(Sebastian) Okay. Here's from Custer Kriptis in Greece.
"Thanks a lot for this class. Are you providing a downloadable form of this course for people that completed it?
"I would personally want to archive the lectures, given that I might want to refer to it in the future. Thank you, professors."
(Peter) Yes, I think we covered that before. I think they'll always be there on YouTube. But we should package them up
in a way to make it easy to gather them all together at once. So we'll look into trying to get that done.
(Sebastian) Yeah. I also hope that the website will just stay up, and then people can just always go back.
I find that the course the way we lay it out is not meant to be an encyclopedia or a reference book,
so if you want to go back, it might be hard to find stuff, as you probably experienced.
It's really meant to be for one-time consumption, you just go through it for the experience.
So we might, in future months and years, add more information on how to find things and how to access things
very quickly once you've learned it and just want to refresh it. That's clearly, I think, a deficiency of the current system.
Evo from Houston, Texas says, "I agree with you to try things out instead of just reading books.
"However, could you say a bit about types of materials you should/could study in order to get a broader foundation
"in the field of A.I.? And thanks a bunch for this great initiative. Evo."
Well, obviously there's one book that stands out among all others, one book that captures 95% of the marketplace,
one book that has been translated into more languages than any other book,
and clearly, chapter 26 is the best chapter, because I wrote it. (laughter) But this book so happens to be written
by a person that sits on my left, you wouldn't believe it. And I highly recommend it.
Peter, you said that you would take the proceeds of the sales this quarter and donate them?
(Peter) That's right, and donate them to charity. My publisher did notice that the sales were going up
because you all were buying books, so thank you. (Sebastian) That's really generous of you.
(Peter) Yeah, well, it's generous of all you to buy the book, and I hope you got some good use out of it.
(Sebastian) So you can tell I'm not doing this for Peter's personal profit, but that's actually a really good book,
and I think Peter put a lot of work into the third edition. (Peter) Yeah. (Sebastian) I think more than the first two editions.
(Peter) That's right. (Sebastian) It's amazing how the later editions consume more work.
(Peter) The first edition was easiest, because Stuart Russell and I were doing it, and we just said,
"We'll put what we put into our class when we teach it, and if you don't like it, too bad," and so it was easy--
We just wrote it down. (Sebastian) And then all the professors (unintelligible).
(Peter) Yeah. And then we felt like we were doing this for the world and not for ourselves, and that made it much harder.
And we had to sort of double-check everything. (Sebastian) Yeah. But I've taught from this book many times,
and I find it amazingly well-researched. It is a big book. It's not 500 pages, it's... can you tell how many pages it is?
(Peter) 1,100, I think. (Sebastian) Oh my God, 1,100 pages. Well, it's going to keep you busy for a while.
(Peter) Yeah. (Sebastian) The other thing that's nice about it is there's lots of references in there to other texts.
I also wrote a book that isn't quite as popular as Peter's. It's more specific, on probabilistic robotics,
so it looks into robotics from Bayes network perspective.
And if, after this, you care about robotics and perception and real world systems, you can buy it.
The sales didn't go up as a result of this class. We didn't use it as text at all.
So if I teach a future robotics class, I might use that as a text.
(Peter) And to answer Evo, I think you should also look into reading journal articles.
Once you start getting into a field, then you should become competent to read the current research,
and that's really where things are happening. And if you want to keep up with where the field is going,
going to have to be able to learn to read those as well as read the textbooks.
(Sebastian) Anul Lorena from Mexico City--hi, Mexico City--says, "Dear professors, you talked about
"top-notch lists of students. Do you think it would be possible to know, after giving us the final score,
"the overall place you got among the 40 or 80,000 students? That would be encouraging for future courses."
And the answer is absolutely yes, we'll let you know your rank. We're kind of dabbling with the idea of
contacting our top-notch students to solicit--for those of you who want to do it--to send us your c.v.
We have good connections to a local search engine company and they're always looking for great people.
It'd be a pleasure, actually, to assist some of you in finding you jobs, to be honest.
