Open source tools, AI for Dota, and enterprise ML adoption

Bandwidth for Changelog is provided by Fastly. Learn more@fastly.com we move fast and fix things here at Changelog because of Rollbar. Check them out@rollbar.com and we're hosted on Linode servers. Head to linode.com changelog this episode is brought to you by DigitalOcean. They now have CPU optimized droplets with dedicated hyper threads from Best in class Intel CPUs for all your machine learning and batch processing needs. You can easily spin up their one click machine learning and AI application. This gives you immediate access to Python 3R, Jupiter Notebook, TensorFlow, Sidekit and PyTorch. Use our special link to get a $100 credit for digital Ocean and try today for free. Head to do co changelog once again.co changelog. Welcome to Practical AI, a weekly podcast about making artificial intelligence practical, productive and accessible to everyone. This is where conversations around AI, machine learning and data science happen. Join the community and slack with us around various topics of the show@changelog.com community. Follow us on Twitter. We're at Practical AIFM and now onto the show.

Well, welcome to Practical AI. I'm joined by my co host Chris Benson, who is a digital transformation and an AI specialist. And I'm Daniel Whitenack, working in AI for Good. And we're going to do another News and Updates and Learning Resources episode for everyone. I think there's been a pretty good response on that. There's a lot of news to keep up with in the AI world and always great to have a few more learning resources at your fingertips. So. Hey Chris, how you doing?

I'm doing pretty good. I'm excited about it. We got some cool stuff to talk about this week.

Yeah, for sure. As always there's surprising things each week and things that are sometimes expected, but yeah, it's always exciting regardless.

Absolutely. So what, have you seen this last week?

Yeah, the first one that came across my path was this survey that O'Reilly did called the State of Machine Learning Adoption in the Enterprise and they have a blog post about it, but then you can actually download the full report. I think you have to put in your email or whatever to download it, but it's free and it is pretty interesting. If you're working in particular at a larger company and interested to know kind of how the landscape of machine learning is playing out in larger companies, I think it's really relevant. They ask a bunch of different questions, everything from what people's titles are, you know, where they're located. But also I think some Interesting things. The one that piqued my interest was actually who's building the machine learning models within an enterprise company. The biggest percentage of that were kind of embedded data science teams, but then kind of went down the percentages from there through external consultants, all the way down to cloud ML services. And I was actually pretty surprised. There was only like 3% of people using Cloud ML services, or at least that's how I read the information really. And that was actually really surprising to me. I don't know if you have thoughts on that. I would have expected that to be much higher just because it's so easy to use these services.

Maybe that's because we're in our little AI bubble and we think about this and are acting in it all the time. But I admit we're hearing constantly from these cloud providers about their services and I guess I'm a bit surprised too that it's not a higher uptake.

Yeah, I don't know. I mean, I think for anyone out there that in particular, if you're maybe in a software engineering role or in a. Primarily a team with primarily software engineers, I think using these cloud ML services and black box sort of models like from Machinebox, if you remember back to our episode to using those sorts of things, are incredibly powerful, where you can think about, you're writing an application and you can just think about, oh, I want to integrate speech to text, or I want to integrate image object recognition in my application. Well, you don't have to build a model, you can just utilize one of these services. And I think that's incredibly powerful where you can think more about the functionality that you're trying to enable rather than the neural network architecture or whatever it is.

Absolutely. And for listeners, that was episode two on Machine Box, if they want to reference that, it was a great episode. So I imagine that that will change dramatically over the next few years with everyone trying to get. This is. Maybe this is one of those moments where companies are still certainly trying to figure out how to incorporate AI into their strategy and maybe we'll see a much higher uptake when that occurs.

Yeah, for sure.

So I ran across several articles and the first one was called autocarras, the Killer of Google's Autom. And part of what caught my attention was probably the provocative title. But they start off talking about kind of telling again what Google AutoML is and that it's based on the neural architecture search that Google developed, which is really about searching for an optimal neural network architecture or model to do a particular task on a given data set, then I think the reason they introduced that is they wanted to say, hey, there's this new thing called autocarros, which is an open source Python library. And that's what really caught my attention is one of the thesis of the article was that in addition to this autokaris library just being a great library to get into that it's open source. And they really noted that if there is anything that could give something like Google automl a run for its money, it's the fact that as you get these high quality alternatives in the open source world that are available, instead of paying Google $20 an hour to use, you know, to use their AutoML implementation, that is probably a trend that will be a powerful thing to come in the years ahead. And I know that both I love open source and I know you do too. And so that really caught my attention.

