It was a very exciting week in the world of artificial intelligence, which received a huge push forward
Nimrod ben Moshe, Tamara Traubman, Haaretz
AI - artificial intelligence, is both a dream and a nightmare at the same time. We all dream that one day we will have a robotic servant who will understand exactly what we say to him and know how to call for help on his own in case of need. But at the same time we are afraid of that intelligence, mainly because we know it is the secret of our power, what separates us from the rest of the creatures on earth. And because of all the movies that convinced us that the first thought a computer would have would for some reason be "we need to kill all humans".
Still, scientists all over the world are working hard on the development of artificial intelligence and in the last week two amazing breakthroughs have been made. The first is a new programming language called ISO 18629 (good name for the band). This programming language allows the computer to "visualize" the results of its actions.
When we give a computer a command, it is able to calculate in advance what will happen when it completes the command. This allows us to give the computer a command like "package the product, but only after you have painted it". ISO 18629 allows the computer to "imagine" itself packing the product and then coloring it, or coloring the product and then packing it. At the end of the process, the computer can decide on its own that the second option is more suitable for the general purposes of the plant.
Forward thinking and the ability to plan is an ability found, as far as is known, only in humans and a limited number of great apes. Now, also with computers.
Rock Paper Scissors
By Tamara Traubman, Haaretz
They call him "the agent", and he is a computer that knows how to play rock, paper and scissors. No one taught him the rules of the game - he learned them himself, by watching and imitating people.
The "agent" is a computerized system developed as part of the CogVis project at the University of Leeds in Yorkshire, England. He "learned" to play the famous children's game by watching people playing the game and looking for repeating patterns. In this way, says Dr. Chris Needham, a member of the group that developed the system, he deduced what the rules of the game are.
Although computers today know how to play many games, even sophisticated games like chess. However, these computers operate according to fixed rules entered into the system by a programmer. The problem with all seemingly intelligent programs is that as soon as they encounter an unexpected situation, which is not included in their database, they do not know how to react.
"The computer was not programmed with the rules of the game," said Dr. Needham in an interview with Haaretz, "instead we equipped it with the necessary tools for learning." He did the rest by himself. First, the computer watched a video, in which two people played rock, paper and scissors with cards on which the three objects were drawn. The players were asked to announce each time which "hand" won.
Advertisement After a few rounds - according to Dr. Needham, "not much longer than it would take a human" - the computer was able to accurately announce by itself who the winning card was and who the losing card was. According to him, in just ten minutes the system gathered the data it needed, and in a few more minutes it correctly identified the results of the game.
"A system that can observe events that are part of an unfamiliar script, learn and participate as a child would, is almost the holy grail of artificial intelligence," said Dr. Derek Magee, another member of the team of researchers from Leeds, in an official announcement they published. "We may not have achieved this challenge yet, but we have certainly gone a long way towards it."
Needham does not believe that it is possible to build a computer that really thinks. He is also not interested in such a computer. "It may be possible to build robots that look like they can think," says Dr. Needham, "I think we can build robots that can learn how to behave - but not robots that actually think."
According to him, "We want to build machines that behave as required. Machines are good at doing uninteresting tasks. If we can teach a machine to perform a task, such as industrial control, then we will no longer have to program it - which can be difficult and time-consuming."
An artificial intelligence expert
https://www.hayadan.org.il/BuildaGate4/general2/data_card.php?Cat=~~~195198357~~~207&SiteName=hayadan
