Dr. T's AI brief
dtau...@gmail.com
How Game Theory Changed Poker
The Wall Street Journal
Oliver Roeder
January 13, 2022
Researchers at the University of Alberta's Computer Poker Research Group in Canada pioneered game theory mathematics that has transformed how professional poker players approach the game. Poker's mathematical complexity rivals or surpasses that of chess while adding randomness and hidden data, bringing it closer to the "real world" that artificial intelligence scientists want to control. Many poker-playing algorithms incorporate the minimization of regret, a mathematical concept for decision-making in uncertain environments. Game-theory optimal poker players hire programmers to analyze their game data, finding "leaks" or errors in strategy, and to conduct game-theoretical analyses, calculating optimal plays in any of the innumerable situations that can confront a player.
*May Require Paid Registration
Economists Pin More Blame on Tech for Rising Inequality
The New York Times
Steve Lohr
January 11, 2022
Some economists blame escalating inequality on the automation of tasks formerly done by humans, in addition to excessive technology investment and supportive public policies. Stanford University’s Erik Brynjolfsson warns of technologists, business people, and policymakers falling into “the Turing trap,” the assumption that artificial intelligence (AI) can match human performance, which leads to AI systems that replace people rather than augmenting their performance. Massachusetts Institute of Technology's Daron Acemoglu and Boston University's Pascual Restrepo determined "so-so technologies" that replace workers without raising productivity underlie sluggish productivity growth. Acemoglu endorses directing technology development along a more “human-friendly" path that works for, and with, people.
*May Require Paid Registration
Scientists Use Summit Supercomputer, Deep Learning to Predict Protein Functions at Genome Scale
Oak Ridge National Laboratory
January 10, 2022
Scientists at the U.S. Department of Energy's Oak Ridge National Laboratory (ORNL), Google's DeepMind, and the Georgia Institute of Technology (Georgia Tech) are inferring genomic-scale protein functions using supercomputing and deep learning tools. The researchers used ORNL's Summit supercomputer, the fastest system in the U.S. and second-fastest globally, to model the full proteomes for four microbes, two of which generate valuable materials for manufacturing plastics, while the other two can break down and transform metals. A key computational tool developed at Georgia Tech, the Sequence Alignments from deep Learning of Structural Alignments, can compare genetic sequences by implicitly understanding protein structure, even when they have only 10% in common.
Robotic Arms Use ML to Reach Deeper into Distribution
The Wall Street Journal
Jennifer Smith
January 10, 2022
Robots increasingly are being used in warehouses to sort, pack, and prepare orders for delivery as logistics operators faced with labor shortages turn to automation to meet high demand. Advances in computer vision and software have allowed warehouse robots to take on more tasks previously handled by human workers. Puma North American Inc. is using robotic arms from Nimble Robotics Inc. to prepare clothing and shoe orders at a California distribution center, with plans to implement robots at another facility in Indiana. Puma's Helmut Leibbrandt said the robots can work two consecutive shifts and perform with about 99% accuracy, on par with human workers. Hasan Dandashly at logistics and manufacturing automation provider Dematic Corp. said it makes the greatest financial sense to use robots to pick orders in 24/7 operations with a limited number of products. Said Dandashly, "I don't think we are on the verge of not having human pickers anytime soon."
*May Require Paid Registration
Outsider Wins DARPA Challenge to Predict Where Floats Drift at Sea
New Scientist
David Hambling
January 10, 2022
A satellite engineer without oceanographic or meteorological expertise won the U.S. Defense Advanced Research Projects Agency (DARPA) Forecasting Floats in Turbulence Challenge, outperforming 31 other teams in predicting where floats will drift on the open sea. California-based Chris Wasson forecast the locations of 90 devices drifting in the Atlantic over 10 days, based on the previous 20 days of movement and meteorological information on currents, wind, and waves. Wasson simulated the effect of wind and surface currents on each float; for the first 20 days, he compared forecasts with actual positions to refine the model, using machine learning and mathematical modeling. Said Wasson, “Machine-learning approaches may suggest non-obvious solutions to problems and analytical methods can help to validate and explain those results."
AI Tool Could Help Diagnose Heart Failure
Imperial College London (U.K.)
Ellyw Evans
January 7, 2022
Combining a smart stethoscope with an artificial intelligence algorithm for early point-of-care heart failure diagnosis could improve patient outcomes at less cost, according to researchers at the U.K.'s Imperial College London (ICL). The algorithm, in conjunction with a stethoscope that records electrocardiograms and heart sounds, could determine a heart is exhibiting weak pumping action within 15 seconds, yielding 91% sensitivity and 80% specificity versus routine diagnostic tests. ICL's Patrik Bachtiger said, "This super-human capability to screen patients at any point of care, including the general practice surgery, can overcome the unacceptable reality that 80% of patients with heart failure are currently diagnosed through an emergency hospital admission."
Health Datasets Could Help AI Predict Medical Conditions Earlier
Financial Times
Madhumita Murgia
January 3, 2022
The Nightingale Open Science health dataset cache launched in December by the University of California, Berkeley's Ziad Obermeyer could help train artificial intelligence to forecast medical conditions earlier. The datasets, each curated around an unsolved medical mystery, include 40 terabytes of imagery from patients, with each image annotated with the patient's medical outcomes. Obermeyer compiled the datasets over two years from hospitals in the U.S. and Taiwan; he made them free to use, and intends to expand the trove to Kenya and Lebanon in the months ahead. "What sets this apart from anything available online is the datasets are labeled with the 'ground truth,' which means with what really happened to a patient and not just a doctor's opinion," Obermeyer said.
*May Require Paid Registration
AI Hiring Bias Spurs Scrutiny, Regulations
Bloomberg Law
Erin Mulvaney
December 29, 2021
Artificial intelligence (AI)-related hiring discrimination has prompted regulatory action, with New York City banning employers from using automated employment decision tools for screening job applicants in lieu of a bias audit. Meanwhile, District of Columbia Attorney General Karl Racine has announced proposed legislation to address algorithmic discrimination by mandating annual corporate technology audits. The U.S. Equal Employment Opportunity Commission's Charlotte Burrows said up to 83% of employers, and as many as 90% of Fortune 500 companies, use automated tools to screen or rank job candidates; she warned these technologies "could be used to mask or even perpetuate existing discrimination and create new discriminatory barriers to jobs." Civil rights groups like the Surveillance Technology Oversight Project (S.T.O.P.) worry that New York's measure could enable more AI bias, and have proposed banning biased technology altogether.