Yeah, and maybe, you know, that is what is partly factoring into this trend of, you know, people not using the cloud ML services much as we might have thought. Simply just because there's so many great open source packages out there, which a lot of them like, like you're kind of mentioning here, don't require you to think through an entire neural network architecture, but allow you to use a lot of things out of the box, utilize pre trained models, utilize things as a service via JSON API or whatever it is. And so, you know, maybe that is partly why that we're seeing that trend. I noticed that this one, you know, it's not from Google, but it is on, you know, Open source on GitHub, it's extremely active as it looks like it has, you know, over 2,000 stars. And so yeah, good, good catch. So the next one that I found kind of follows in that trend as well of open source tooling. And this one's called Netron, I guess a play on Neutron I assume. But a lot of people are probably familiar with like Tensor Board, which is like a visual tool attached to or very tightly integrated with TensorFlow which allows you to kind of visualize the architecture of your neural network along with many other things. But this one caught my attention just because of how well supported all of these different types of models are from all sorts of various frameworks. So Netron is also a way for you to visualize your neural network, the structure of it and various things about it. But it says right now that it supports Onyx format, Keras, CoreML, TensorFlow Lite. It also has experimental support for Caffe, Caffe2, MXnet, TensorFlow, JS and TensorFlow. And so this is really like a pretty cool thing and a trend that I think I've mentioned on this podcast before that I really enjoy seeing in the community is this kind of idea of interoperability where this is a tool for visualizing your neural network and it doesn't really matter which framework you're using, but there's interoperability with a whole bunch of them. So it's definitely worth looking at, especially if you don't want to tie yourself into a particular framework or set of tools.

Yeah, I'm looking at the GitHub page while you're talking about it and it looks pretty great. It has almost 1400 stars and 131 forks at the time that we're recording. And it looks very active in terms of there's a lot of recent updates to it. So in addition to all the different frameworks it supports with models and stuff, I'm definitely going to try this one out myself for sure. So the next thing will take a little bit of a turn. I'm often really interested in some of the non technical posts as well about how AI is affecting the world in different ways. And I came across one that is from Harvard Business Review and it's called what's the Purpose of Company in the Age of AI? And you know, I'm often talking about digital transformation in general and how AI affects that. And so that, that caught my attention. But it was interesting to see Harvard Business Review seriously considering that with these new tools, how does that literally change the business functions? Technology aside, that AI is introducing and they kind of, to summarize it, they kind of came down to that. From a business strategy standpoint, AI was really going to have an effect in one of the four following ways. And one is really AI being used to exploit existing advantage and to make that existing advantage more pronounced. They also talked a lot about trade offs in terms of long term and short term. And we're all used to hearing about companies that get caught in the short term, you know, for stock market, for quarterly reporting and stuff like that. But that AI can be used to figure out what trade offs make sense for moving companies more into a long term perspective and that there may be some insights there that would not otherwise be available. The third one that they mentioned really appealed to me and I know it's going to appeal to you and that they're talking about, they define it as a moral or spiritual call to action. And I know you explicitly always talk about AI for good and that's a big important thing for me as well. And as I think about my future and the organizations that I want to be part of, knowing that that organization stands for something more than just making a profit, a motivating factor. So I found it really interesting to think of AI being used to promote that bettering the world approach. And then the final thing was really kind of being the. They use Steve Jobs and Elon Musk as examples as really people who are going to really meld the world into their own view rather than trying to fit into the world. And you might say that that is innovation for the purpose of creating value. But I, I just like seeing this conversation where companies are recognizing how important this is and recognizing it's not just another tech coming in, but it's actually something that's going to affect the core way they operate.

Yeah, this is great. I think this is a really great perspective, especially the things like you mentioned. I mean, you know, the things that kind of pique my interest and coming from the AI for good perspective in particular, where, you know, it talks about a moral or spiritual call to action. And I know I'm motivated primarily by my Christian faith in terms of infusing a morality into the technology that I build. And I had a lot of those conversations with people lately where the morality, if there is some in AI, is really driven by that of its creators. And so to be a part of the development of AI is also to be a part of that kind of moral piece of it. And we really need to be having those conversations. And it's great to see that there's a lot of people having those conversations in various circles. And it's great to see that.

And I'm going to surprise and embarrass you and note to our listeners that Daniel is speaking at an upcoming conference that has a faith based perspective on technology. And what is it called, Daniel?

