人工知能とは人の延長線にある? Artificial intelligence is an extension of humans


 少し前にビデオ講座でAIに関する知識を入れました。と言っても、大学生レベルらしいので実用的に活かせる場面もないわけですが、とにかく驚いたのは、学習目的で利用する場合に限り無料で利用可能なアプリとか、クラウドサービスとか’思った以上に揃っていること。

自然言語処理、画像解析、深層学習などワクワクする専門用語が満載でした。また、反対に意外だったのが、教えるのは人だったということ。(これも、深層学習などではその様な人間の先生も必要ないくらい進んでいるのですが、そこらへんをやるとサンプルを集めるだけで時間がかかりすぎて、まだ現実的でない様子なので今はなし)短時間で、判定効果を高めるには教師あり学習と言って(画像解析などがわかりやすい例だと思いますが)写真に写る犬と猫を見分ける方法。まず、人がみて、画像に写るのが猫か犬か画像を判断し、AIがどちらかわかる様に記号をつけてゆき、一定数用意する。それを前処理で学習した上で、その後、本番画像を判断させてゆくもの。

I learned about AI in a video course a while ago. That said, it seems to be a college student level, so there are no scenes where it can be put to practical use, but what surprised me was that apps that can be used for free only when used for learning purposes, cloud services, etc. It is aligned with.

It was full of exciting technical terms such as natural language processing, image analysis, and deep learning. On the contrary, what was surprising was that it was people who taught. (This is also advanced to the extent that such a human teacher is not necessary in deep learning etc., but if you do it there, it will take too much time just to collect samples, so it seems that it is not realistic yet, so it is not now) To improve the judgment effect in a short time, it is called supervised learning (I think that image analysis is an easy-to-understand example), and it is a method to distinguish between dogs and cats in the picture. First, a person looks at the image, determines whether the image is a cat or a dog, adds a symbol so that the AI ​​can tell which is the image, and prepares a certain number. After learning it by pre-processing, it is made to judge the production image after that.




画像は計算式で解析され、その結果が三次元グラフに展開される様で、三次元空間に答えを示す点を打ってゆく。三次元グラフを見れば点がたくさんある部分が見てとれて、なんとなく点の色が濃く見える領域がわかるというもの。これで、写るものが異なれば、濃くなる領域が異なるので犬か猫か判断できるというもの。多分あってるはず。

The image is analyzed by a calculation formula, and the result seems to be expanded into a three-dimensional graph, and points are struck in the three-dimensional space to show the answer. If you look at the three-dimensional graph, you can see the part with many points, and you can see the area where the color of the points looks dark. With this, if the image is different, the darkened area will be different, so you can judge whether it is a dog or a cat. Maybe it's right.


 少し脱線しますが、もっと興味深いのはイギリスのネット講座で「クリエイティブAI」という講座があった事。日本でのビデオ講座を受講する少し前に、知り合いの方がその記事を取り上げていた(推していた)ので、全部英語のネット講座でしたが興味本位で講座を始めてはみたものの、残念ながら今は途中放棄した状態になってます。それでも、使う人が意図しないが、好ましい答えが出る様にチューニングできればこれまさに「クリエイティブAI」として広く受け入れられそう。

It's a little derailed, but what's more interesting is that there was a course called "Creative AI" in the UK online course. Shortly before taking the video course in Japan, an acquaintance picked up (recommended) the article, so it was all an English online course, but I started the course with an interest in mind, but unfortunately now Is in a state of being abandoned on the way. Even so, although it is not intended by the user, if it can be tuned to give a favorable answer, it will be widely accepted as "creative AI".



話を元に戻すと、AI(artificial intelligence)はまだまだ人を教師として、早く物事を判断する便利なプログラム、ということができると思っていて、先に出てきた深層学習も最初から自律的にアウトプットできるわけではなく、十分に人の関与、チューニングが必要であること。もしくは、先に出てきたイギリスの通信講座例を考えてみると、予想外の答えを出すマジックボックスとでも言えるかも知れない。ここまで考えてみると、人により十分干渉すべき超便利な道具、と解釈できそうだ。やはり、今はまだ人が色濃く反映される存在だと考える次第である。

Returning to the story, I think that AI (artificial intelligence) can still be a convenient program to judge things quickly with people as teachers, and deep learning that came out earlier is also autonomously output from the beginning. It is not possible to do so, and sufficient human involvement and tuning are required. Or maybe it's a magic box that gives an unexpected answer when you consider the example of England. If you think about it so far, it seems that it can be interpreted as a super convenient tool that should interfere with people enough. After all, I think that people are still strongly reflected now.

閲覧数:15回0件のコメント

最新記事

すべて表示