![\"Besmira](\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2019\/12\/Research_Podcast_Besmira-Nushi_Site_10_2019_1400x788-1024x576.png\")
<\/p>\n<\/p>\n
With all the buzz surrounding AI, it can be tempting to envision it as a stand-alone entity that optimizes for accuracy and displaces human capabilities. But Dr. Besmira Nushi<\/a>, a senior researcher in the Adaptive Systems and Interaction group<\/a> at Microsoft Research, envisions AI as a cooperative entity that enhances human capabilities and optimizes for team performance.<\/p>\n On today\u2019s podcast, Dr. Nushi talks about what it takes to develop collaborative AI systems and unpacks the unique challenges machine learning engineers face in their version of the software development cycle. She also reveals why understanding the \u201cterrain of failure\u201d can help researchers develop AI systems that perform as well in the real world as they do in the lab.<\/p>\n Besmira Nushi: <\/span>W<\/span>hat I\u2019d like AI to be, I<\/span>\u2019d<\/span> like it to be a technology that enables everyone, and that is built for us. It\u2019s built for people. My parents should be able to use it, an environmental scientist should be able to use it and make new discoveries, or a policy maker in order to take good decisions.<\/span> <\/span><\/p>\n Host: <\/span><\/b>You\u2019re listening to the Microsoft Research Podcast, a show that brings you closer to the cutting-edge of technology research and the scientists behind it. I\u2019m your host, Gretchen Huizinga.<\/span><\/b> <\/span><\/p>\n Host: <\/span><\/b>With all the buzz surrounding AI, it can be tempting to envision it as a stand-alone entity that optimizes for accuracy and displaces human capabilities. But Dr. Besmira Nushi, a senior researcher in the Adaptive Systems and Interaction group at Microsoft Research, envisions AI as a cooperative entity that enhances human capabilities and optimizes for <\/span><\/b>team performance.<\/span><\/b> <\/span><\/p>\n On today\u2019s podcast, Dr. Nushi talks about what it takes to develop collaborative AI systems<\/span><\/b>,<\/span><\/b> and unpacks the unique challenges machine learning engineers face in their version of the software development cycle. She also reveals why understanding the \u201cterrain of failure\u201d can help researchers develop AI systems that perform as well in the real world as they do in the lab.<\/span><\/b> <\/span><\/b>That and much more on this episode of the Microsoft Research Podcast.<\/span><\/b> <\/span><\/p>\n Host: Besmira Nushi, welcome to the podcast!<\/span><\/b> <\/span><\/p>\n Besmira Nushi: Thank you. It\u2019s great to be here. I\u2019ve been following the podcast in the last year, and<\/span>,<\/span> you know<\/span>,<\/span> it\u2019s always interesting. Every new episode is different.<\/span> <\/span><\/p>\n Host: You\u2019<\/span><\/b>ve been<\/span><\/b> following the podcast<\/span><\/b>!<\/span><\/b> That\u2019s nice!<\/span><\/b> <\/span><\/p>\n Besmira<\/span> <\/span>Nushi<\/span>: <\/span>I have!<\/span> <\/span><\/p>\n Host: Well, I\u2019ve talked to you before. Last time<\/span><\/b>,<\/span><\/b> you were on a Research in Focus panel at Faculty Summit in 2017 and you talked about ML troubleshooting in real<\/span><\/b>–<\/span><\/b>time systems. <\/span><\/b>L<\/span><\/b>et\u2019s go there again. As a senior researcher in the Adaptive Systems and interaction group, you work at what you call the intersection of human and machine intelligence.<\/span><\/b> <\/span><\/p>\n Besmira<\/span> <\/span>Nushi<\/span>: <\/span>Y<\/span>u<\/span>p, y<\/span>u<\/span>p.<\/span> <\/span><\/p>\n Host: Which I love. So we\u2019ll get to your specific work in a minute, but in broad strokes, what\u2019s going on at that intersection? What gets you up in the morning?<\/span><\/b> <\/span><\/p>\n Besmira Nushi: Well, the intersection is a rich field and it really goes both ways. It goes into the direction of how can we build systems that learn from human feedback and input and intervention, and maybe learn from the way people solve problems and understand the world<\/span>?<\/span> And it also goes in the other direction, in like<\/span>,<\/span> how can we augment the human capabilities by using artificial intelligence systems? How can we make them more productive at work? And putting the best of both worlds together.<\/span> <\/span><\/p>\n Host: Let\u2019s talk a little bit more about this human-AI collaboration. You<\/span><\/b>\u2019ve<\/span><\/b> framed it in terms of complementarity<\/span><\/b>\u2026<\/span><\/b> <\/span><\/p>\n Besmira<\/span> <\/span>Nushi<\/span>: <\/span>Mmm<\/span>-h<\/span>m<\/span>m.