Data scientist Anna Bethke had approached management with a new position – to become the head of Intel AI for Social Good to bring in new positively impactful projects to the group and company as a whole. Anna shares her story and lessons learned along the way.
Angie Chang: Hi, Anna.
Anna Bethke: Hello, how are you doing?
Angie Chang: Good. So we are back. I’m going to … there we go. Snazzy background. So we are recording the videos, this is a common question we get, and they will be available later on our website, girlgeek.io. Please tweet, the hashtag is GGXElevate.
Angie Chang: We’ve been sharing selfies of viewing parties, since it’s International Women’s Day, of women gathered, and allies, in rooms and offices, and coworking spaces around the world.
Angie Chang: Next up is Anna. She will be talking about how she developed the AI for Social Good program at Intel.
Anna Bethke: Cool, thank you. I’m super excited to be talking with everyone. I’m assuming my slides are showing, but let me know if that’s not the case. Awesome. I wanted to get into what does AI for Social Good mean, as well as what are some of the projects that we’ve been doing here at Intel, and things that I’ve seen elsewhere in the space, because it’s one that I’m super passionate about, and love.
Anna Bethke: But just first want to talk a little bit about how I got to where I am today. So I am from Colorado, and I also think of things in a geographic sense. Then I studied aerospace engineering at MIT out in Boston, and started on my career path as a geospatial data analyst in sorts, taking imagery from satellites, taking information from that, and then writing some algorithms to find different patterns of life, and anomalies.
Anna Bethke: Did something slightly different, but also geospatially related at Argonne National Labs, and that was in the Midwest, which was lovely, was really close to my husband’s family, but not quite the right fit for us. Moved over, and was doing a data science consulting type of gig at Lab 41, and landed at Intel doing data science work there.
Anna Bethke: So before I took up this role, I was primarily looking at natural language processing, deep learning techniques. How do we make these faster, what are the different things that we can do, what is the state of the art? It was very interesting, but I just have been so inspired by a lot of different projects, and I’ll talk about some of these groups a bit later.
Anna Bethke: Now I’ve been being a volunteer for Delta Analytics, as well as Data and Democracy, two groups that help pair you with some different projects that you can help a non-profit with, or help move the bar on what is important in the world.
Anna Bethke: That’s sort of how I define, and how AI for Social Good is defined, and it’s easiest to talk about specific projects than to say what this is, because AI is super nebulous. What is good is also something that has some subjectivity to it, but basically, the idea is how do we utilize AI hardware, and software, which is a lot of different techniques, and these technologies to really positively impact our world?
Anna Bethke: The thing that I find really promising and fascinating about this is that we can have a very large impact. This is a smorgasbord of some of the projects, and I’ll go into depth for a few of them. But there’s a lot of different verticals. Healthcare is a large one where we can start to be able to take these image segmentation networks, or object detection type of networks, and say, “Okay, well where is a potential tumor?” Or, “Where is a disease?”
Anna Bethke: “Is this something that looks benign or …” what’s the opposite? “Benign or …” sorry, I can’t say that word today. But basically, “Is this cancerous or not?” Taking these types of ideas from our research areas, and putting them into the field.
Anna Bethke: It’s very wide and varied. For earth or our different types of work there, we can do a lot of things, too. So ways to protect our natural resources, ways to protect, also, our man-made resources, so one of the projects was looking at restoring landmarks such as the Great Wall of China, or how do we map our structures and buildings so that we can have a better disaster response?
Anna Bethke: And then ourselves. How do we protect our kids against online threats, or physical threats? This is some work that the National Center for Missing & Exploited Children has been doing for years, and how can we help them with technology so that they can help protect and find potential perpetrators faster?
Anna Bethke: And then how do we help ourselves create these online communities that are better, so like preventing harassing text online, and even mitigating and stopping it.
Anna Bethke: That’s a high level. There’s more on the website, but the interesting thing is really once we get to dive into these projects. One of the ones that I think is really interesting, because it came from one of our software innovators. So these are basically entrepreneurials, individuals that are external to Intel, and they have these ideas of ways that can really help society, or these different projects, and there’s a link at the end of this presentation that you can get more information on this particular project.
