Create a software that guesses your future in the next three minutes
Create a software that guesses your future in the next three minutes
Predictive technology is not something new. For example, when we write a message, the application suggests what we will put next, and although previous generations this may seem "magic", the fact is that most people perceive it as normal and without much mystery. But, what would happen if your mobile told you what action you are going to perform within five minutes without even you having raised it?
This is the objective of the software developed by scientists at the University of Bonn, in Germany. "Our intention is to predict the time and duration of the activities minutes or even hours before they happen," explains the leader of the research team Jürgen Gall. And, for this, it has begun by providing an "aid" to the program, based on artificial intelligence : when introducing the beginning of an activity, it was intended to list a sequence of future actions based on those first steps.
Specifically, they trained a software to guess what a chef would do after showing him several videos of people preparing breakfast or a salad. Afterwards, they showed him the images of another totally different person cooking and he was able to predict the next steps in an autonomous way.
It may seem unspectacular , since the logical thing is that if you prepare a breakfast and you put cereals in a bowl, then spill the milk on top. However, this project is a first step to open the horizons of predictive technology and to be able to make future models of, for example, the future of the planet in future years or that your robotic butler of the future can bring you a lime at that time where stress does not let you keep working. For example.
These first technological predictions gathered a precision superior to 40% in the short prognosis periods (the first 20 seconds), but they diminished in efficiency according to they made longer projections. In fact, if they extended to three minutes, the hits dropped to 15% . Although it does not impress at first sight, these could be the first steps for artificial intelligence to predict human actions. In fact, the results will be officially presented at this year's IEEE Conference on Computer Vision and Pattern Recognition in Salt Lake City , where predictive software will be one of its stars.
There is still a lot of ground to travel, especially since the ultimate goal is for programs to accumulate enough experience on their own without the need for training. That is, that they train for themselves thanks to the cases that they are registering, so that each time they refine more in their "prophecies". But it is a good start.

Comments
Post a Comment