<\/span> <\/span><\/p>\n Host: <\/span><\/b>\u2026b<\/span><\/b>ecause humans and machines have different strengths and weaknesses. And you\u2019ve also characterized it as putting humans and machines together to<\/span><\/b> quote-unquote<\/span><\/b> \u201coptimize for team performance.\u201d<\/span><\/b> <\/span><\/p>\n Besmira Nushi: Yes!<\/span> <\/span><\/p>\n Host: So, e<\/span><\/b>laborate on that for us. How should we understand AI as collaborator versus AI designed to work on its own?<\/span><\/b> <\/span><\/p>\n Besmira Nushi: You know, people and algorithms, they have very different skills. We\u2019re really good in reasoning and imagination. And machines are good in processing these terabytes of data for us and giving us these patterns. However, you know, if we can use the machine capabilities in an efficient way, we can be quicker and faste<\/span>r,<\/span> as I said. But then<\/span>,<\/span> on the other hand, you know, these are concepts that, if you think deep about it, they are not that new. In the sense that when we invented personal computing in the 80s, this is one of the reasons why it became so successful<\/span>,<\/span> because the personal computer was suddenly this \u201cbuddy\u201d that could help you do things faster and quicker. But then there is another thing that enabled that development in those years and really, I think that that is the field of human computer interaction.<\/span> <\/span><\/p>\n Host: Yeah.<\/span><\/b> <\/span><\/p>\n Besmira Nushi<\/span>: <\/span>What HCI did in those years, is that it made the interface understandable from a human perspective and it really made the computation technology accessible for everybody. So now we see billions of people around the world that use some form of computation without making any significant effort.<\/span> <\/span><\/p>\n Host: Right.<\/span><\/b> <\/span><\/p>\n Besmira Nushi<\/span>: <\/span>And I think that today, we are in front of such forms of developments in artificial intelligence. We are<\/span>,<\/span> in <\/span>a <\/span>way<\/span>,<\/span> in <\/span>the<\/span> position that we can innovate in the way how people <\/span>interact with AI technologies, but we still need to make that leap and make AI accessible for users.<\/span> <\/span><\/p>\n Host: Right.<\/span><\/b> <\/span><\/p>\n Besmira Nushi: And this is what I mean by the fact that<\/span>,<\/span> so far, we have been optimizing AI for performance only<\/span>,<\/span> and performance when the AI is designed to play alone in the field.<\/span> <\/span><\/p>\n Host: Yeah.<\/span><\/b> <\/span><\/p>\n Besmira Nushi: But if it has to play together with a human, there are other scores that we need to think about. For example, one of them is interpretability, in that people should be able to understand how a machine <\/span>makes <\/span>a prediction. Another one that we focus a lot on is predictability of errors. And what this really means is that, if I\u2019m working with an AI algorithm, I should be able to<\/span>,<\/span> kind of<\/span>,<\/span> understand that that AI algorithm is going to make mistakes. And this is important for me as a user because, if I have the agency to make the final decision at the end, I need to know when it\u2019s right or wrong\u2026<\/span> <\/span><\/p>\n Host: Right.<\/span><\/b> <\/span><\/p>\n Besmira Nushi: \u2026so that I can correct it at the right time, as it goes.<\/span> <\/span><\/p>\n Host: Let\u2019s drill in on the topic of AI as a collaborator then.<\/span><\/b> <\/span><\/p>\n Besmira Nushi: Okay.<\/span> <\/span><\/p>\n Host: We\u2019ve talked a little bit about AI working alone and it\u2019s designed to optimize for performance and speed. How do <\/span><\/b>you <\/span><\/b>then go about training ML models with collaborative properties in mind instead of optimizing for speed and performance? What are tradeoffs in the algorithmic design and how do you go about enforcing them?<\/span><\/b> <\/span><\/p>\n Besmira Nushi: Right, right. Yeah, so, you\u2019re right that it always a tradeoff. <\/span>It is a tradeoff for the machine learning developer to decide which model to deploy. Should I deploy a model that is fully accurate by its own or a model that optimizes team performance<\/span>?<\/span> But the difficulty is that this tradeoff is not always as visible or as easy to access. So I see the work that we do in our group as an enabling paradigm that allows you to explore this tradeoff or<\/span>,<\/span>Related:<\/h3>\n
\n
\nTranscript<\/h3>\n