Anna Bethke: Basically, this guy Peter Ma, he was looking at the issue that every minute a newborn dies from infection caused by lack of safe water or an unclean environment. This is worldwide, and it’s a very large issue. This was the World Health Organization, but the current systems that we have there are very expensive.
Anna Bethke: They require manual analysis, so you can’t just take your machine, and bring it from one village to another village, and that’s just not possible. But you want to make certain that the water everywhere in the community, and you’re going to need to be measuring this multiple times, because the water quality can change.
Anna Bethke: So what Peter did with some expertise help from Intel, is he built a convolutional neural network, basically. A computer vision model that is able to take a water sample, and using off the shelf products, as well as this Movidius Neural Compute Stick, which you can buy this commercially from a lot of different sites.
Anna Bethke: Don’t know if I have a link on it here. Oh yeah, if you go to the AI for Social Good website, then you can get to see more information on this product, and that has more information on how you can buy these NCS devices. They’re really low weight, you could actually see one in the image, I believe. That USB stick, basically.
Anna Bethke: He was able to build this entire prototype for less than $500, and now it’s even smaller, and less expensive, and it’s more than 95% accurate. So it might not be perfectly accurate, but it tells you a lot of information, and can really start to help communities know where their water is safe to drink, and where is problematic, which can greatly improve people’s lives.
Anna Bethke: Another really interesting project that I love is with a company called Resolve, and we’re building, with them, Trailguard AI, and the idea here is the camera that’s in the picture is this motion capture camera. Motion capture cameras are great. Scientists have been using them for a very long time to be able to monitor the health of animals, and where are animals located.
Anna Bethke: Park rangers are also starting to use this to be able to say, “Okay, when are there poachers in an area,” and how do we help this poaching epidemic, and really turn the tide on it? Because basically right now, National Geographic has identified that an elephant is poached every 15 minutes, or a rate of 3,5000 a year.
Anna Bethke: This presentation that I’m giving is about 15 minutes long, so in all probability, an elephant could’ve been killed during this, which is just really sad to think about. There’s not a lot of park rangers, it’s a massive area that they’re trying to cover, so what can we do more?
Anna Bethke: What we did with Resolve was embedded also the Movidius Vision Processing Unit, so this is the same chip that’s in that USB stick that was in the last project. But basically, we can run an object recognition network on the Edge, here.
Anna Bethke: Everything that is being processed is being done on this VPU, and basically what happens is an image is taken, that goes to the chip, the chip is able to run this CNN for object recognition, and in this case, we’re looking for people in particular, because this is what the park rangers are very interested in.
Anna Bethke: If there’s a person or a vehicle, then it’ll ping the park rangers, and this is really–
Anna Bethke: –Both reduce false alarms, about 75% of the images that the park rangers would’ve originally gotten, wouldn’t have anything in them. So basically, if a tree moves, then this camera goes off, because you want that to be very sensitive.
Anna Bethke: Now, by just sending the 25% that have a person or an animal, so I think people are only in about five or less percentage, depending on the camera, of course. You can greatly reduce the false alarms, as well as extend the battery life.
Anna Bethke: So we’re expecting these to last a year, which really helps, because then it’s harder for the poachers to find. If you’re always blazing a trail between all of these different cameras, then it’s pretty easy to figure out, a poacher can figure out where they are, and avoid them.
Anna Bethke: It’s really interesting. Something that we already have, both of these last two projects, as well as the next one, these technologies that we already have, like object detection is something that is getting more robust, that we’ve been using for a lot of different applications that we’ve been researching for a number of years now.
Anna Bethke: So how do we use it, though, for these really impactful purposes? One of the last projects I wanted to just mention before going into some of the stuff that I’ve learned while building up a program around these types of projects is the Wheelie. This is a project that we worked with a company called HOOBOX, and basically, they are robotics experts.
Anna Bethke: In the last example, Resolve are conservationist experts, so they bring deep knowledge about what the issue is, and we bring the technical expertise. So what HOOBOX saw, was there are a lot of people worldwide that are suffering from spinal cord injuries, and there’s more and more every day.