It's called Faith Leads. It's in Nashville and having a lot of great conversations recently with, with people in that context. That's going to be a great conference. So I recommend you check that out if you're at all interested in that kind of intersection. But I know that there's a lot of people having these conversations. I've seen a bunch of articles with people talking about how your worldview and how your morality is kind of infused into the technology you build. And not only is a separate piece of who you are, but it actually can kind of mold in very naturally with the technology that you build. I think that helps people also feel like they can put their whole selves into the technology that they're building and create a lot of passion for the things that they're building, which is super important.

And to finish that up, it's so good to see. When I saw that you tweeted that, I wasn't surprised and I was very happy. And as we look and other people we know in the space are really focused on using AI for good, I think it's a fantastic counterpoint to the predictions of AI as a, as a, as a scary thing in so many people's minds. I just love seeing these great use cases for using this technology to better the world. And I hope that our listeners will help us spread that across the AI industry in general.

Yeah, yeah, for sure. And I definitely recommend, I mean, I think there's a couple of great links as well. I know in our episode three, we talked to Amanda and Latifah and Peter about the great work that they're doing with TensorFlow helping African farmers. If you're at all interested, I just encourage you to listen to that episode. It's incredibly inspiring just to utilize the skills that we have for helping people improve their quality of life in a.

Real practical way and literally saving lives in Africa there. I mean, literally saving lives. It was an emotional episode which. So yeah, absolutely, they should go listen to that.

This episode of practical AI is brought to you by Hired. One thing people hate doing is searching for a new job. It's so painful to search through open positions on every job board under the sun. The process to find a new job is such a mess. Yes, if only there was an easier way. Well, I'm here to tell you there is. Our friends at Hired have made it. So the companies send you offers with salary, benefits and even equity up front. All you have to do is answer a few questions to showcase who you are and what type of job you're looking for. They work with more than 6, 000 companies, from startups to large publicly traded companies in 14 major tech hubs in North America and Europe. You get to see all of your interview requests. You can accept, reject or make changes to their offer even before you talk with anyone. And it's totally free. This isn't going to cost you anything. It's not like you have to go there and spend money to get this opportunity. And if you get a job through Hired, they're even going to give you a bonus. Normally it's 300, but because you're a listener of practically I, it's 600 instead. Even if you're not looking for a job. You can refer a friend and Haier will send you a check for $1,337 when they accept the job. As you can see, Hired makes it too easy. Get started@hired.com PracticalAI.

Well, I'll bring us into the next thing that I found here, which is not totally unrelated but in a different vein, and that's that Julia. The Julia team if you're not familiar, Julia is another programming language that's very prominent in scientific computing and in certain communities especially, I think more in the academic sense. But it's kind of starting to filter into industry and they just released version 1.0. So first of all, congrats to the Julia team. That's really great. And I have to say, if you haven't taken a look at Julia, just try a few examples. I think that you'll appreciate what they're trying to do. I was at juliacon, I think it was last year, and the community is just really doing some amazing things in kind of distributed computing, large scale scientific computing, but also in terms of machine learning and utilizing GPUs and a bunch of different things. So we'll link to the blog post about Julia 1.0, but just encourage you to give the team congrats on Twitter and also try out a few examples if you've never used Julia.

All right, I will definitely leap into that. That's a huge congratulations to hitting 1.0. I've also been watching Julia develop over the last few years and it's just an impressive language and it's coming along at a perfect time for that. I actually have my next article is actually kind of a counterpoint to the Harvard Business Review. And I just mentioned before, this one is a blog post by a person named Ian Hogarth, if I'm pronouncing that right, called AI nationalism. And the crux of it is, whereas the Harvard Business Review talked about the changes that companies are having to do think about in terms of their operations, this blog talks about geopolitics and economic concerns and how AI is really going to be driving entirely new types of geopolitics in the years ahead, to the point where it will transform not only economies, but military strategy and thinking. And the author actually goes so far as to suggest that AI policy may eventually be one of the most, if not the most important parts of government policy because of the profound impact that it has on government operations. And you know, as we're in this moment where not only in the United States, but around the world. We're having all sorts of tumultuous politics with people on different sides, and obviously it's at times very hotly contested. We seem to have left the period of politics being a slightly kinder, gentler thing to do in the past. But as we think about how AI may affect this, I just found that an interesting thought process. And he goes on in this article in quite a lot of detail in a bunch of areas, far more than we can cover in the podcast. So I, I would certainly encourage listeners to take a read and be thoughtful about it.

