 Hello, and yeah, welcome here to the Susie AI and personal assistance session. It's actually above, so I hope we will have some discussions and input here from different people. I also have set up like a hangout that we can have later with Michel Christen, who is the main developer of the Susie AI server mainly. And some of you might know him from other projects, so we'll come back later to this. The goal of this session is basically to bring different people together who are interested in personal assistance. And yeah, I've prepared some slides for the beginnings, talking about Susie AI, but we can also come back and talk about other personal assistants that are out there in the Linux community already. So let me talk a little bit about Susie AI. Susie AI is a smart framework for conversational assistance. And we say we need a free assistant. Like in the proprietary world, there are a number of assistants, but we don't have really a comparable free and open source assistant right now. Why do we need it? Because nowadays people are talking to devices. The conversational web, it's the future. Many of you know that probably we have a lot of Star Trek fans always in the community. Everyone knows like these examples from Star Trek, so that's pretty obvious to many here in the room probably. And then also, businesses need a way to interact. So I think the free software community and the free software projects were very successful because they also provided new opportunities for businesses. So if we are able to do this, then this will provide a lot of new business opportunities. But of course, it's expensive to develop our solutions. So if we collaborate in the free community also in this field, it will be a big opportunity. So people don't want to buy into a solution with an external proprietary vendor lock-in. We know this situation from years ago when Microsoft came out with this browser and like whenever we wanted to go to a banking website there was like this question, ActiveX not available, your browser is not compatible and so on. We are going back to this time, we have now Google Assistant for example, Google Home. Everything we do for Google Home is only compatible, only works with Google Home. Everything we do is Alexa, it's only working on Alexa and so on. For Apple, Siri, it's only working for Apple. We don't want to go back to this time, so we need a free assistant. So that's why we are proposing SUSE. Here are a few numbers. It's estimated that already by 2020, it's just in two and a half years, 75% of U.S. households will have smart speakers. And we have a few solutions out there, Cortana, Vivi, Microsoft, Clover, Bixby, Google Assistant, Amazon, Alexa, Siri, Jasper. I hope to get some feedback from some in the audience later on, which one have you used already, what is your experience with them and where are they suitable and what can we do to bring free assistance forward. For the conversational web, now we at SUSE AI, we develop a free and open technology platform with a full open stack. This is not just a chatbot that we want to have, we actually want to have a full open stack, so it's ready for business to customer, B2C, B2B. It should work offline and in-house, so you should be able to deploy it at home, you should be able to deploy it in your company, that is the goal. And it should be pluggable to existing APIs and data sources. So for example, if you have like an API in your company, if you have any service that you use, you should be able to use that API in your solution for a conversational client. And yeah, what we are doing is a simple yet powerful Wiki style authoring. This is our idea, we really want to make it easy. I will demonstrate the skill authoring later. So, and another thing, like I know many geeks here and they say, oh yeah, I really love this thing, but I will not install this at home because these devices that you for example set up at home, these personal assistant devices, they always telephone home elsewhere, always connected to the internet, so really it must be ready for offline networks. We can come back to the pros and cons later. So, we have already a fast growing and vibrant developer base and here's a few screenshots of what we are doing. So, we have here the Android client, we can already see a few features and Susie AI works on any device, supports text, voice, media, images and videos and of course we also want to support interactive forms. A lot of work is still to be done, but many things already work. Here is our screenshot from our WikiStyle authoring page and there's an underlying git, but the front end is just similar like to Wikipedia. You can click on it and edit these skills. Another feature is private skills and there's also reason for private skills, for example, like people are using private information or private keys and things like that. But these skills are open so far. So, what components do we have? So, these services here on the left, web service APIs and big data index, these are external services can also be deployed from you, so it depends but could be like external APIs. And then we have the Susie AI system. So, there's the Susie AI server. You can access the server on api.susieai. It will give you adjacent outputs. It's just like a form field and you get adjacent output and then there's an underlying git, their console services, their system components, part of the server, like for example the authentication, authorization and so on. And then when you make a call to this server, this server supplies adjacent and it's read by one of the clients. So, which clients do we have? We have a web client, Android client, iOS client and also clients like MagicMirror. Actually, when I check the stats, I don't know if you know the MagicMirror. It's kind of a project where you stand in front of a mirror and behind is a screen and you feel like you can talk with this mirror and there's some interaction happening. So, we integrated