An esteemed group of computer scientists at the University of Massachusetts Amherst has engineered a revolutionary AI framework called DISCount. This AI tool marries the swiftness and vast data processing prowess of artificial intelligence with the dependability of human analysis, enabling swift and accurate results from extensive collections of images.
DISCount emerged as a result of two contrasting applications. The objective was to address the challenge of rapidly detecting compromised buildings in crisis zones, as well as estimating the size of bird flocks accurately.
The team had been collaborating with the Red Cross for years, aiding them in developing a computer vision tool to accurately count buildings affected during catastrophic events such as earthquakes and armed conflicts. Simultaneously, they have been assisting avian scientists from Colorado State University and the University of Oklahoma in interpreting radar data to gather precise estimates of bird flock sizes.
Guided by the belief that computer vision could potentially solve these problems, they continued their research into devising automated tools that can offer a higher degree of reliability. Their intention was to create tools that non-AI experts could use.
DISCount is a versatile framework that can pair with any existing computer vision model. It operates by scanning vast data sets — for instance, all images taken from a particular region over several years — to determine a specific subset of data suitable for human inspection. The model can sieve out images that best illustrate the extent of damages to buildings in that area. This lets a human researcher review a smaller set of images to perform a manual count of the damaged buildings.
DISCount operates with any image-based AI model and offers a prediction of the accuracy level of human-derived estimates. The tool significantly outperforms random sampling for the tasks it has been delegated. Due to its ability to provide a range of accuracy, it empowers researchers to make informed decisions about the reliability of their estimates.
The breakthrough recognition of DISCount's model stems from the realization that the choice isn't between human intelligence and artificial intelligence. Instead, it's about combining the best of both to build a faster, more comprehensive, and accurate tool.
DISCount's pioneering effort has been rewarded with the best paper award for AI's social impact by the Association for Advancement of Artificial Intelligence.
Disclaimer: The above article was written with the assistance of AI. The original sources can be found on ScienceDaily.
