In the new Natural Artificial Intelligence research program, scientists use strategies from nature to make artificial systems more intelligent.
There are wonderful solutions to problems that scientists are working on in nature. Consider image recognition: although humans are adept at identifying images, machines are not. Birds will fly in groups without colliding, posing a significant threat to drones. As a result, scientists will learn a lot from nature. Artificial intelligence, on the other hand, may give insight into how natural processes function. Scientists cannot unscrew a person’s development, but they can use models to determine which form of language learning is most efficient.
If the receptionist is a robot in the future, and a chat robot staffs customer service, you should be able to communicate naturally with those robots. But how can one teach an automated machine to use language naturally? Paul Vogt predicts that ‘By learning to connect in the same way we do with young children.’ He is a researcher who focuses on language acquisition and evolution. He also uses artificial intelligence techniques.
“The goal is a system that can learn from people, is flexible and adapts to its environment”
Vogt and colleagues from Delft and Tilburg create computer models of a family in this study. ‘We make computer children and computer parents.’ We allowed them to play language games. Then we’ll look at how computer children communicate with one another.’ Human experiences would be fed into the computer family, taking cultural distinctions into account. ‘In an earlier research experiment, we made video clips of parents interacting with their child. ‘We annotated these images and translated them into facts that a machine would understand.’
Learning gestures on a computer
The goal is the development of an intelligent system that is good at interacting with people. ‘A system that can learn from people, is flexible and adapts to its environment.’ In addition, Vogt hopes for more insight into how children learn to communicate. ‘Once the system works, you can also test which factors determine good language acquisition. That in turn gives us insight into how that works for children. ‘
Would the valuable furniture collection still be intact? That was a major concern when a fire in 2008 destroyed the Delft Faculty of Architecture. Due to the risk of collapse, no one was allowed to look during the first days. Guido de Croon therefore flew a drone past the building, hoping to see the furniture through the windows. ‘Unfortunately that did not yield any results. I would have preferred to have flown into the building, but that was not possible with that drone. ‘ Later it turned out that the furniture was largely undamaged.
Bees as a source of inspiration
Fortunately, there is evidence that it is possible. After all, bees do all that drones may soon be able to do in nature. They are less than a tenth of a gram in weight and have no brains, but they can swarm and navigate just fine. Bee research provides plenty of inspiration for De Croon and his colleagues. According to De Croon, nature serves as a warehouse full of good ideas for industrial technology. He describes how drone makers are now making effective use of a trick that aids insects in landing. ‘Insects use a ruse’. The world travels faster and faster across their picture as they get closer to the ground. They make use of this fact; they are aware that they must slow down. They always land softly this way.