Yeah. And I definitely will take a look at that and take that perspective in. It's always good to have those checks and balances and make sure you're hearing different sides of the story. I would agree with you in the sense that especially in light of, I think all of us, when we were watching, for example, the Facebook hearings at Congress, were just kind of, at least all of us work in this industry and are made maybe of a younger generation or just cringing at the fact of how little is understood at the government level about these new techniques and what's really driving, really driving decision making in industry, which is so different. And as that filters into government, I can't help but think that there's really going to be some profound changes at every level of government in terms of how they go about their decision making and how they manage companies that are utilizing these technologies as well.

Yeah, I agree. I remember a big part. Aside from the actual reporting on the hearings themselves, some of the senators really, really took a beating when it became how apparent it was that they weren't familiar with the implications of these technologies. And we live in a time where we can't really afford to ignore these things anymore. I mean, they're not just technology. They affect the way we live our everyday life and what can and cannot happen, for sure.

So now onto what really matters in the global context, which is video games. So I don't know if you're a video game connoisseur. I haven't been a big video gamer since the days of Super Nintendo, but there is this very popular game called Dota. You probably, probably most everyone has heard of it, so I probably don't need to. Need to mention anything about it. But OpenAI developed five. My understanding is that it's, it's called OpenAI 5, which is a team of five neural networks. So this is a team game. And so they have a team of five neural networks called OpenAI5. And what, what happened is they played a Tournament recently where they, they played against the team. This says of the 99 percentile Dota players. So essentially some of the best in the world. Right? And they won. I think it was two out of three. Correct me if I'm wrong in our community, but yeah, I think there's a lot of interesting things about this. I mean it's just kind of entertaining in general as these things are, along with AlphaGo and other things. But I also think it's pretty interesting that this is kind of a. It has that team play element. It also has an element in the game which is like drafting, which I guess is considered to be a pretty hard, hard challenge. And so there's a lot of facets to this that are pretty interesting. And so either if you're interested in DOTA or if you're interested in these sorts of game playing neural networks, this is a really unique one and one to look into. And there's some videos online and all of that good stuff that you can dive into.

That's really interesting. It's funny how. And I'm not a big gamer either, so I won't really go too far because I'm out of my depth. But we've seen so many demonstrations of AI capabilities in recent years through gaming in different ways, in different capabilities. And so I'm always wondering what's going next. And so this is a fascinating one, I guess. Are we wanting to move into some learning resources at this point?

Let's do it. Let's learn something.

Okay, sounds good. You want to go for it?

Sure, yeah. So what I was going to point out this week for people to try out is the Pytorch tutorial. So if you go, if you just search for Pytorch tutorials, of course we'll include it in the show notes, but search for that. The reason why I mentioned this is I'm actually going through some of these now and have been for a bit in preparation for some workshop materials that I'm putting together and some online course materials which hopefully I'll be sharing in a future learning resources episode. But yeah, this has been really great for me to learn a little bit more about Pytorch, which I've thoroughly enjoyed working with Pytorch so far. It's been very natural, at least for me and in my background and the tutorials really help with that. So they have a bunch, you know, deep learning with PyTorch, a 60 minute blitz. PyTorch for former Torch users. There's probably not a ton of Those. Maybe there is, maybe there's more than I think. But there's also learning Pytorch with examples, transfer learning tutorial, data loading and processing tutorial. It's really, you know, practical things and they give you patterns. What I love is that they give you patterns that you can kind of reuse in your Pytorch programs. So you're, you know, you're not always starting from scratch. You kind of, kind of do a lot of copy paste and go from there.

Yeah. This is a really rich set of tutorials that they have here as I'm looking through it. And you know, as we've talked about different frameworks over time and we really have a great set of resources by each of the frameworks. This PyTorch one is awesome. We've talked about TensorFlow and Keras and various others in the past, but if someone out there is listening and thinking about looking at this, see one that feels right and just dive into it. There's so many great examples where you can take whichever framework, Pytorch or others, and start doing stuff that just jump into the pool and start doing stuff and you'll find that you can get productive pretty quickly there.

Yeah, just, just jump in and try to run something that works. Right. And then go from there. You don't have to feel like you need to understand every little piece of it before you run something. Just take one of these and try it, run it, see, see if it works and then try to start thinking about, oh, when could I use this with my own data? Or what if I modified it to do this or that and you know, take that approach, I think is really useful.

