Experiences Working in AI - Irina Kofman

Thank you for having me today and thank you for the really warm. Welcome. So as mentioned I work at Facebook, director there and in particular I work in the Facebook AI lab which is really exciting. But before I dive into what do we do at Facebook that involves a I wanted to share a little background about me and how I got so lucky. I consider it lucky. I'm not perfect. Thank you. I'm not an engineer. I'm not a researcher and still I get to work in AI which is pretty cool and I got to be involved in some really cool projects. I'll share with you. So I emigrated from the

Ukraine as a refugee at five years old and that's really important to me and it's part of who I am. It's also part of how I'd approach to project fi work on and making sure things I work on have a global impact and think about users around the world and although I came from the Ukraine or Leon. I was exposed to engineering actually back there. And when I came by my father who was an electrical engineer and a professor has been exposed so early I swore I would not be intact. I want to be a lawyer a doctor anyting but in time and here we go on my Facebook and before that. I have to

spend a lot of time at Google actually almost 12 years there. So who remembers Radio Shack remember Radio Shack? So I spent a lot of my childhood at RadioShack given we adjust immigrated here. My parents didn't have a ton of money to buy me all these toys and gadgets. But my dad thought it'd be great to spend every weekend trip to RadioShack to buy parts build little circuit boards with LED lights and wires and test the current and switches expose his daughter to engineering just like him what we learn though is that I enjoyed way more planning the trip. What was the

budget how many parts are we going to buy and then on the trip managing with the store and the customer service agents what I wanted if they could order extra parts for me. I was five years old at this time. And then when we get home with the products, I would tell each family member with their role was to build and then I will track their building and give them timeline. So I'm a program manager. It was a given that that's what I would be. But how did that get meet a guy? That's really cool that the last speaker talked about product management program management. So you got a little bit

of glimpse of the value hopefully of those are parts of the team. Superfly jump into a i and how it got there. I want to show you Through The Eyes of a little kid. This is my son Logan. He's about five in this picture and we used to take him to bring your kids to work day at school and my husband worked at the time as well. And on these trips, we'd want to make sure that he didn't think mommy and daddy were going to work to jump and bounce houses eat cookies and build Legos, which is mostly what happens at Bring Your Kids To Work Day. So on the way there I ask my kid. What is Daddy do

for a living just to prepare him in case he met one of our managers and he said Daddy bills Google. It's spot-on. My husband works on Doodles at that time. Then I turn to what does mommy do and remember this is when we're setting up the Google AI team the first year, we don't really know what's going to happen. All I knew about. My job is that it was get shit done. That was basically it with the chief of staff of the organization my kid paused pondered for a minute and came back with Mommy must be daddy's boss. That was it all that might be true

at home. That was definitely not the case at work. But the evolution of AI that point not many people really believed in it have been around for a long time. It was kind of this current wave that was still in. And from there the next year when we took him I asked him again. What does Mommy do it has developed into robot. She must be working with robots with getting somewhat closer. And then I think a few months later didn't even wait a year. I hear him on his headphones playing video game and yelling at the AI completely screaming and telling me the AI gets to train the AI gets

to train a video game a way longer than we let him play. So obviously it's going to be better than him. So the AI world. The world got a custom today. I started my kid. So what was my actual role at this point? I kind of figured out what do I do and a lot of it centered around people operations and figuring out how to establish a space for our Engineers to be able to innovate and support open research. It sounds a little crazy. But even at a big company, we were kind of operating like a start-up I was moving from the search organization, which was so established to starting a small

teen. I think we started around 8 people to setup a Google AI team at that point. I also worked on this table showing the AI Residency program to bring people in and enable those newer 2 ml to get involved start being mentored by researchers. I also wanted to see if I can manage to technical project like I used to know on those shopping trips to RadioShack and one of the first projects they manage is a teepee. I'm a technical program manager was the release of tensorflow machine learning framework and I continue to do that in my current role. I also helped establish the MLS are in

a steam at the company thinking through as we release all this great AI technology. Are we doing in the responsible way including publishing the principles and responsible machine learning development and then last but not least in all of this. I wanted to make a positive social impact. I remember the day I decided to move from search to Ai and we thought of the dining room tables my parents you thought I was crazy and if you were here earlier this morning, I heard one of the speakers mention how being a little bit crazy. Discomfort is actually should build courage and that's exactly

what it said and someone that's a point tell me if you're a little scared. It's probably worth doing so I jumped into a i So how does that apply to my role at Facebook now? AI is actually quite a Facebook at a family of apps it's used throughout them and we've deployed it at scale. It's an important technology at our company and we continue to think about it as we build product we work with the product team throughout the company. Here's one example, so your newsfeed and ask customization is actually using Ai and we try to provide an experienced 2 / 2.7 billion users on a monthly

basis that's relevant to them. We also try to filter out any ads and content that might not be relevant. But I have to tell you I'm wearing rothys. I don't know how many of you are familiar with them there vegan shoes made out of plastic water bottles. So my husband is begin love this and let me spend way more than reasonable on a pair of shoes or amount of shoes. And also they're super comfortable from running around huge campuses such as Facebook and I would have never known about this had it not been for an ad on my Facebook feed other things Warby Parker glasses, but I'm wearing it

the ads ranking algorithms knew that I was interested in quality glasses that were fashionable and affordable. So earlier I mentioned that I immigrated to the US so I have family all around the world. How do you communicate with them and with natural language understanding your own machine translation were able to build communities across languages and currently this technology Powers more than six billion translations a day. Another really important part and I talked about having a positive social impact globally as well as our users and making our products accessible.

