Return to Video

Office Hours 9 02mp4

  • 0:00 - 0:06
    (Peter) So here's Michael Grotik in Barcelona, wants to know, "What kind of computer processing do you need
  • 0:06 - 0:10
    on an autonomous vehicle? Do you need fancy CPUs and a lot of RAM?"
  • 0:10 - 0:17
    (Sebastian) Well, you need a lot of processing. We tend to have usually a quad core computer or two on the device.
  • 0:17 - 0:26
    Most of the processing is actually sensor data, so camera data tends to occupy at least one or two cores of a CPU.
  • 0:26 - 0:32
    And then exadata, in our case, comes a million points a second, occupies another core.
  • 0:32 - 0:38
    And then it goes down into the processing needs. So our current (unintelligible) cars can safely live
  • 0:38 - 0:42
    on a single quad core computer and drive just fine.
  • 0:42 - 0:49
    It's actually not an exercise in computing. Evaluating a Google instant query takes, I don't know, I'm guessing,
  • 0:49 - 0:54
    10,000 times as many computers, for a short amount of time, and they're really fast.
  • 0:54 - 0:59
    It's much more a question of algorithms. To make sure that the computer software actually makes the right decisions
  • 0:59 - 1:06
    is much harder. So if someone gave us a supercomputer, we actually wouldn't know what to do with it on a car.
  • 1:06 - 1:11
    So here's by Pravin in Chennai, Indiana--actually, India.
  • 1:12 - 1:16
    "This is definitely one of the best classes I've ever had. Thank you, instructors. It would be really great
  • 1:16 - 1:22
    "if you considered doing a class in advanced A.I. What are the research options that can be formulated
  • 1:22 - 1:26
    "by some of the ideas described in this class?" Peter?
  • 1:26 - 1:31
    (Peter) Yeah, so they keep on asking for the advanced A.I. class. I guess we'll have to get around to that.
  • 1:31 - 1:33
    (Sebastian) Do you think there's a message in there, "Let's do it?" (Peter) Yeah, we should do it.
  • 1:33 - 1:38
    (Sebastian) You know what? I like sleep, I like my family. (Peter) We're going to sleep for a little while,
  • 1:38 - 1:43
    but then we'll think about getting back to it. (Sebastian) And I love all of you who are seeing this online.
  • 1:43 - 1:47
    You guys are my heroes. I've received some unbelievable emails that really broke my heart.
  • 1:47 - 1:53
    I received recently an email from a mother of three children who was raising their kids and doing their midterm exam
  • 1:54 - 2:00
    with one hand while having a teething child in her other hand, and it was just completely mind-blowing,
  • 2:00 - 2:06
    what effect this class has had, and how many of you feel so positive about it. I'm really humbled by it.
  • 2:06 - 2:09
    I just want to express this.
  • 2:09 - 2:15
    The second part of the question was--let's see, let me scroll up again--
  • 2:15 - 2:20
    "What are the research options that can be formulated from the ideas described in this class?"
  • 2:20 - 2:27
    I think we covered, basically we did a short coverage of all of A.I. That's not quite true. There's areas of A.I. that are missing.
  • 2:27 - 2:33
    Like genetic algorithms and ontologies and knowledge-based systems. And we probably brought to air
  • 2:33 - 2:38
    some of our personal biases in the direction of the material. But if you go to a modern-day A.I. conference,
  • 2:38 - 2:44
    like, there's a couple (unintelligible) I'm sharing in 2013, in China, you would find that this is kind of basically
  • 2:44 - 2:51
    what's happening in the field. And all of those we touched just superficially. So if you read Peter's amazing book,
  • 2:51 - 2:56
    you'd find in the book alone you get twice, three times as deep a treatment of all of those.
  • 2:56 - 3:01
    The key thing for researchers, in my opinion, just pick a problem, something that bothers you.