I have to admit, I've been impressed by some of you, many of you, and the amazing high-quality answers
you've given to a really challenging set of exam questions. At some point we looked at the number of--
the ratio of top-notch students in this class outpaced the ratio of top-notch students at Stanford.
So it would be a pleasure. So please stay tuned, and if you are doing really well, you might receive email by me.
(Peter) And we'll be careful to respect your privacy and not spam you too much, and let you make the contact
if you want to, but we may be able to help-- (Sebastian) Absolutely. (Peter) --make that connection.
(Sebastian) This is really meant just as an assistance if you care about it; if not, just ignore us, please.
Vicki in Brazil. "Hi, I just wanted to thank you, professors, for this excellent opportunity that I had.
"I really appreciate the subject and the challenges it has brought. Please let us know of any incoming classes.
"Professor Thrun and Professor Norvig, your passion is contagious."
(Peter) Well, you're welcome again, Vicki. It's too easy. There aren't any hard questions this week.
(Sebastian) Yeah, no one says, "This course sucks." Except once you said you didn't sleep much.
(Peter) That's true. (laughter) (Sebastian) Well, I have to admit, these comments are moving me, honestly,
they make me really emotional about this class, and make me feel really close to you guys.
So here's one last question I'd like to ask, and I tried to find one that's a little bit less about the wonderful experience.
Here's one by Neco from Moscow. "I want to ask about neural nets, evolution in programming, genetic algorithms,
"fuzzy logic, fuzzy time series. The professors never mentioned these topics at all over the class. Why is this?
"And will these topics be covered next classes?" And he actually hopes that they will.
And the answer is that we only had ten weeks, and so we can't cover everything.
Like Sebastian said, my book is 1,100 pages. That was too many, and all of A.I. is even larger than that.
So we had to come up with a sampling. And in each area we chose a couple examples, and we left out other examples.
So, like why didn't we do genetic algorithms? So genetic algorithms are certainly very popular, they're a big part of A.I.
It's a commonly used tool. But the way we see it, genetic algorithms are another type of search algorithm.
And we introduced a couple search algorithms, and since it wasn't a class just about those,
we didn't want to exhaustively cover all of them. And so we covered what we could in the amount of time we had.
(Sebastian) One of the most popular tools right now are probabilities. And of course, probabilities in machine learning,
probabilities for inference, and probabilities and uncertainty. So for example, when it comes to fuzzy,
I think these days probability has become somewhat more popular than fuzzy logic, which tries to achieve similar things--
not exactly the same thing, but it's often used to represent uncertainty.
And as Peter said, I think there's many topics we had to leave out. We did want to go in-depth in some topics
and really teach you the skill of using it. We didn't just want to make an overview class.
And you can take my word for it, we also left them out at Stanford. So Stanford students have no clue about these things, either.
And these are important concepts and they're worth studying, and I think a lot of interesting stuff is happening in these fields.
And so our apologies for not dragging this class on for another eight weeks or so.
Well, I think that's our last office hour. (Peter) That's it. So, thank you all for watching and participating in the class.
Good luck on the final exam. I know you'll all do great.
You've done great so far, and it's been a pleasure working with you.
(Sebastian) So it was great having you in class. We've been blown away by the wonderful feedback we received.
I think it was one of the best ways to spend my time, and I'm so glad I could reach so many of you.
When I was a student, I couldn't get access to good education at the scale that we aspire to provide to you for free.
So thank you so much, and I hope you stay tuned for possible future classes.
(Peter) And it's been a great experiment, I've learned a lot from doing it. I hope you've learned.
And I hope the field of A.I. and the field of education and online education can take some lessons from this.
There's still a lot we have to analyze to kind of understand what went on.
We know we worked really hard and didn't sleep very much. We don't quite understand all the lessons
from what worked and what doesn't work. We're going to go back and analyze that and try to figure it out,
and then try to figure out what the next steps are in the next class.
And I hear, Peter, you're going to do a TED Talk, right? (Peter) I am. (Sebastian) That's wonderful.
So Ted is this big conference that takes place in February--technology, entertainment, design.
