For every person who is afraid that robots are suddenly going to become sentient and we’ll have to fight a liquid robot sent from the future to kill us and our moms, this story will give you hope that we still stand a chance.
MIT Tech Review reports that the Allen Institute for Artificial Intelligence held a global contest to see whether modern AI could go beyond “superficial intelligence” — i.e. recognizing a photo of a dog or completing repeating tasks like playing endless games of chess. That test ended up being a real test, in 8th grade science to be exact.
“The most awesome stage”
Last year, Facebook's VP of Design thought the TNW Conference main stage was the best she'd ever been on.
The winner, Israeli contestant Chaim Linhart, used a combination of machine learning techniques and hundreds of thousands of practice questions to prepare his AI for the exam. The machine took the multiple-choice test and answered 59 percent of questions correctly — earning itself a big fat F. Or, perhaps, a Gentleman’s C.
Still, that score is major for an AI to accomplish, even if the test was modified to exclude diagrams and other variables that your average middle-schooler could handle with aplomb.
Moreover, it shows how much larger the channel of intelligence will be to cross for AI. Intelligence isn’t about responding to stimuli — it’s also about critical thinking and decision making. Now, when the robots begin to strategize, that’s when I suggest you panic.
➤ The Best AI Program Still Flunks an Eighth-Grade Science Test [MIT Tech Review]