IT Home News April 13 Weve seen the power of artificial intelligence (AI) in many news reports. Googles Alpha Go has defeated two top chess players, Li Shishi and Ko Jie, and won Go, which has always been a dominant game for human beings.
_Graph from Pixabay
But according to foreign media Medium, Google DeepMinds performance in a new study was somewhat shocking. In a new paper, researchers took DeepMinds high school math exam to analyze the mathematical reasoning ability of neural models, but DeepMind failed.
Using mathematical problems, the researchers tested two LSTM models for DeepMind. According to reports, these topics are based on the National School Mathematics Course (for 16 years and younger), covering algebra, arithmetic, calculus, comparison, measurement, polynomial operations and probability, etc. The test accuracy of AI model is 35% for 40 problems, and it is a solid fail.
This study has been published in arXiv. Here are 40 senior high school math questions used in the study.
Source: Editor-in-Charge of IT Home: Wang Fengzhi_NT2541