Just in Time
Oct 08, 2024
Transcript automatically generated by Google Pixel for this presentation
[Speaker 1]
But we waiting baited breath for Okay, thank you. Hi everyone. I'm Stephen Downes. It's a pleasure to be here with you today. Thanks for coming to this session. I see, there's 50 of you. It makes me feel relevant.
[Speaker 2]
I
[Speaker 1]
Appreciate that. I got a packed agenda today. Um so without further Ado I'm going to share the screen. Hopefully my Head and Shoulders still shows up when I do that. Um, screen, one share. All right. And you should now be seeing the cover to my slides. Um, Yes, you are. You are seeing the cover to my slides. Um, So this talk is called just in time. Creating Dynamic. Open learning resources. Using gai uh, just a caveat off the top. It won't be 100 g. A i So, um, It'll be other types of AI as well, but but you get the idea, right? Um, So, For all of history open education resources which I'm now calling open learning resources to kind of shift the field a bit have been about static learning resources as the image suggests kind of like books on digital Um, you know, and people have even used the language of books talking about publishing them. You know having libraries over open education, resources Etc. Lots of problems with this. These problems drive up the cost and make it difficult for us to use. Open learning resources in a scalable way. There they become out of date. The technology becomes obsolete. Remember flash everyone uh Finding them is a problem. Uh, localizing them is a problem. Making sure they're accessible as a problem. There's the cost, the production overhead and then probably the biggest issue, the general failure to adopt and reuse the learning materials at all, which has been well documented. So, These are well-known issues. I'm I'm not going to discuss them at any length. Uh, but Proposing and talking about kind of a different way of thinking about open learning resources today. Uh, which is the dynamic open learning resources. I'm gonna go to the fly. Brain simulator, maybe you, you saw this present or this, this paper that was published just this week on nature. This week, maybe late last week. Uh people at Princeton and others built a computational model of the fly visual system. It's the theme of this slideshow. You see the image of it in the upper left. Um, consistent with available connectome data. The connectome is the list of all the neurons and all the connections in a brain. In this case, a fruit fly brain And uh, Has what they call biophysically, plausible. Neurodynamics, etc, etc. The details don't matter. Well, I mean the details matter a lot but not for this presentation. It's an incredible piece of work. I'm just blown away by, I love this piece of work. Now. As part of this project, what they built was something called the flywire connectome data Explorer. And, Here it is. Um, And, uh, you, you can go into it, have a look inside the connectome. And here's all the people who are involved in it, The brain, oops. Female. I don't know. Uh, brain and nerve core. Etc. Um, And it's like this online simulator. Oh, here we go. This, I clicked on the wrong link. Here we go. This is the online simulator All right, you can go into it. You can see stats about it. You can see some annotations. Uh I I like you know, I like the network graphs because you know I'm kind of a network person. I've always been a network persons on this and this stuff, just, you know, makes me happy. So, All
[Speaker 3]
Right.
[Speaker 1]
So we see that It's a huge Advance. And what I really love about this project is that it wasn't just some scientist guy working in a lab, wasn't even, just a bunch of professors working in Labs connected across the country. In fact, it was I think it was like, something like, 120, 125, different partners in the collaboration.
[Speaker 2]
But
[Speaker 1]
They are also supported with artificial intelligence, and there were also supported with something like 50, 000 Gamers. Who went into?
[Speaker 2]
The
[Speaker 1]
Visualization that you just saw and worked with the data validated the connections, validated the predictions. Etc, did a whole bunch of stuff with this that actually made this project possible and you can read all about it. Here's the article from Princeton, it's very Pro Princeton, but it's a good article. So, here's the eyewire game. Uh, this is a project from 10 years ago that inspired Flywire, I wear. Yeah, yeah, yeah. And a lot of this happened before, AI really came into the picture. And again, though they did the same sort of thing, they mapped neurons in a mouse retina. Painstakingly assembled, millions of tiny, little puzzles to solve the 3D structure for each Mouse. Neuron Fabulous, interesting. Craig I I wish I had been part of that. I Can Only Imagine what it was like to do that. I can only imagine what it feels like today. Having been a part of that. Been a part of Flywire. Played this game helped develop this connectome and been part of this release. I'd feel like you know, especially if I was a student, I'd feel like my life work is complete and I haven't even graduated yet. And that leads us to the idea of dynamic, open learning.
[Speaker 2]
Right.
[Speaker 1]
Dynamic open learning to be contrasted with static, open learning, which is what we saw in the first slide with the open educational, resources, stacked up like books.
