Adaptive robots won't be around anytime soon

[zakiyatasnim]

Let’s say you put the first shelf in,” says Nikolaidis. “The robot doesn’t understand anything yet. Then you take the second shelf. And now you start placing the third. There is a high probability that you belong to the group of users who collected all six shelves in a row. At the same time, the probability that you will change your preferences is quite low.”

When the robot recognizes a sequence of human actions, it presents greece number data the person with the item it knows people in the same group chose next. Experiments have shown that this way the machine is able to quickly and accurately adapt to the person's style.

An image recognition algorithm is built in a similar way. To make it detect cats, you need to feed the neural network lots of images of the feline family. Because the algorithm has seen a lot of examples, it is now able to generalize from them.

If you show it a picture of a cat it hasn't seen before, it can use that knowledge to confirm that it's really a cat. This robot does the same thing. But instead of static images, it uses examples of action sequences—the sequences in which people would assemble shelves and supports.

This research is not aimed at developing highly specialized robots that will go around homes and help assemble shelves. Nor is it aimed at creating machines that can perform such complex tasks on their own. It is designed to teach robots to cooperate with humans, but at the same time not to complicate their work.

There are concerns that robots will steal jobs from humans. However, the reality is that a machine is more likely to work alongside you than replace you entirely. At the moment and in the near future, humans will be much better at some tasks.

No machine can replicate the dexterity of the human hand and solve problems the way humans do. Robots are only good at unconscious tasks. For example, on an assembly line, robotic arms install car doors, but working with small parts requires human intervention.