Michael Bernstein describes crowdsourcing platforms that harness global workers for tasks from document editing to writing symphonies, but warns of quality control and malicious actors. Álvaro Soto explains how machine learning lets robots perceive and navigate real-world environments autonomously, moving beyond factory floors toward household tasks like cooking.
Chapters
Hey, Michael Bernstein from Stanford, welcome.
Thank you.
I know that you are investigating a field that is fairly new to computer science, this whole notion of crowdsourcing. You've heard the term, but until recently, crowdsourcing is just sort of unorganized, you know, it's just sort of an open call, and that's changing.
Yeah, so we're thinking a lot about ways in the field to help crowds organize more effectively, get things done that they might not be able to do natively if they— we just sort of gathered a bunch of people together and said, try to do this. So we're trying to think about all sorts of things ranging from writing stories, writing papers, editing things, writing symphonies. What could we help crowds come together to do, and what kind of infrastructure can we build to help them get it done?
So give us an example of how that works.
Sure, so I think You're familiar with many positive examples of crowdsourcing. Things like Wikipedia started there. And in the research field, what a lot of us are starting to think about is taking advantage of these platforms that exist in order to pull people together, for example, to do things like to take text and to identify— like have a copy editor. So you take your document that you've written, ask the crowd for help identifying typos, weird language, or maybe you want this part shorter. Or maybe you want a better photograph out of your camera. Or maybe you're a blind user and you want to ask visual questions like, Which way— is there a bench across the way from where I'm standing? All of these things are reaching out to crowds in order to help end users.
Right. And if you want to build the next Wikipedia or something, maybe there was an interface for that.
Yeah, so a lot of what I think we need to be able to do is to make anyone be able to access the intelligence of the crowd, to get perspectives from people who you might not have been able to reach out to before, see people from different countries, different cultural backgrounds, and get their feedback on your thoughts, maybe inspire new ideas for you.
How did you get into this? Because I know this is fairly new. You didn't, like, go to college thinking, this is what I'm going to do, crowdsourcing.
Sure. So, you know, in fact, there's a very interdisciplinary field that's growing around crowdsourcing. For myself, I came out of a sub-area called human-computer interaction, where we think about how to design computational systems. So how to design user interfaces and the entire interactive worldview around what you're using technology for. And then I started getting interested in how to do this at large scale. How do we design for tons of people at once? And as this started happening, you know, you see lots of successes around platforms like Amazon Mechanical Turk or oDesk and many other ones that really are making it possible now for the first time to actually help direct crowds. And so we just launched straight into it.
Is the sum always greater than, you know, is the whole greater than the sum of the parts, or are there times when crowdsourcing, you know, maybe doesn't lead to the results that people expect? Yes.
So, you know, you have to be careful. I asked 1,000 people online to flip a coin, and about 2/3 of them said heads. So, you know, you have to actually be careful and think deeply about how you're designing what you might call an algorithm that's running on top of the crowds to help guide them as they work so that you get honest results, real hard work, people actually contributing perspectives that they really believe in, everything like that. So we— there's a lot to be done there.
And you don't want somebody hijacking the process either.
No, this is very possible. Yes.
Well, very interesting research. I sure appreciate you coming and talking with us about it. Thank you, Michael. My pleasure.
Thank you.
Also, another guy who's right around the wings here— they keep coming— is Álvaro Soto from Chile. Welcome.
Hi.
Hi. I know you've been to the Faculty Summit before, but you are here because your topic is highly talked about. This term machine learning, you've heard it all day long. Machine learning, what does that mean to laypeople and how are you using it?
Well, the machine learning is just to create tools such that the computer can analyze data and then extract patterns and knowledge so they can understand, you know, what's going on from that data.
You're teaching computers, basically.
That's basically, yes.
And you're also building robots, I understand.
Oh, yes, yes. I'm mostly working with the cognitive part of robots, so it's related to machine learning. It's also how a robot can learn from the environment to recognize things or navigate autonomously through the environments, for example.
You're building cognitive robots?
No, I mean, I'm mostly interested in the part of the cognitive part of robotics. You, in robotics, you can analyze, like, the motion or different applications. But I'm mostly working with the cognitive part of the robot. So how the robot can understand the environment so it can take decisions and act autonomously.
So for example, you're teaching the robot to navigate and travel?
Yeah, yeah, for example.
Yep. And you're also saying that instead of like industrial robots where you can change the environment so that the robot can adjust, you're building robots where you can't adjust the environment.
Oh, that's right. Yes. And the idea is just to make robots work in natural environments where The robot can adapt to what is going on in the real world, not to adapt the world to the robot, but the robot adapts to the world by understanding what is going on, the semantic understanding of the real world.
That would be nice. And so how do you see that eventually being applied in the real world?
Well, it has many applications, you know. I mean, maybe you can have a robot in your house and then you can— the robot can make some tasks for you and make your life easier.
Oh, that would be nice. Can I have a robot cook me dinner?
Oh yeah, yeah, probably, right, in the short future.
Well, we already have robots that vacuum and robots that do—
Exactly, that's right. There's some robots, I mean, and also for that robot, you still need some adaptation of the environment. But, you know, as the technology progresses, you can find a robot that can do this in a more natural way without any adaptation or with any simplification of the real world.
This is just like The Jetsons. You're building Rosie the robot who can, like, go around and take care of everyone.
Yeah, I hope, I hope so someday we can have that, you know?
I would like the first one, please. Thank you, Álvaro.
