Artificial intelligence predicts human actions better than humans themselves

MIT specialists at the Massachusetts Institute of Technology described the successful launch of a prototype of their new Data Science Machine. A learning artificial intelligence system automatically finds predictable patterns in datasets. In the competition, which was attended by over 900 teams, AI proved to be excellent, surpassing two-thirds of the participants.

People and AI took part in the three stages of the competition. In two of them, the accuracy of AI conclusions was 95% of the accuracy of the result of the team that won first place. In the third - 87%. At the same time, the participating teams took several months to calculate the results, and the system managed them in no more than 12 hours.

“Our Data Science Machine is a natural complement to the human mind,” says Max Kanter, whose graduate work at MIT is one of the main components of AI. - There are so many different data in the world. And now they do not bring any benefit. Maybe we can do something that will allow us to start processing them and we will move forward. ”

At MIT, students work with online courses, so good statistics have been collected on their use. At one stage of the competition, it was necessary to predict whether the student would quit working with an online training course in the specified time period. The basis for the analysis was detailed statistics on the work of this student with the specified course.

Thanks to the analysis of data using AI, it turned out that the most important factors influencing the question asked are the number of days remaining until the date of the assignment and the total time spent on the site. Moreover, it was precisely such parameters that were not specifically presented in the statistics, but they could be derived from other data.

The other two stages included checking the crowdfunding project for “interest” for visitors and predicting whether the buyer of the online store would become a regular.