Here's an example of computer vision understanding images and we laying what's the number for the visually impaired? so how many of you believe recommendations in restaurants from a total stranger versus your friend who you know is a foodie by using natural language understanding again and systems they can show post that are relevant to you from people, you know and trust The next one so I talked a little bit about AI for social good and thinking about blood donation matching this another understanding the language and don't putting

donors together with those meeting that we actually signed up over 50 million people here. I'm going to turn into a completely different space Integrity AI tools for Content understanding also helps to proactively identify and remove an appropriate policy violating content that includes share spam pornography and bullying. For example, these tools help us block and remove approximately 1 million accounts a day. I don't know how many of you have seen this how many of you know about the portal video conferencing units that Facebook produces some of you so it's

a device that has a smart camera in it and use the computer vision models running locally on the device. So it's not sending it anywhere your videos anyway to be changing and you can see it following this person on the screen and what's interesting here is I have my father-in-law has mobile. He's kind of Moberly challenge at this point as a disability and when my son communicates with him across the country, it's really hard when you have a kid fidgeting a little bit moving out of the picture and my father-in-law can't move the camera. It's very hard. Now, they can talk almost like they're in

the same room. So it's enabled us to connect right to the mission of what Facebook does allowing us to connect with others. Unless example I fear of technology is Oculus go experiencing interactive VR with your own hands without having to hold controllers or wear gloves and you can imagine how much coke red I get with my kid. Now when he thinks this is what I work on. It's a big a lots of progress from being Daddy's boss. So I talked a lot about what we do with AI but how do we actually do that behind all the work that I've mentioned a huge collaborations that happened both

internally and with the external Community including with Academia? Just some examples so fast MRI for anyone that had an MRI before it could be a pretty uncomfortable situation. It takes time. You're in a confined space. You have to keep really still and that's just hard. So we collaborated with the NYU school of medicine to create an investigative. How do you make MRI scans 10 times faster? But what's really interesting about this project is everything we gathered we made publicly available through a data set. So others can continue this research and move it forward. We see a lot of value

of working with Academia and the broader community. Another example is a defect detection challenge that we launched late last year. This was a challenge that brought the community together from multiple industry Partners academic Civil Society organizations and Media Partners, like BBC and New York Times identify manipulated media and particularly deepfake together. We released the largest data set of its kind against allowed researchers in the broader Community to help us research and come together for a problem. We're all seeing. We also collaborate with

in the company. It's not just with external Partners. I talked a little bit about engineering. We have a fundamental AI research team and we haven't applied research team that work together and partnership are supported by design teams product management and analytics and it's really a group effort to get our projects done. Announce we decide what we're going to do data really informed those decision. We bring data into the ideation process use it to power opportunity sizing and enhance it be learning plan. We set up the right metrics and counter metrics to ensure our products are

cheating the outcome we desired and we try to understand the growth that drive and setting the right targets to ensure that we reach the maximum skeleton and finally and I think really exciting to me is knowing when it's time to imagine the future and understand the behavior is that will be present later on an amplified with our research. So what is the biggest challenge for me right now? And I think it's this creating a i responsible. You must have heard about it today and throughout the last two days and it's really at the heart of everything. We do at Facebook. We're using AI to

benefit billions of people around the world and to do that. Well, we need to ensure that the system work fairly and equally for everyone involved. So what's my role today? I talked to this is basically pretty much the same size. But now I actually know what I'm doing and it's still revolves around people and program operation setting up an organization that allows Innovation to happen freely open research to happen Mighty Mouse support spiceworks, which is the Facebook platform for a Mal and open source as well. We work unfairness and responsibly I throughout the

company is not just the Facebook AI team's responsibility and we continue to work with in AI for social impact and making sure that everything we do has a global impact as well as helping our users day today. Emily that would leave the talk with one more example for you all this is a picture of some of the posters we see around the campus as you walk around her campus, there's tons of posters everywhere. You see some examples here nothing a Facebook as someone else's problem Bill harder proceed and be bold and these are some of the examples you see and for me, although it's a big company

and I come from another big company. It has the spirit of being almost like a start-up of do something bold. And when I walk in every day, would I remind myself is what would you do if you weren't afraid it's on one of our many of our posters actually and I encourage you to do the same. So thank you for having me here today with the pleasure.