  • 3:01 - 3:05
    Like if you're walking to your house and you're bothered by the fact that you have to flip on the light switch,
  • 3:05 - 3:10
    just solve it, okay? If you open your fridge and you have like old produce and you keep forgetting that
  • 3:10 - 3:15
    the salad only lasts for a week, and you want to make it so that in the future, that your produce doesn't get old,
  • 3:15 - 3:21
    just solve it. And if you solve this problem through A.I. and you can leverage some of the skills you learned in this class,
  • 3:21 - 3:27
    and it'll be amazing. (Peter) Yeah. So I agree with that. We talked about that in a lot of office hours,
  • 3:27 - 3:32
    of just picking out a problem and doing it. I think another way to look at it, though, is you can look at it from the side of,
  • 3:32 - 3:39
    "What do I want?" or you can look at it from the side of, "What do I have?" and that if you have a lot of data available,
  • 3:39 - 3:45
    if you have an autonomous car, then do something with that. If you have a copy of the web, then do something
  • 3:45 - 3:50
    with that data. So you can approach it in either direction, or maybe have the two meet somewhere in the middle.
  • 3:50 - 3:55
    (Sebastian) And there are so many other options. If you have a smartphone, you can make it learn how fast you walked,
  • 3:55 - 4:01
    how far you walked. The problem of finding out what the angular orientation is of a smartphone is an open problem,
  • 4:01 - 4:07
    because compasses aren't very reliable. Little small things sometimes are just really amazing to do A.I. about.
  • 4:07 - 4:11
    Here's a question from Washington, D.C., by Terry, who's a Stanford alum.
  • 4:11 - 4:17
    "Thank you for this great experiment." Of course, I love reading these questions. They're all about how great the class was.
  • 4:17 - 4:24
    "It's been long since my days at Stanford. I'm so proud that my alma mater is doing such pioneering work.
  • 4:24 - 4:29
    "And makes sure that it should catch up with other institutions." So here's the real question, Peter:
  • 4:29 - 4:34
    "How does this online teaching experience compare to your regular teaching experience?"
  • 4:34 - 4:42
    (Peter) Well, like most things in life that are worth doing, it ends up being a lot harder than you ever thought it was going to be.
  • 4:42 - 4:46
    I thought that doing the online work would be a little bit harder because you've got to prepare the videos,
  • 4:46 - 4:51
    and if you stumble in the middle of a video, you've got to rewind and do it over again.
  • 4:51 - 5:00
    What I didn't realize is how hard it is to anticipate what you all are thinking, and try to put that into the video in the right way.
  • 5:00 - 5:05
    When you're in class and I'm talking and things aren't going right, I can see the puzzled faces,
  • 5:05 - 5:11
    and then I can back up and try something new. But with the videos, I don't have that immediate feedback.
  • 5:11 - 5:17
    So I have to anticipate what could possibly go wrong, and try to get that all right the first time.
  • 5:17 - 5:24
    And so it really leads to a little bit different material in terms of what you teach, and a very different approach
  • 5:24 - 5:30
    in terms of the quiz questions that we came up with. And I think we were really breaking ground as we were going,
  • 5:30 - 5:35
    trying to think of what are good mechanisms for coming up with good quiz questions?
  • 5:35 - 5:39
    And I hope we did okay, and I know we can learn more and do a better job with it next time.
  • 5:40 - 5:47
    (Sebastian) Yeah, I learned a good number of things. I learned that, I think the digital medium is a much more amenable tool
  • 5:47 - 5:57
    to interaction than I thought. I got a lot of feedback that even though technically--I'm not really talking to you right now, personally--
  • 5:57 - 6:03
    But we still got a lot of intimacy in the setting, by asking questions, and we got a lot of engagement
  • 6:03 - 6:08
    by students solving problems. It was much better than, I think, the Stanford classroom,
  • 6:08 - 6:13
    where I'm basically just lecturing students and they all get the skill of listening, but they don't get the skill of problem-solving.
  • 6:13 - 6:20
    In fact, all my Stanford students have extensively watched our online videos, and they all say their experience
  • 6:20 - 6:25
    is substantially better as a result. So I'm getting to the point of asking, "Why am I lecturing?"
  • 6:26 - 6:29
    And I might just not lecture again in the same class at Stanford.
  • 6:29 - 6:38
    The second thing I've learned is that while I'm kind of primed as a teacher to make really hard questions,
  • 6:38 - 6:44
    and at Stanford we do this a lot, and it's frustrating for the students, I've learned this is much more about
  • 6:44 - 6:49
    empowering the students, you guys, than showing what a smart question I can ask.
