When you think of a smart home, you think about your fridge always being stocked, your bath always being ready on time, your shower always being the perfect temperature, and your TV always being tuned to the right channel – very rarely does the idea of an assistant chef enter the equation. But researchers at the University of Bonn in Germany have decided to add that to the list, too.
Led by Professor Doctor Jurgen Gall, a working group from the University’s Institute of Computer Science have developed an AI whose purpose is to predict future activities by analysing past similar situations. In tests, the AI “watched” 40 short videos – each approximately six minutes long – of people preparing salads, and then was shown only the first 20% to 30% of new salad-preparation videos and asked to predict the rest of the actions that would occur. Each video included about 20 actions.
The software had a 40% accuracy rate for short forecasting periods, but the accuracy dipped dramatically the longer it had to continue with its predictions. Now, 40% accuracy over a period of a few minutes may not sound terribly impressive but the results need to be taken in the context that one, this was a software trying to predict human behaviour, and two, it’s a first foray into the new field of activity prediction.
Watch the short results video below.
Via Science Daily.