, which is usually the most affordable form of transit but is difficult to navigate, and integrated it with open data about Cape Town’s trains and buses. Where Is My Transport identified and logged 657 unique minibus taxi routes and made this data available so it could be used by city officials, taxi associations, and commuters to make transit more accessible.
5. Sensing Internal Bleeding with Smart Pills
Scientists at the Massachusetts Institute of Technology have created an ingestible capsule that uses bacteria to spot gastrointestinal bleeding and wirelessly transmits this data to a smartphone app. The scientists genetically engineered a safe strain of E. coli to produce a bioluminescent response when it comes in contact with heme, a molecule found in blood, and added this bacteria to a permeable capsule containing a small phototransistor and microprocessor. When the capsule comes into contact with blood in the stomach, the bacteria light up, the phototransistor registers this light, and the microprocessor transmits this data.
Microsoft is developing a dashboard capable of scrutinizing an AI system and automatically identifying signs of potential bias. Researchers behind the project are designing the dashboard to work with a variety of different kinds of algorithms and help developers more easily spot how their systems could produce unfair or discriminatory outcomes. Additionally in May 2018, Facebook announced that it was developing a similar system called Fairness Flow that could automatically flag if an algorithm is making potentially unfair decisions based on a person’s race, gender, or age.
7. Building a Better Lie Detector
Researchers at the University of Arizona and San Diego State University have developed a lie-detecting AI system called Automated Virtual Agent for Truth Assessments in Real-Time (AVATAR). AVATAR analyzes real-time video of a person to analyze biometric information such as eye movements, posture, changes in voice, and other physical cues that could indicate a person is lying more reliably than a polygraph test. The system is between 60 and 75 percent accurate at detecting lies, while humans are only up to 60 percent accurate at judging truthfulness.
8. Tracking Your Diet with Teeth Sensors
Researchers at Tufts University have created a tooth-worn sensor that can monitor data about a wearer’s food intake. The two-millimeter sensor can detect changes in temperature and pH as well as identify the presence of sugar, salt, and alcohol, and uses RFID to transmit this data to a smartphone app.
9. Building a Robot to Replace Bees
Engineers at West Virginia University have developed an autonomous robot called BrambleBee that can navigate a greenhouse and pollinate flowers. BrambleBee first creates a 3D map of a greenhouse to identify the most efficient route between plants, uses computer vision to identify the positions and conditions of flowers, and then uses a robotic arm to brush each flower’s pollen.
10. Finding the Best Bomb-Sniffing Dogs with Brain Imaging
Cognitive behavioral researchers at Auburn University that oversee Vapor Wake training, which trains dogs to identify explosives for law enforcement, are using fMRI scans to identify which puppies are best suited for the job. The researchers subject puppies to various cognitive tasks, such as locking a treat in a box to see how long it takes for the dog to recognize it needs help to get the treat, while an fMRI machine analyzes the dog’s brain activity. The researchers have been able to identify correlations between different patterns of brain activity and certain desirable cognitive traits for dogs.