Susie AI into the MagicMirror for fun and there's a community using it. And then we have plugins for Chrome and Firefox and so we have a basic Linux app which is using Electron, so basically a web app integrated into Linux. Another thing we have is Messenger bots. You know this nowadays, Facebook for example can be used as a Messenger bot. We have Jitter, we have Slack, Telegram, Viber and Line and you can find them all on GitHub and deploy them yourself if you like. Okay, so who's our community? We have more than 100 active developers on Jithub. We also got a lot of Google Summer of Code students this year and yeah, we're showing Susie AI wherever we can. Yeah, here's Honfuck with the FOSSAsia community. So actually Susie AI is hosted on the GitHub repository of FOSSAsia, so we're taking advantage of the existing communities and developer base and try to get everyone excited and showed at make affairs at many different events and yeah, it's pretty fun to get together with all these people. Who are the competitors? So let's come back to this. Most of the competitors are actually proprietary closed source systems. So we have Harman, we have Sony, LG, Panasonic, Amazon Alexa. On most of the devices on the left we see actually Amazon Alexa is running on them. Recently Sonos also released a device with Amazon Alexa but they also said they will furthermore release it as well with Google. So basically it's just a computer and you put the software on it and of course we think it's better if there would be a free and open system running on all of these devices that should be our goal. Mike, it is possible to install Susie on any of these devices? Yeah, it's not possible to install yet. I've talked with Mani Huang who is the original hacker of the Xbox and he said it should be possible. It sounded like I can do it in a weekend but I think he can do it in a weekend but not me. So that would be so cool if people actually come out and try to install Susie on Google Home. But we need a few more things for that and I will come back later to this. So these are the proprietary devices that we have here. Then we also have smartphones of course. Any device now using Google comes already pre-installed with Google Assistant and Apple with Siri. So we have an Android app but usually when you have the Google Android phone you press on it and it's already coming up. Whereas with Susie AI people need to start the app. So competitive advantage, what's our competitive advantage nevertheless? We can potentially run on any device or any platform because it's open source. So people can adapt it to their platform if they like. It plugs into existing solutions. So if you have an API and you want to use a voice assistant you can just install Susie AI directed to the API and use it right away. And you could think of custom date sources. So in the server actually we have already much more features than you can see available through the front. For example you could use databases. There's already something implemented called LSTM, Long Short-term Memory. Some of you might be familiar with it. So all this is possible. And you could think of using Susie AI, for example, translation experts or mechanical experts. So you could think this further, not just have assistants that have specific skills but that are actually experts in one area. You have a database of knowledge you could connect it. There's no vendor lock-in, in-house offline deployments you can make them. And we believe it's increased security through an open review process. Of course people have to review it, right? Many different versions and we have a community backing. Okay so here are a few links. I will share them with you also after the talk and now I would like to show what we have done so far. So here is our skills Susie AI website. And our intention here is that people can create skills. We make it a little bit similar to the Google Home and Alexa website. We believe we don't need to reinvent your eyes. We just want to keep it simple and make it available. So there's a very simple skill, for example, here. Let's click on it and let's see how it works. Donald Trump quotes by Trump. You can see it's pretty standard. You can rate it, you can provide feedback. There's time-wise usage, there's information on device-wise usage, there's country-wise usage. So this is a page of a skill. And yeah, you can have here quote by Donald Trump and this is the skill. So how, I don't know, if anyone of you has done this before with Alexa on Amazon, you will see there's already one difference here. We provide much more information about the skill. So we keep it public. We let the public know what's going on with this skill. Then another thing here, and this is the main difference is you can click on the skill and see how it is written. And our core developer here, main developer Michael Christen, he really invented a very simple language. And I will show you that language also in a moment. But as you can see here, it's super simple. Basically here we just call an API and that's it. That already makes a very simple skill. So it's a Trump quote. Now, where can I find the tutorial here to how to make skills? Basically, just go here, when you make a skill, go here to the I, click on it, click on it, and it should have worked. Okay, let's try that in a different way. So basically on the left, you can create a skill. Okay, so that worked until yesterday. Now some JavaScript thingy doesn't work. So we're deploying live from GitHub to Torreal. And you can just go and search it online and also find the tutorial that we have. So it's like how many steps? 