| Robot Society 1 |
February 26, 2018
|
| Recently, he frequently watched news about AI robots on TV and read about them in newspapers. He was concerned about the future world. "Will robots take all of our jobs?" "If they become superior to humans in many points, will they start to order humans around?" So, he decided to study about the possibilities of a 'robot world' through the Internet during one of his holidays. After a while, he got some knowledge about AI (artificial intelligence) and AI robots: 1. Before AI appeared, programmers programed, and robots only moved as they were programmed. 2. The notion of AI was first introduced in 1956 by a scientist in the US. 3. AI's biggest feature is 'deep learning'. 'Deep learning' means that computers think for themselves. AI would find features in common from a lot of information and lead to some theories. For this, AI needs as much information as possible. The Internet world has enabled this. For example, for an AI robot to say, 'This girl is pretty.' AI would first need to get an uncountable amount of girls' pictures with comments like, 'She is pretty,' or 'My girlfriend is pretty, you see?" From there,  AI would start to find the features that those pictures had in common. Then it would get some theories for 'pretty'. Of course, all comments are not always correct. The more information AI gets, the more the AI's notion becomes closer to human's general notion. 4. Sometimes programmers help the AI units to get information suitably. If you want to make a poker computer game, you can program the computer to get poker game videos from the Internet. The AI learns how the game proceeds and how to win. You don't need to program all the rules of the game because the AI can learn intuitively from the games themselves. 5. After AI robots received eyes (cameras), their ability to learn broadened extremely. AI robots started to move by themselves. Robot arms could recognize parts of products and build them suitably. Auto drive cars could avoid hitting walls. 6. AI robots learned languages. AI searched for the words 'apple' with the pictures of 'apple' on the Internet. The AI first got the features in common of those pictures. Next, the AI found sentences that used other words featuring them, like 'round' and 'red'. When the AI found the same shape for example the 'earth', it could make a new sentence like 'The earth is round.' The same thing happened to 'red'... They also studied about the function of 'is'. From many sentences that used 'is', the AI could guess 'is' is familiar to those types of words. The AI doesn't need to categorize words as 'nouns' or 'adjectives', it just learns how the word is 'usually' used. So, it is different to say that AI 'understands' a language. They 'get' how the sentence is made. After the computers learned two languages, the programmer gave them many books that were sets of books written in two languages. The computer started to compare each series of the two books. And learned the rules about how to translate between these languages. 7. If an android robot --human-shaped and human-like moving robot--says to you, "Hey. You look sad. Did your girlfriend say something wrong?", it doesn't mean that the AI robot understands your feelings. The robot just pretends to feel sympathy or empathy and just reacts as most people react about the lives that they view through the Internet. The robot got your looks by its camera and analyzed your face that showed 'sadness' because the type of 'sad' pictures were shown with the word 'sad' on the Internet. The computer might get that a 'girlfriend' sometimes says bad things to their boyfriend from a lot of comments on the Internet. He was very relieved. Before he saw on YouTube that a female android replied to an interview very naturally like a human. He had thought that at this pace he would have a chance to get a humanoid girlfriend, but in reality they were just 'pretending' that they were like a woman. 8. About a 'Go' game. 'Go' game is played with white round stones and black stones. Before, computer engineers and scientists had thought that computers couldn't beat human players because 'Go' is too complicated. If the computer calculate all the possibilities on the 19 × 19 cross points (= 361), and if the computer needs 0.01 seconds for each play, it would take 10 billion years! But, eventually, one AI computer defeated the best Go player in the world. The computer got information from many professional players' games, and basing its strategy on this 'experience' the computer could simulate each move an uncountable amount of times. One specialist said that since an AI won in one of the most difficult games, in other competitive worlds, for example investments like the stock markets, AI would have success. On this day, after studying about AI robots, he remembered an old saying, "Any difficult sentences can be understood, if you read them 100 times." Maybe AI robots could read a long story 100 times in a second. He sighed. |
| |
*superior :より優れている
*feature :特徴
*in common :共通の
*general :一般の、大衆の
*notion :考え、意見
*intuitively :直観的に
*proceed :進む
*broaden :広げる
*adjective :形容詞
*pretend :振りをする
*sympathy :同情
*empathy :共感
*saying :ことわざ