  • 6:49 - 6:55
    So I've really focused on trying to get material together to make this experience as positive for all of you
  • 6:55 - 6:59
    as I possibly could. And I'm not sure I succeeded, but I worked really hard.
  • 7:00 - 7:05
    The final thing I've learned is I get about maybe one invitation per day to give a keynote at a conference,
  • 7:05 - 7:10
    and I've said no to every single one of them since the class started. And the reason is,
  • 7:10 - 7:17
    I can fly to India, I can fly to China, I can talk to maybe three, four (unintelligible) 2,000 people.
  • 7:17 - 7:26
    But talking to 30,000 people is such an amazing (feast). I feel my personal time is much better spent making this class
  • 7:26 - 7:33
    than pretty much anything I do right now. Possibly with the exception of changing a lot of transportation.
  • 7:33 - 7:37
    It's been really amazing. I mean to some extent, we're sitting in front of a video camera right now,
  • 7:37 - 7:42
    and we can't really see all of you. But then we're getting so many beautiful emails and so many beautiful questions
  • 7:42 - 7:46
    on this forum that I can actually see everyone, and I'm going to try my best to respond to emails.
  • 7:46 - 7:52
    I haven't responded to every single one of them. But I've often spent an hour or two a day just responding to email.
  • 7:52 - 8:01
    And we try to respond to you through these questions. It's been really gratifying what kind of scale we can achieve with this new medium.
  • 8:01 - 8:08
    (Peter) All right. Here's one from Andy in New York. "I made a bot with (Dred), Sonar and Bluetooth for $200
  • 8:08 - 8:15
    "to do some mapping experiments, and Bluetooth allows processing on a P.C. Did not have laser rangefinders.
  • 8:15 - 8:18
    "Can you suggest other inexpensive platforms for experimenting?"
  • 8:18 - 8:27
    (Sebastian) Yeah, there's a company called Neato that does a vacuum cleaning robot, and they built
  • 8:27 - 8:35
    an insanely cheap laser rangefinder. Rumors have it that the laser rangefinder is made for about $10 apiece.
  • 8:35 - 8:41
    If you build a robot for $200, that's wonderful. I would love to see more insanely low-cost robots in the market
  • 8:41 - 8:46
    that people can program and play with. There's Lego Mindstorm, there's a few other kits that people can use,
  • 8:46 - 8:51
    and they generally tend to be out of people's price range. I think if you had platforms--
  • 8:51 - 8:56
    In fact, if you had a platform that people could buy for very little money, I'd be much more enticed to teach a robotics class
  • 8:56 - 9:01
    because I think (unintelligible) could be a really essential experience in robotics.
  • 9:01 - 9:06
    So (unintelligible) question. Monica in San Francisco, California--hi, Monica, hi, San Francisco--
  • 9:06 - 9:12
    says, "I've decided to take this class because I studied A.I. in college and used Peter Norvig's book at the time.
  • 9:12 - 9:17
    "I've been beyond impressed with the enthusiasm and passion of the professors, and the quality of the content.
  • 9:17 - 9:22
    "Thank you." Not a question. (Peter) You're welcome. That's the answer.
  • 9:22 - 9:29
    (Sebastian) Garachi in Budapest. "This was definitely one of my best classes ever. I really enjoyed your big picture approach.
  • 9:29 - 9:34
    "Now it is high time to get my hands dirty programming some of the concepts I've learned.
  • 9:34 - 9:40
    "Well, after the exam period, that is, and of course, past the holidays, Christmas. Thanks for the awesome experience."
  • 9:40 - 9:45
    (Peter) You're welcome, too. (Sebastian) Yeah, it's great. Do get your hands dirty. We really regret we didn't
  • 9:45 - 9:48
    have a programming component, and it's going to happen in the future. (Peter) Yeah.
Title:
Office Hours 9 02mp4
Video Language:
English
Team:
Udacity
Project:
CS271 - Intro to Artificial Intelligence
Duration:
09:48
Udacity Robot edited English subtitles for Office Hours 9 02mp4
sp1 added a translation

English subtitles

Revisions Compare revisions