17 steps tutorial now that we have. And here you find step by step how you can make a Susie skill. There's a skill format. And we organize these skills in a kind of schema. And then there are different tutorial levels. So the simplest way to make a skill is just write a statement or question and an answer. Roses are red, Susie is a hack. So this is the question, this is the answer, and it will work. Then there are further ways. What is your favorite dish? And you can receive random answers. Random answers just like this straight line in between your answers always receive randomly one of these answers. Tutorial level number three, may I star you? Yes, you may. Tutorial level number four, the same principle. May I get a star? Yes, you may get. And then here we are starting to use query patterns in answers. Tutorial level number five, multiple patterns in queries and answers. For star, I can buy a star. Yeah, I believe then dollar one dollar is a good price for two dollars. So you can see we're advancing more and more on each tutorial level. And I recommend to you to go through these tutorial levels. And yeah, you can do like pretty cool things with this already. So you can have conditions for answers and rules as functions. You can have a bad JavaScript. Okay. You can call an API. So here are all examples. And yeah, you can, it's not just text that you can use as a skill. And like because we have visual assistance. Yeah, you can also use tables, pie chart, RSS, web search, map, images and video. So you can also supply that. Check. Okay. So tables also possible. So I recommend you to go through these. So what do we have? We have like a different clients. And maybe one thing more I can show you is the API. So that's another thing I talked to you about is basically you get information. So if I say hi, I will get a Jason here. And yeah, and that can be interpreted by any client. You can also go to the app. So actually our app is still unreleased. We're still like hope to release maybe next week or the week after. So it's pretty quickly done. So and yeah, community wise, please also check it out on GitHub. You always have these insights. We can see like the activity. You can see polls, contributors. So we really like a lot of things are going on. You see like it's like getting more and more. It's really rising. So basically you just go to the first Asia GitHub repository. Type in Susie and you will see all these repositories coming up. And basically like, yeah, we have mergers all the time like 10 minutes ago, 20 minutes ago, 12 hours ago, 12 hours ago. So it's really like a high pace development. So right now we are not claiming that we are in production stage. We're claiming we have a software deployed the software for testing online. Yeah. So at this stage are we now, but like most of the time things work. And usually you can go to chat dot Susie AI. But like the last pull request unfortunately created some issues. But you can test it for example on the Android client on the Susie client. The project that we are also focusing on right now is so we have it on all these devices, but we don't have a device itself just like Alexa and Google Home. And so we are right now trying really to make this happen as well. And with a Raspberry PI and a microphone and speakers and we need to pie head. So I think most of the time we spent on actually trying out what works what doesn't work. For example, I don't know if any of you is also working with hardware here. And we tried out to make it work with a USB microphone, but the quality of the voice interaction is really going down. So you need a good microphone. But on the other hand, we want developers to not spend, have to spend too much money in order to get like a prototype working. And so it's always like pros and cons and we are figuring out at the moment what is the best way to make it work. Also, there are a lot of things that actually only come to your mind once you develop a voice kit. For example, when you want to say stop, for example, you play music and you say stop, you actually need to implement like an action on top of another action. You need to stop an action and then this action might continue later. I would love to talk to anyone who has already experienced with this. So I think we are ready for questions and maybe I can also plug in and see if Michael can join us. So this is one thing we haven't thought about yet. Can we plug in sound here? Is that possible into the computer? Maybe not, right? With a mic next to it. Okay, good. Sure. Sure. So let's call Michael and see if he can join us. That would be cool. Thank you very much. Yes. So Michael, so great that you join us. So we just had a presentation here about Susie AI and maybe do you want to say a few words about how the development works and what is the idea of the different components, how they work together? Can you hear that? No, hang on Michael. Let's check the microphone. So I think we have to increase the sound a bit. Yeah, just speak again. Can you hear me now? Can you hear? Yes? Like a little bit? No? How is it? I'm going to talk next to chat Susie AI web page again. So Mario, if you want to show this, it should work. Mario? Yeah. So maybe... Am I loud enough? Is it loud enough? No, not really. I can somehow hear it better than nothing. Better than nothing. No, maybe not. Yeah. Okay, so maybe that was a good try, but I think it's difficult to make it work, right? No. Yes, no, it's like we don't have a connection to the thing. We just have a microphone on the speaker. So Micha, it seems like not possible here. We don't have a connector. Okay, thank you. Okay, Micha, great. So we saw your pictures, but the sound is not working. So let me get back to trying out the Susie AI system. Thank you. So here is, for example, like a chat system that we implemented as a web service. And now, like, you can go to this services on chat.Susie.ai. And yeah, so if you are not sure, like, how to try out... So we have one very nice small feature on our Skills page. So what you could do is you could, for example, try any skill and click on the skill. For example, so click on the quote here. Can you see that quote here, for example? So you click on the skill, then a new window will open and you can try out any skill. That's