Humanitarians Ai Course on Courses: How to Build AI Courses by Learning-by-Teaching

Intro: What are “Courses” vs YouTube playlists?

[0:00]Okay, bear here. I'm introducing courses. What courses are is they're basically a YouTube playlist except they're far more structured. What I mean by that, so when I create a course for a um course at Nor Eastern, it has a syllabus and modules. You're going to learn this and this and this and this and this and this. So I've done that to for I've set up basically course outlines for maybe a 100 courses in AI. So you can just take a look at of them anywhere here on you know here's a course outline for for AI for personalized learning. A lot of these courses already have hundreds of notebooks. For example, for this one I do. So, I'm going to start adding those notebooks or have volunteers start adding those notebooks to this particular course. Many of these courses are courses that I've taught at Nor Eastern. Um, and so I have a lot of materials for them there. Uh, but these are all

Syllabi, modules, and 100+ AI course outlines

[1:05]Courses. And so, the idea here is for people to learn by teaching. So what I'm going to do is I'm going to show people how to make courses by making a couple of courses. The two courses that I'm going to make is how to use AI to get a job if you're an international student and that's related to the 80 days to stay initiative. And the other course I'm going to do is more of a personal one and that's my curiosity on AI music. So my understanding of AI music is something like Spotify gets, you know, at least 100,000 MO songs a day. And so even if your music is great, it won't be found. And so I get 200 spam messages a day about these people claiming that they can put me on a playlist and, you know, get me hurt. Being a researcher, I'm skeptical. And so what we do is we do research. So the way we do a course and so for this particular course here, the course here is how to build a bot which finds

Course #1: Using AI to get a job (80 Days to Stay)

[2:20]Legitimate playlists. So what we do is we outline you know each module. These may change because what happens is you learn by doing. So, I'm going to learn how to make a bot that verifies legitimate uh playlist plus ones that have reach. Reach is fairly easy to to measure. Basically, has a lot of followers, but it can have a lot of followers because it's entirely fake and bots. There's a a music service called Soundcloud, which is just all bots and fake crap. Uh, and so a lot of this in this world is a lot of just fake stuff. It's it's like a lot of industries where it's just a lot of sleazy people doing a lot of sleazy things. But there is great music in the world and there's great artists in the world and there are people who make wonderful things that don't get hurt. So the purpose of this course is how do I make a bot? And there's, you know, hundreds of thousands or more playlists. There are 100 thousand probably songs a day or more that go on Spotify and other things. So this has to be done via, you know, AI and machine learning. Human being can't bet this level of

Course #2: AI music & the playlist discovery problem

[3:41]Information. You can't hire enough people to listen to a 100,000 songs a day and determine which ones are best. So these are algorithmic problems. It's an algorithmic problem that we can generate so much music so easily with all the tools. Uh, and we've been doing that for decades. For for you people who don't know about sort of the history of the music industry in 1990, not that much. I'd have to look up the stats, but maybe 25% or less music used autotunes. 2025. I'll have to look up the stats, but uh the stats, but I'll be surprised it's less than 99%. Most music that makes it on the Billboard list is highly highly processed, not

Just with AI, but with prone tools and all these other tools. So, you know, sort of tech and music has been around for decades. the issue of that these things like Spotify and Dieser and Apple Music or whatever getting hurt in a just a sea of of content is a fundamental issue since this is what I do. I teach AI and machine learning at Northeastern. I did my Harvard at postto in computational neurology. I have master's degrees spread visualization and NDA. So what we're going to do is we're going to look at this problem. It's an interesting problem to me personally because I make music. Um and so how do how do we figure out

Fake playlists, bots, and spam in the music ecosystem

[5:20]Let's assume you are a person and you like your music. You think your music is good. You listen to some music and it has a million followers and you go, "Oh my god, how did that thing get a million followers?" We're going to try to figure out how that happens. How How does this happen? How does some great music, some great music just sit there and other pretty mediocre music sort of gets on the top 100 Billboard. And so this is a research problem. So the first thing I did was I asked Jumbai and Flaud and Chachib to do research on that. I put this up in notes. So here's here's the notes that I have so far. So the first thing we have to figure out is well how can I access this data? That looks fairly easy. Uh for example, Spotify