Anna Bethke: But the offerings for mobility devices can be expensive, complex, difficult to use, and there aren’t as many options as one would like. We developed the Wheelie 7 with them, and it lets users choose the most comfortable facial expressions to command their own wheelchair.
Anna Bethke: You can basically say, if I smile, go forward. If I raise my eyebrows, go backwards. If I open my mouth, go right. If I stick out my tongue, go left; things like that. This is really nice, because every person has different abilities, so by giving choices that extend the range of people that this’ll be really effective for.
Anna Bethke: And again, facial gesture recognition is built on a lot of deep learning applications that we have already looked at in-house, and so how do we apply it to this? What are the different things that we really need to think about in order to make sure that this product works for everybody in a very safe and reliable way, and that people’s privacy is protected, and all of the things that we really need to be considering while looking at this type of project.
Anna Bethke: This all of course is run on the wheelchair too, because you don’t want to be sending this information to the Cloud. That would take a long time, and if you are telling your wheelchair to stop, you want it to stop immediately. So this is run on the Intel NUC, it’s this little miniaturized PC with a customizable board. I think it’s four by four inches, so really small, can fit on the wheelchair, and it also doesn’t pull a lot of electricity, and the facial gestures are captured by this 3D RealSense camera, and that gives more information about the facial gesture than just a normal 2D camera.
Anna Bethke: Again, all these devices are things that we are commercially selling, which is great, because it’s things that are already built, and we can just improve them, make them better by seeing how they work in this new and different environment.
Anna Bethke: This project in particular was supported through a couple different projects. The Software Innovator Project as well, that was that CleanWater example that I gave a couple things ago, and our AI Builders Program. This one’s interesting, it’s sort of tuned for startup companies, and I’ll have a link for that one, too.
Anna Bethke: What does it really mean for me as being the head of this effort? I do a lot of different things. One is research and development, this is a slightly older picture of my cat, is a bit older, but I get to play with code, and do some literature research. That’s a little less now that the program is running, and getting a lot more interest in it, but it’s something that I try to carve out as well.
Anna Bethke: Sometimes I miss doing more of the technical work, but this is something that I am exceedingly passionate about, so I don’t mind not getting to code every day anymore. Another one is Connector. This was a breakfast ideation session last week, or a couple weeks ago, where I talked with a lot of different people like, “What can your companies do? What are the different ways that we can really just raise the bar, even just a little, with our technical expertise, with what our various organizations, or ourselves can offer?”
Anna Bethke: The last is Advocate. Talking at conferences, speaking to all y’all. One of the things I really hope to communicate is that this type of work is really important, and also really interesting and fun, and has a lot of very good business use cases that might not be as prevalent either.
Anna Bethke: I think a lot of us really want to do this type of work, because it’s the right thing to do, but there’s a lot of benefits too. It’s great for marketing, of course, as you could likely imagine, but it’s also really great for hiring and retention.
Anna Bethke: A lot of people want to be doing these types of projects, so the more that we offer them, the more that our companies can hire in this type of talent, and keep us all happy. Then the third for us specifically, being a hardware company, I think I eluded to it a bit, is that we really see a larger number of use cases of how can we apply technology, and then that helps us make certain that we are designing our technology in such a way that it is robust, and that there is a larger user base, basically.
Anna Bethke: So I’ve been learning all these different things along the way, but it’s interesting. There’s some other things too, so I wanted to share a few lessons. Asking for work that inspires you. The role that I’m in now didn’t exist, and I am so grateful for my manager, as well as the leadership here, that they’ve been really supportive in me taking on this position.
Anna Bethke: Helping me get the resources that I need, as well as helping to find what are the things that we can do now, in a couple months, where are the places that we really should be looking?
Anna Bethke: The second is that there are a lot of people who want to help, whether it’s helping plan meetings, whether it is doing the engineering work, being a contact coordinator. A lot of the projects that we have been developing are ones that one of my colleagues, or a colleague’s colleague has a friend who is doing this thing, and they are having this issue. Can we help out with that?