Okay. I ran across something about, it's almost, it's almost meta about learning. It was on Lifehacker and it' to get started in machine learning, learning and robotics. And before we go on, next week we're going to be interviewing Chris DeBellas on robotic perception using Mask R CNN. And with that in mind as a future thing and thinking about crossing machine learning and robotics, this was interesting not only because I knew that was coming, but also because it talks about these two gentlemen. And I'm not, I'm not going to say their names because I'll butcher it terribly. The link is in the show notes, but notice that one of them at least was 20 years old, maybe both of them, and they're, they were trying to get started in this and they were involved in a hackathon and they just kind of talked about some of their lessons learned about how to get started in this field. And there's so many people, whether you're 20 or 40 or 60, there's so many people that are starting to move into this that I thought that they had a really great perspective. And a couple of the key things that they said that if you're starting out, they refer to it as cross the streams. And what they meant by that is to think out of the box and not think about the problem you're trying to solve in the way everybody that came before you might have solved that. With previous technologies, with new advancements happening so fast and with robotics, it may not just be from an algorithmic standpoint, it might be the sensors that you're using and where sensors are applied and how they're combined. And they basically said go for something that other people aren't necessarily doing and see if you can make it work. And then the next thing is they said is get an assignment and that is to make it real. They were involved in a hackathon and in that perspective they had a set time limit to knock some code out. And with that time approaching quickly, you have to produce whatever you can in a short amount of time. But that forced them to really think quickly and act on it quickly and see what they could produce. And that assignment they said made a big difference. And finally when you have your assignment, they said break down your project. Instead of being overwhelmed and saying, oh my gosh, we've taken this very ambitious assignment on in terms of how we're going to approach and we have a set timeline, they just said break it down to pieces just like you would if you were a software engineer or any one of many other things. It's a project. And a project is a big thing that's composed of lots of little things, things. And they said that they just, they would basically divided and conquered the project and were able to use open source tools like, like Pandas, which I know you mentioned in our last conversation. And they were out able to turn out a good product. And I just thought it was a great attitude with some great practical advice for doing practical AI at an entry level. And I wanted to share that with our, our listeners.

Awesome. Yeah. And, and there's a bunch of resources if you're looking, maybe you're not, you don't have a hackathon near you, you or something like that. If you're looking for a project to get started with or you feel like you've built up some skills and want to try them out, there's a bunch of ways to do that remotely as well. Of course you can try out different competitions and stuff, but there's also things like Datakind, if you're not familiar with that, that's a way to volunteer on real projects related to social good things like poverty in global warming and public health. There's a lot of ways to get involved and if you're, if you're interested and need some help in finding some of those opportunities, make sure and pop over to our community. We have a Slack channel. You can find that@changelog.com community there's a practical AI channel in that Slack and we'd be happy to talk through some of those things and. Or if you have interesting news coming your way, keep us up to date with what you're finding interesting and we'll look forward to hearing from you.

Yeah, we do have a great group of listeners that are active on those in slack. Also on LinkedIn, we have a practical AI group on LinkedIn which we actively monitor and we're on social media. So there's a lot of great ways of reaching out. We are very accessible and so don't hesitate to reach out to us and give us your feedback, suggestions and we're looking forward to this conversation.

All right, sounds good. Well, I'll see you or talk to you next week, Chris, about Mask r cnn and I'm looking forward to figuring out what that is and having that discussion.

Me too. I'm looking forward to it. Have a good week, Daniel.

Yeah, you too.

Bye. Bye.

All right, thank you for tuning into this episode of Practically AI. If you enjoyed this show, do us a favor. Go on itunes, give us a rating, go in your podcast app and favorite it. If you are on Twitter or social network, share a link, Facebook on with a friend, whatever you got to do. Share the show with a friend if you enjoyed it. And bandwidth for changelog is provided by Fastly. Learn more@fastly.com and we catch our errors before our users do here at Changelog because of Rollbar. Check them out@roar.com changelog and we're hosted on Linode Cloud Servers. Head to lino.com changelog check them out. Support this show this episode is hosted by Daniel White Knack and Chris Benson. Editing is done by Tim Smith, the music is by Breakmaster Cylinder and you can find more shows just like this@changelog.com when you go there, pop in your email address. Get our weekly email keeping you up to date with the news and podcasts for developers in your inbox every single week. Thanks for tuning in. We'll see you next week.