Does have an API. I'm going to learn the API. I'm assuming they all do. Apple these are they all probably have APIs. So we'll dig into the API how to garbaging garbage out. First thing is we need to we need to get playlists. We can also use tools like uh Google search and for to search for people on and Twitter API. I'm sure Twitter and other people YouTube has an API and I'm sure there are people who do music playlist uh music channels on YouTube. So we'll go through all those and find all of those. So basically you will find sort of everything that has an API or scrapable looking for music playlists and what we want to do in effect is there are people who care about music so they care about their playlist they care it carefully because clearly a lot of

Framing it as an algorithmic + research problem

[7:01]Playlist which if you give me $15 you're on this playlist it should be fairly easy for Spotify to detect it. So, they probably do. So, you're probably actually hurting yourself if you get put on a crappy place. So, in order to figure out exactly what Spotify does, we need to do research. So, Spotify, if I email Spotify, hey, I'm a computer scientist engineer. Tell me how your algorithm works. They won't, but we can go publicly. We can also test it. We can just test things and see what works. And so, that's what we're going to do. And we also say does it make sense? So we're going to get into the detail. We're going to do research on Spotify and all the other algorithms. Exactly mathematically how do they work? What do they do exactly? But the idea is they look for signs. These

Are called metrics. A metric is something that you can measure of what they think is real engagement. So, for example, if you if you're on a playlist and then you save it, they probably indicate that in effect like a like. If you listen to it again and again and again and again, if you listen to it for two seconds and go to something else, they probably view well, you didn't like it. But I don't know exactly what they they do. So, they they look for a lot of metrics. How many different listeners? How diverse are the listeners? Do you complete it? You add it to your own playlist. You save it, etc., etc. And then what'll happen is you get paid to get on a playlist. It's unlikely that actual people do that and that may actually end up hurting you. You may get

APIs, scraping, and finding music playlists across platforms

[8:45]Some fake traffic for a while, but it's probably going to hurt you. But we I don't know for sure. So, this is a research branch. What we are going to do is we're going to research and build. And the other part of this project is teaching people how to make courses. The first thing you do in a make a course is well what are you going to teach? These courses are also very hands-on. Meaning most of this is going to be me doing it and then showing you what I did and talking about it. So the first step here is module one. It's sort of what is the problem. So that'll be the next module one will be what is the problem as let's develop

Other tools to make it easy and I'll go over these other tools in in other videos but these are tools to make it easy for you to create an explainer video and I'll go over this tool in a later one. I try to keep these short. So to summarize I we have a courses initiative at humanitarians AI. Humanitarians AI is a nonprofit which is a bridge program. People learn by doing with AI and machine learning from the time they graduate until they get a a real job. Often people tell me they've learned more in our program than they did in their master's program because they learn by doing. They build things. And then we're setting a new initiative called courses. And again, it'll be up

Metrics: saves, replays, completion, and real engagement

[10:16]To the volunteers to teach whatever they want to teach. There are already a 100 choices for them. They could teach music medicine, social media, creative, environmental, cult of AI, effective prompting, neurodyiverse learning, chat bots for nonprofits, regular machine learning, research methods, causal AI, interactive story, whatever they want. but they'll have to come up with a module every two weeks if you're going to do this as part of a volunteer and teach somebody something. The model is pretty simple. First, you do it, you learn it, and then you explain it. So, the expectation is and then you have a sort of a clear outline of what you're going to talk about. So, for example, if you decide to volunteer for air and marketing, well, this and then this and then this and then this and then this and then this and this. Of course, these things may change as

You as you develop something. Oh, there really should be something between this and this. So then you add it. But you don't know that until you do it. That's the number one thing. You learn by doing. So in order to teach people how to make courses in whatever related to AI, I'm going to do two courses. One is how do we use AI to find legitimate playlists? And this is a niatic problem because there's so many of these playlists and so much music every day that goes on these services that this is an algorithmic problem that it's something that has to be solved with algorithms. The other one is how to use a get a job effectively get a job. You can look at the 80 days to stay thing now but that's also going to be a course. So I'm going to create two courses. The courses are

Why pay-to-play playlists likely hurt more than they help

[12:01]Going to be me doing things sort of essentially video blogging about them and creating notebooks and code that actually does that. What's different with a course is all the courses will eventually have chat bots associated with them and they'll have learning materials. So what's different about a course and sort of just a YouTube playlist is there's going to be a GitHub associated with that course. that GitHub is going to have, you know, code. It's going to have examples. It's going to have work problems. They have a textbook. It may have a chatbot, but it's going to have other tools for people to go to to actually do. You learn by doing. So, that's it for this. This was my first uh courses on courses. I'm going to teach how to make courses by making two courses with the tools that I created to make courses. Okay, that's it. Take care.