[Speaker 2]
Um,
[Speaker 1]
Etc. Uh, Dynamic learning. Dynamic, open learning. Is. Like what they did with the fly. Just realizing that my wife yesterday watched the movie, The Fly. That's kind of an interesting image. Um, And so accidentally, I watched part of the movie, The Fly and I don't watch movies like The Fly anyhow, totally off topic. Um, So, this is From a quote, from a page from the University of Victoria. Dynamic learning is fueled by research inspired insights and hands-on experience. The interplay of ideas and action gives our students a powerful edge of expertise. To be proof, right? You know, I mean the the research needs to be done. Uh, interestingly, when I looked up Dynamic, open learning because it's the topic of this presentation. I found almost nothing online looking for this phrase. I won't take credit for it because clearly, it did exist before this presentation. But, uh, this does seem to me like an open area of Investigation.
[Speaker 2]
So,
[Speaker 1]
Part, did I skip a slide? Nope. Okay. Part of dynamic open learning and part of what makes it challenging and what makes it interesting. Is that it's based in community connections. And this is, this is from believe it, or not Calgary, Economic Development. Like I say, I did a search for this and this is one of the things that turned up, but they they get the idea. Um, you know, uh, Dynamic open Learning System will require active involvement from key members of The Learning System, namely Learners, duh, Educators and employers and I would add Pretty much everybody else in the community. Um, you know you look at the fly project, we're looking at scientists, we're looking at computer technicians, we're looking at biologists, neurophysicists, Etc. Huge collaboration across a wide community. And this would be the case for pretty much any kind of Dynamic open learning that you would do. What are the characteristics?
[Speaker 2]
Of
[Speaker 1]
Dynamic, open learning. And here, I just want to give credit because It's kind of fun to do that. Uh, this is Cindy underhill from open. Ed 2012. Can you believe it? Diagramming something from Alan Levine and it turned up in my search. For. Dynamic, hope and learning. So, Um, Basically having looked at different discussions of dynamic open learning. Here are the characteristics one uses? Rio data. That's really important. Yeah, you could fake the data but now now you've got a case of static open learning You know, the best dynamic open learning to my mind re uses real data, transparent processes. So you can see the data coming in. You can see what we're doing with the data, we can see whatever manipulations we're doing, we can see how we're making decisions in the project, Etc collaborative and Cooperative, most people just say collaborative, but I think we need to be clear that both collaboration and cooperation are required collaboration, is where everyone's working toward Or the same goal cooperation is where people are working toward different goals but they find exchanges of mutual value. It's interactive. It's constructive you you know it's more than just playing a game. You're actually doing something with purpose to create something and IT addresses real world problems. Now, this is hard. Uh, it's really hard and that's probably why we haven't seen very much in the way of open. Dynamic learning. Uh, it's really hard because It's hard to set up the data. It's hard to set up the interaction with the community. Uh, it's hard to do the conversation the coming to conclusions, Etc.
[Speaker 2]
Um,
[Speaker 1]
I found a project called the edgy caliber Condor project working toward Community Driven development of educational material,
[Speaker 2]
And,
[Speaker 1]
Uh, it captured a lot of the issues involved with creating Dynamic open learning, for example, transactive memory, you've got this community going. And it's been going for a year or so and then somebody new joins the community. Well, they've missed the year or so of stuff that happened before they joined. How do you deal with that? That's that's an example of the sort of thing shared models. Now, Uh, here. Well, in this article, they're actually talking about shared mental models. I wanted to take that away from mental models and just make that shared models in the community, for example, ontologies that you can work with Dynamic Network form formation. You can't have 50 000 people for example all working together in a single environment, that's craziness. So you need to be able to subdivide that build smaller groups, focus on specific projects. All of that needs to be done hard to do people who are organizing this conference. Can they test that? It's hard to get all of that set up. And then how do you make decisions again hard to do? So, There are a lot of roles for artificial intelligence. In the creation of dynamic, open learning. Now, I know, I know. Everybody's got criticisms of artificial intelligence these days, and I can imagine the other presentations going on at this conference and we'll talk a bit about that at the end. But And I also want to say it's not just generative AI, we're not just talking about some AI that's going to create some content for us, but that is one of the rules. Uh, data collection, and interpretation is a rule transaction, logging and summarization.
[Speaker 2]
Um,
[Speaker 1]
You know, I had as I was talking, I had somebody request the little thing pops up on the screen that I turn on the AI tools, right? We're working in AI supported environment as we work here. In this Zoom environment, it will summarize the content to highlight key points or maybe it'll translate text Etc. Dynamic Network formation another role for AI as I mentioned, and consensus identification. So, Oh, that seems very fairy right. Maybe someday in the future, we can do this. Well, part of my argument today is we can do this. Now, we can make all of these things happen. Now these tools exist, I'm not gonna go into every one of these tools. Uh because I just don't have anything like enough time to do that but here's one type one set of tools data, collection and interpretation. This is a well-established set of tools. That. That exist have existed for a while. Um, Automatic image. Analysis is one that I've been playing with recently see if I can just pop into the right place. This link might not work for you if you're trying it yourself.