how easy it is. So what do we see here? Quote by Donald Trump is what I said, and then the answer is... Okay, and then you can also like or dislike here. So usually what we get when we have these kind of rules, so people see, okay, you make skills here, where's the AI? Yeah, I can see you have a rule. But now we compare this with Alexa and Google. What do they have? What are the most popular skills? Setting alarm, yeah, translate something. What's the weather? Yeah, so where's the AI here? Yeah, so we already get very far with standard rules. And this also helps us to solve the chicken and egg problem. I mean, if you do real AI, like training data and so on, you actually need data, you need users, you need to train it. But here we can just start with rules, and this is what others are doing as well. So that's what we do here. But the first step in order to actually go to feedback is also to provide users a way to give feedback to specific conversations. So if you have a lot of users and they give feedback to conversations, we can analyze this data. It's the first step. There are many more ideas, and I'm sure we have experts here in AI, so would love to talk to you about this. Yeah, and yeah, I recommend to you to try it out, go to the skills, check if they work, if they don't work, help us to debug them. As you know, you can click on it and simply try it out and edit these skills. Tell me a joke. Okay, I click on it, and the system will tell me a joke. Whiteboards are white because Chuck Norris get them way. Okay, so also like in Susie on Chrome, the voice also works already. This is, I'm now using Chromium. Unfortunately, it doesn't support voice yet. We'll give you an error message. Yeah, it seems like many websites have this problem, so something we also need to solve in the free software community. So they're always pros and cons, but like actually this is already an advantage to the proprietary competitors because they don't have a web client. Right, so we do this. If you go around, you will see and discover much more features. For example, you can create also your skill bot, which you can host on your website. This was like in cooperation with Google Summer of Code students. So, yeah, create your private bot, change colors. We have two views, a view for UI or a code view always in each steps. You can choose your own icons. You can configure them, allow bots or only on specific websites. And then you simply deploy them by copying a short script into your page. It's all work in progress, but I think pretty cool. So it will enable us to use free and open on each website. So I think we're ready for questions and maybe like other feedback, other people would like to show something if there is any question. So does anyone already use a personal assistant somewhere? Where do you use it? To turn on your TV? Which one are you using? Google Home. So Google is a Chromecast device? Yes, I have a Chromecast plugged into the TV and it's always on wall power. So now Google can turn on my TV when I ask politely, which saves six or seven seconds each day. Which country are you based in? United States. So United States is always kind of a privileged country with these services because they always have all the services first. I'm based in Germany and I tried for example something else which is like play a song on YouTube on my device, on my Google Home. It said to me for a long time, oh that's not possible, it's not enabled yet in your country. It changed around four weeks ago. Now it tells me you need a Chromecast device which basically means I need my computer to stream to my speakers, my Google Home speakers. Why would I need Google Home if I can't just plug in speakers? So it made me feel immediately the restrictions that these companies put on me. Why am I allowed to play a YouTube video on my computer? Why am I not allowed to play it directly on my Google Home device which is right next to it? Of course, I understand the business logic behind it but I don't like restrictions. So a thing that we will show in the upcoming months when we get our device ready is play YouTube on Susie AI smart speakers. Because that's something that we can do and anyone can write skills or write plugins or whatever they like and do what they want. They don't have to wait until Google accepts your skill. Another thing that we did, we actually developed a skill for Susie AI and for Amazon Alexa. Unfortunately, they didn't accept it. So we're still trying to figure out that they accepted but how cool would that be? Use a Google Home and then switch on Susie on your Google Home, right? So yeah, this is one thing. I can show you the feature here on things that we achieved already to... Okay, so usually what music do you like? Yeah. Let's try some craft work. You're from Germany. Ah, craft work. Okay, let's see if that finds craft work. Okay, so it will play... Very good. It will play a craft work song. So that's already one thing, so we can do already one thing in any country that Google Home is right now not doing. So even though we're still at the beginning. Yes? Some of us have been staying away from these devices, be it Google Home or Alexa. We are even worried that our phones are listening to us and I have even experienced this that I'm talking to a friend about a particular company which is not a big company and next day I see the same ad on YouTube and this is happening. So when you are writing the skills, where exactly is the data stored and what kind of privacy respecting services or restrictions are there and how the user privacy is handled. Could you touch on that? Absolutely. Actually, these are many questions and one question. So firstly, it's free and open source and we deploy right from GitHub. We have a development branch, we have a master branch. We deploy for it. So you can see the code, you can see what we are doing. There is however one thing, you are calling external APIs and if you enter like any