It's a meeting where the brightest and the best in the world report on their experiences,
and Peter's going to be the person doing it. (Peter) So by February I have to understand what just happened.
(Sebastian) Yeah, I'm happy to help you. All right. Goodbye. (Peter) Goodbye and thank you.
ギリシャのカスター・クリプティスからです
“この講義に感謝します
受講後にコースはダウンロードできますか?”
“個人的に講義を保存して今後も参照したいです
講師の皆さんありがとうございます”
前にも答えましたがYouTubeで公開しています
ですが一度にすべて閲覧できるように
まとめた方がいいですね
そうなるように検討したいと思います
そうですね ずっと公開されていてほしいです
いつでも復習できます
私たちの講義は百科事典や
参考書で勉強するのとは違います
復習したくても同じような講義を
見つけるのは大変です
これは一度限りの聴講ですね
経験を得るための通過点です
数カ月か数年後かに復習したい内容の見つけ方と
効率的なアクセス方法を追記するつもりです
今のシステムではそれが欠落しています
テキサス州ヒューストンのイーヴォからです
“教科書を読む以外のやり方に共感しています”
“ですが人工知能の分野の幅広い基礎を学ぶ際の
教材の種類について”
“アドバイスを頂けますか?
すばらしいご指導に感謝します”
他の本よりも明らかに秀でていて
市場の95%を独占しており
どの本よりも多くの言語に
翻訳されている本があります
特に第26章は最高です 私が書いていますからね
信じられないかもしれませんがこの本は
偶然にも私の隣に座る人が書いたものです
強くお勧めしますよ
そういえばこの四半期の売上は
寄附すると聞きました
ええ 慈善事業に寄付しました
出版元によると売上は好調なようです
皆さんが買ってくれたからです
ありがとう 感謝しています
買ってくれた皆さんのおかげです
本は学習に活用して頂けたらと思います
ピーターの利益のために勧めたわけでないですが
これは実際にとてもよい本です
ピーターは第3版に多大な労力を注ぐと思います
そうですね 第2版よりもです
最新版の制作にさらに力を注ぐのは
驚くべきことです
第1版はスチュワート・ラッセルと一緒に
書いていたので一番楽でした
彼に“教える際に講義で扱う内容も
盛り込もう”と言ったんです
すると多くの教授が本を使い始めたんですね
自分自身ではなく読む人のことを考えていたので
執筆はなかなか大変でした
すべてをダブルチェックしましたよ
そうですね 私は何度もこの本で教えました
驚くほどよく調べられています
かなり厚いですがページ数を教えてもらえますか?
1,100ページだと思います
すごいですね 忙しくなるのも納得です
この本のよい点は他の文献を
数多く参照していることです
人気度は落ちますが私も本を出しています
確率的なロボット工学に特化した内容です
ベイジアンネットワークの観点から
ロボットを見ています
ロボット工学、認知、実世界のシステムについて
今後も興味があるなら買うのもありだと思います
この講義は売上には貢献しませんでしたね
本を使っていませんからね
今後ロボット工学の講義で教えるなら
教科書として使うかもしれません
イーヴォへの回答ですが
雑誌の論文を読むのもお勧めします
ある分野に足を踏み入れたら現在の研究や
物事が起こっている場所が実際にどこなのかを
知る力量を持つべきです
その分野が向かう場所へついていきたいなら
教科書と同様に論文を読む必要があります
メキシコシティのアノール・ラレーナからです
“講師の皆さんは”
“優秀な学生の話をしましたが
期末試験を採点したあとで”
“4~8万人の受講生の中での順位を知れますか?