Anna Bethke: Those connections are wonderful. Low hanging fruit are wonderful as well. I say wonderful a lot. It’s very true. A lot of times we really try to go for, I think they’re called moon shots, but what is the coolest and the best thing that we could possibly do?
Anna Bethke: And while those are important too, there are things that we can do today, or in the next month, that are potentially quite easy for us to move the bar a little bit, but can have a really large impact in somebody’s life, or some animal’s life, the environment’s state.
Anna Bethke: Those are important to continue to look at, and to consider. But it’s also okay to say no, and this is one of the hardest things, and it’s really actually necessary to say no sometimes. I have been having to come to terms with the fact that I’m not able to do everything that I want to do, and we’re not able as an organization, or a company, to do everything, to help everybody, and so it’s sort of making sure along the way, that I’m preserving my own health and wellbeing, and sanity, and spending time with my cats, and my husband, and all of that at the same time, too.
Anna Bethke: And then who can help? Redirecting it to other resources, or saying, “I’m sorry. I can’t help at this moment, but maybe this person can.” Doing that redirect. But yeah, it’s hard. Then finally, you’re here for a reason. Whatever position you’re in, it’s awesome.
Anna Bethke: Imposter syndrome is one of my good friends now. I have definitely doubted myself along this way, when I was a data scientist, now as a head of a program, it crops up all the time. This is something that I remind myself of, and I think that we all should elevate each other.
Anna Bethke: I love communities like this, because I really feel like we do that. And then actually last, is that if you’re helping to debut hardware, there’s a high likelihood that they will take a picture of you holding the hardware.
Anna Bethke: This picture, I had no clue was going to be taken. This was right before the holidays, and I had just repainted my nails, and I had never painted them this bright and shiny, but I’ve come to terms with this too, and loving it.
Anna Bethke: I think it’s funny, so I have to laugh at myself a little bit, but I actually really like the super pink sparkly nails. How do you get started in this? Hopefully I have shared enough inspiration, and project examples. There’s more at AI for Social Good, at intel.ai, and this really follows a similar process as most other projects, so getting your ideas, finding the partners.
Anna Bethke: So the partners are someone that has the ability to really implement this into action, and that really varies, and then getting your research together, the data, the compute. There’s some things that could help, like at this AI Developer Program, and that actually gives you links to both the Software Innovator Program, as well as our AI Academy, if you’re a student, but there’s some Dev Compute there, which could be helpful.
Anna Bethke: And then your algorithm development, so how am I actually going to analyze all this data, and make sense of the world? And then testing and deployment. This is really important, of course, to make sure that the system is working before it goes out into the wild. Aibuilders.com is the startup company connector, if that’s something that you’re in, and then one of the really important things as we’re doing this, is to talk about, and think about how do we do project management, project deployment in a responsible way, so there’s a bunch of different resources that are out there. There’s a lot of toolkits, so this goes everywhere from checklists like Deon from DrivenData, which just [inaudible 00:23:00]. These other things you should be considering as you go through, to more algorithmically based mechanisms like the IBMs 360 Fairness Toolkit, or the What If tool from Google.
Anna Bethke: Take a look if you are doing a project. Either for a socially impactful, or anything else, and there’s a large discussion around this right now, which I love. I’ll leave you all this sample of various social good volunteer organizations.
Anna Bethke: This is definitely a growing area of interest, so something in the Bay that I’ve been involved in is Delta Analytics. This is mostly San Francisco, but the other ones are either completely based in the United States everywhere, or also global, so DataKind, Data for Democracy, Code for America, Visualization for Good is really cool if you’re more on the visualization side, and then there’s a lot of different hackathons and challenges that you can join, too.
Anna Bethke: So yeah, that’s it.
Angie Chang: Thank you, Anna. This has been a great, very informative, resourceful talk. There’s been a lot of chatter and questions. I don’t know if we have time for … I’ll send you the questions, and you can maybe answer them on Twitter. I think you’re pretty active on Twitter, and we can get all the questions answered, with helpful links.
Angie Chang: Our next session will be starting soon, so thank you so much for joining us.
Anna Bethke: Thank you, yeah, I’ll definitely answer them there.