private data and that is like related to APIs, let's say IPs, depending on what you do, for example you imagine there is a form element, I could say a form element, show me a form of a company that asks you to fill in something that could be shown in Susie. And then you say submit or you would still submit or say you use an email feature here, contact form, you could use this as a contact form on your website. It is clear that this is submitted to the company or to the one collecting the form. I think every user should be aware of it. And in regards to what Susie collects, Susie collects the interaction with the user and you see that also with Amazon and so on. Like owners of skills can usually see the interaction because they also want to improve their skills. There is a question when people stop to use the skills and so on, so this is a tricky question because if you don't collect data, how can you improve your skill? And also it is a tricky question in the way that some people ask us, please release all the data that you collect as open data. I am like, oh well, it could be private data, there is a European law, I am not actually sure, we are developing it right now. So there are a lot of questions to this that we have to take care of. What I can say is we are still figuring things step by step. However, like I already know right now here that many people will not be able, sorry, many people will not be happy with the choices that we will make at one point. But we are free software. You can deploy it on your own device, in your own cloud and do whatever you want. So it's like WordPress.com, right? You can take it from them or you can deploy it yourself. So this is our offer to the community and the other things we try to sort them out as well as possible. There is an additional point though that I want to make. So we are really trying to be good, but I am not always sure if we like, because many people are very privacy conscious and I am not sure if we can actually always take the exact privacy conscious, because we need data to also get into more into machine learning and improve the flow and so on. So I don't know if you go to the skill tutorial, you will scroll down and the last part of the skill tutorial we haven't implemented yet, which is for example like inter-SUSI dialogue. So two Susis talking with each other and talking with you and all like really cool stuff that people in the community think about. But however, this free software can do what you want and in regards to our device we have a different approach to Google and Alexa. Our device will have a Susi AI server stored itself on the device. This means if you plug it out of the internet it will still be able to provide basic functionality. So right now we are using for example cloud services for speech recognition. If you plug it out it will still be able to use local services on the device itself, like let's say pocket text and flight for voice recognition, text-to-speech-to-text, this kind of solution. It will sound a bit like Stephen Hawking, but it's a start and so it will still work, but you have to be aware. I'm not sure if all the people are always aware. If you want to know how is the weather you need to connect to an online service or you have an IoT device at home but if you want to know certain things you need to connect to the internet and you will get the data just like when you visit a website. It's just like visiting a website but you use a voice client to do this. And in regards to the architecture approach we only develop the parts that don't exist. It is not our goal to develop every part ourselves. So let's say there are good solutions for voice recognition and speech-to-text-to-speech and often the challenge is rather that they don't have the data for certain languages. The solutions are good but we need the data. So our goal is not to develop every part. Basically we focus a lot on the server because that is what doesn't exist. And there's another solution like Mycroft. It doesn't convince us in its approach with the technology and there are other approaches like OpenHab or Jasper. They often like as far as I can see they focus more on IoT and that could be even a part of Susie AI so connecting with IoT services. So let's see what comes out of it. I'm really happy that it's open and that's much appreciated and I understand that you will need to collect data to understand patterns and develop the skills. I just want to make sure that the decisions you make should be able to not personally identify people and preferably not send them ads back which the competitors do. Otherwise Susie AI seems a great project. Absolutely. So yeah, I think there are many... It also depends what choices companies or developers will take that use Susie AI. I imagine it as a platform, like really a framework and there will be different plugins and there will be different skills and choices people make. So basically we want to offer also like a tick that you can say, right, you know that from Linux distributions share anonymous user data. The question is where does... So for me, where does anonymity start? So some people say collecting IP addresses is already not anonymous, yeah, or geolocation. So like we try to be good as far as possible. Click line and let the user know if he's crossing the line. Yes. Okay. Cool. So we'd love to have more discussions with you guys and yeah, I hope we follow up. Thanks a lot for joining and thanks a lot by the way like for all the developers. We have really amazing Google sum of code students this year. Most of them actually in India. We have Jougartfest in India on 30th of September in Hyderabad. So anyone who wants to join us and meet many of the developers please come over to India and thanks a lot for all the contributors and people who make the skills and I'm really excited about this project. Thank you for joining here the session and we continue the discussion. Thank you very much.