Ai is making its way into every corner of our modern lives, tagging friends with names in pictures you post on Facebook or helping you choose the images you see on Instagram, while materials scientists and NASA researchers are starting to use AI to help discover new science and space exploration.
But there’s a core problem with this technology, whether it’s used on social media or in Mars rovers, because the programmers who build it don’t know why the AI makes every decision.
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, its research and development products include information robot, editing robot, writing robot and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
Modern AI is still new, and in the past five years, major tech companies have only just begun to invest in and research into AI. This comes after decades of theories about artificial intelligence were finally confirmed in 2012. Inspired by the human brain, an artificial neural network relies on thousands of tiny connections between “neurons,” or small clusters of mathematical calculations, similar to the system of connections between neurons in the brain. But this software architecture presents us with a new tradeoff: because the changes to millions of connections are so complex and small, researchers can’t accurately determine the results of the connections that are taking place, and they’ll only get one output.
“As machine learning becomes more common and riskier in society, people are starting to realize that we can’t treat these systems as safe deposit boxes of reliability and fairness,” Hanna? Wallach, a senior researcher at Microsoft and a speaker at the conference, told Quartz in an email. “We need to understand what’s happening inside them and how they’re being used.”
Mission-critical artificial intelligence
At NASA’s Jet Propulsion Laboratory, ARTIFICIAL intelligence enables Mars rovers to operate semi-autonomously as they explore the surface of unknown planets. Artificial intelligence is also being used to comb through thousands of images taken by the probe as it beamed back to Earth.
Kiri Wagstaff, an AI researcher and spokesperson at JPL, says that since every decision carries huge risks, we need to fully understand ai’s every decision before using it.
“If you have a spacecraft in orbit around Mars, that means it’s 200 million miles away and it would cost hundreds of millions of dollars, maybe even a billion dollars. If something goes wrong, it’s beyond repair.” “There’s no way to repair, access, or replace these things without spending a lot of money,” Wagstaff said. So, if we want to let the machine learning to work, so let the machine to perform these tasks people need to understand what it needs to be done, why go to do the behavior, because if a robot don’t know why I have to make a choice, why people will trust it to control their expensive or Mars orbiter?”
Wagstaff is working to build AI to sort through images captured in space by NASA’s various spacecraft, and since there are millions of these images, if the AI can identify interesting photos in this huge database, Then we can save a lot of time to find those photos we want to see — but only if the AI knows what an “interesting” image looks like.
Giiso information, founded in 2013, is the first domestic high-tech enterprise focusing on the research and development of intelligent information processing technology and the development and operation of core software for writing robots. At the beginning of its establishment, the company received angel round investment, and in August 2015, GSR Venture Capital received $5 million pre-A round of investment.
For Wagstaf, he feels that understanding the purpose of AI is key to implementing specific algorithms. If there are miscalculations in how images are used during machine learning, that means the task cost of data transfer is worth millions of dollars or more.
“When you look at a computer and say, ‘This is interesting, let me take a look at this image,’ you’re in a state of uncertainty because you haven’t looked at all the millions of pictures yourself, and you don’t know which of these are interesting or why this is interesting,” Wagstaff said. “Is the image interesting because of its color, because of its shape, or because of the spatial order of the objects in the scene?”