A Computer Program Can See the Future
Researchers at the University of Bronn may have developed a computer program that can see into the future!
While the program is not going to determine major events around the world, it is still very impressive.
The program, developed by computer scientists, is able to predict actions that will happen a few minutes into the future. And it was made using cooking videos!
Predicting the Future with Programming
The computer scientists at the University of Bronn had an idea. They wanted to create a computer program that could predict what was going to happen based on the action it was seeing.
They chose cooking videos to create such a program. The idea is that a cooking video is played and the program is able to accurately predict what the chef will do in the next few minutes, based on what is happening in the moment.
It is not just a fascinating concept to have a computer program that can predict a few minutes into the future. There are practical uses.
For instance, a kitchen robot with AI could be able to predict what steps have to take place at what moment, based on the dish being cooked and what is happening. The robot would know when to add certain ingredients, when to take something off the stove, when to stir and when to set the oven for preheating.
These are the types of actions that humans are able to perform as second nature, especially after a bit of practice. But it is not so simple for computers.
Training the Program
When it comes to this program, the computer scientists put it through hours of “training” before seeing if it could predict the future.
The program watched around 40 hours of videos where people were making salads. The recordings were around 6 minutes in length and had around 20 different actions being performed.
The program would “watch” the videos and its algorithm would learn what actions were following each other. It is not just about the order, but the amount of time in between different actions.
Time to Predict
When the program had watched all that training footage, it was time to see if it had learned anything.
The computer scientists would show the program around 20 or 30 seconds of a salad preparation video it had not seen before. And the goal was to predict what would happen for the rest of the roughly four minute video it had not seen before.
The results were impressive. The program managed to predict the correct outcome around 40 percent of the time – for short forecast periods.
When the activity was more than three minutes into the future, the program was only accurate 15 percent of the time. Accuracy refers to both the correct action and the timing of the activity.
The issue in a computer program predicting the future in cooking videos is that every chef has their own method. If one were to watch the same type of salad being prepared by five chefs, they would all do things at different intervals.
But the future of computer learning is here. It is only a matter of time before computer scientists succeed in teaching computers how to predict events with greater accuracy and further into the future!