今後の勉強の励みになると思います”
答えはイエスです 皆さんに順位を
知らせるつもりです それにもし希望があれば
優秀な学生と連絡を取りたいと思いますので
希望者は履歴書を送ってください
私たちは優れた人々を求めている地元の
検索エンジンの会社とよい関係を築いています
皆さんの就職活動を手伝えるのは
正直に言ってかなり楽しみです
私は試験の難問に対して驚くほど
質の高い回答を出す皆さんに
本当に感銘を受けてきました
いくつかの点においても
スタンフォードよりもこの講義の方が
優秀な学生の割合が高かったと思います
本当にうれしいです
私からメールを受け取るかもしれませんね
皆さんのプライバシーを尊重して
余計なコンタクトは控えますが
皆さんが望むなら連絡を取らせてください
関係を築く助けになると思います
本当に関心がある人への単なる手助けのつもりです
そうでないなら聞き流してください
ブラジルのヴィッキーからです
“このすばらしい機会に感謝します”
“テーマとそれに伴う設問はとてもありがたいです
今後の講義について教えてください”
“スラン教授とノーヴィグ教授の
情熱に感化されました”
お役に立ててうれしいです
今週は簡単ですね 難しい質問がありません
“睡眠不足でしょう?”と言われたことを除けば
確かに誰も“このコースは最悪”と言っていません
そうですね
皆さんからのコメントに感動しています
心を動かされますし
皆さんにも親しみを覚えます
では最後の質問です
これは今までとは少し違った内容です
モスクワの受講生からです
“ニューラルネットワーク、
進化的プログラミング、遺伝的アルゴリズム、”
“ファジィ時系列についてです 講師の皆さんは
なぜ講義でこれらに言及しなかったのですか?”
“これらは次の講義で扱いますか?”
彼は実際にそうなることを望んでいます
講義は10週しかなかったので
すべてを扱えなかったのです
私の本は1,100ページありますが
それでも人工知能の分野の一部に過ぎません
その中でもサンプリングを行う必要があり
各項目で例を何個かに絞ったのです
確かに遺伝的アルゴリズムは多用されており
人工知能の大部分を占めています
一般にも使われるツールですが私たちの見解では
遺伝的アルゴリズムは探索アルゴリズムの一種です
いくつかの探索アルゴリズムを紹介しましたが
講義はこれらについてだけではなかったので
すべてを扱わなかったのです 限られた時間の中で
カバーできる部分だけを扱いました
現時点での最も一般的なツールの1つは確率です
機械学習の確率、
推論の確率、不確実性の確率などです
ファジィ論理と比べて
近年では同じような効果のある
確率の方が一般的になっています
ですがファジィ論理は不確実性を
表現するためにしばしば使われます
確かに扱えなかったテーマも数多くあります
深く掘り下げたかったものもありましたし
皆さんがそれを使う技術も教えたかったです
浅く学ぶような講義にはしたくありませんでした
スタンフォードでも同じ部分を割愛したので
学んでいない部分は一緒です
この分野には学ぶ価値のある重要な点が数多くあり
たくさんの興味深いことが起こっています
この講義を更に8週ほど延ばせないことを
申し訳なく思っています
これは最後のオフィスアワーだと思います
この講義に参加して頂いて感謝します
期末試験での幸運を祈ります
皆さんなら大丈夫です
皆さんはとても優秀だったので
一緒に学べて光栄です
講義で皆さんと会えてよかったです
見事なフィードバックに驚かされてきました
最良の時間の使い方だったと思いますし
こんなにたくさんの方とお会いできて嬉しいです
私が学生だった時は無償でシェアしたいと
思うほどのよい講義に出会いませんでした
本当にありがとう
今後の講義を楽しみにしていてください
すばらしい経験でしたし私も多くを学びました
皆さんもそうだったことを願います
この講義が人工知能の分野、教育の分野、
オンライン教育に役立つことを願います
講義の内容についてより理解するために
分析すべきことがまだたくさんあります
睡眠時間を削ってこの講義に力を注ぎました
すべての講義を通して成功した部分と
失敗した部分をまだ把握できていません
分析して明確にするつもりです
そして次の講義へのステップを
見つけるつもりです
ピーターはTED Talksで講演するんですよね?
その通りです
TEDは大規模な講演会を2月に催します
Technology Entertainment Designの略です
世界でも輝かしい活躍をする優秀な人々が
その業績について報告する会議に参加するんですね
当日までに講義の結果を
まとめないといけません
私も手伝いますよ
それではさようなら ありがとう