[Speaker 2]
Um,
[Speaker 1]
Because it goes into my own account, so, sorry about that. But you'll be able to, at least see this on the screen. So, Uh I went into the Google AI School, AI Studio some in Pokeball Obsession that's stupid. Uh I mean I'm not saying Pokeballs are stupid, but it's a pretty trivial example, right? Um, so I took one of my actual photos that I, I was in Morocco, in Rabat, uh, for a conference recently. So, I took an actual photo, I asked it to write a short engaging blog post. Based on this picture, it did that. I wanted more like a caption, it did that. But you know, it's you know,
I didn't like it, I wanted more, I wanted more detail and less flowery language so I tried to you know, if possible providing some scientific geographical or historical background
[Speaker 2]
Tried
[Speaker 1]
For that now. It had some issues, um, it called the five star sided Moroccan star, the star of David, So, not really right. And the sort of thing that could cause well, an international incident if I published it. So yeah, it's still not quite ready for prime time. I'll try something a little more neutral.
[Speaker 2]
Uh,
[Speaker 1]
Here are some flowers. So, we're back to robot.
[Speaker 2]
Uh,
[Speaker 1]
Here's here's a picture of a guard that I photographed. It's describing it. I really wanted to tell me about it. There's the star of David reference that I mentioned. Now, what's really interesting is that this is real time. Um, You know. So, I'll take my My prompt. So here's my prompt. Now, I've been You notice my prompt got a bit more and a bit more descriptive, you know, specific.
[Speaker 2]
So there's
[Speaker 1]
My prompt right, a short detailed and informative caption. I'm going to add an image to this. Uh, so I'll actually upload it from my computer.
[Speaker 2]
Uh,
[Speaker 1]
Let's see.
Oh there. That's a good subject. All right, so here's the image. I actually just tried this. I didn't pre-select this image because I'm living dangerously. All right, there it is. So we can all recognize what it is. Let's see what Google thinks. Now, the important thing here is that it has published, okay, we're up to okay, 6.8 seconds. I wanted it now but it took 6.8 seconds. Okay. An abandoned wasp nest made of papery, material hanged, from a branch of a tree, likely in a temperate climate dead on so far. It's composed of layers of chewed wood fibers, which the Wasp used to build and that's the nest is now empty, nice deduction. And the Wasps have likely flown away for in the winter because there's no leaves on the tree. The nest may be inhabited by other insects or small animals, Dana
[Speaker 2]
Um,
[Speaker 1]
And no. No, International incidents. That's the sort of thing. Now, that's a small thing, right? It's not a big thing, but as part of a big project having that capacity at your fingertips is great. You simple examples and Uh, you know, as part of your project, people go out and do a survey of the, uh, of the ecology in their neighborhood. I did that when I was in high school, We went up behind this High School, went down to the creek collected samples, water samples, plant samples did a report. Etc. Now they can do this and they can actually find out what they were talking about or what they were looking at rather I should say.
Oops. Oh, dang. That's not what I intended to do. This is what I intended to do. Sure, those tools. Back to my presentation. There we go. Dynamic content creation. Um these links do work. Uh there are some things that I did over the last little while. Perplexity is an AI, I did a research uh a bit of a research study on connectivism. Basically I asked it to give me resources on connectivism. These are the results here. Uh, automated summarizing of podcasts. Now, you've many of you will have seen this, but for those of you who haven't, I really do have to show you. So Um, I got this from Darcy, Norman. Um, and he had notebook, LM summarize his dissertation. So I tried to do the same thing with a paper that I wrote on connectivism a while back.
Now, I'm not sure. Were you able to hear that? So I'm going to give me a thumbs up or a thumbs down. Ah, that's too bad. There's probably a way to make that work. But I don't know what it is.
Okay, where's share sound?
All right, let's go into.
Uh, it's so stupid sometimes.
Not seeing it. Oh, there it is. Not in the lower left hand side. All right.
Good now.
Let's see if Peter Finch laughing at it, because
All right. It goes on like that for almost 12 minutes now. As the person who wrote the paper, I I can say a couple things. First of all, uh, the summary was accurate. Um, Second, the summary actually used some examples. Good examples that I hadn't actually used in the paper. So,
[Speaker 2]
I
[Speaker 1]
Was pretty impressed with the result. Now, a lot of people are calling this quite rightly a parlor game because all of these podcasts sound the same. So they're obviously using the same. Same voices, a lot of the same speech generation algorithms interpretive algorithms, Etc to generate this. It is a parallel game. Another thing is that they've been getting worse as they've been getting more popular, Um, if I try to do the
[Speaker 2]
Same
[Speaker 1]
Connectivism one. Now it might not be nearly as good. I did see reports where people did try it on the same material over time and it got worse and worse and worse. So It's not 100 great reliable technology yet. But yeah, who cares, right? For the point that we're making, who cares? Um, it's the sort of thing that you could use especially once it gets good and there's more variety and more speaking tones Etc. To summarize reports meetings data. Remember we just looked at data. Uh, in a fairly intuitive way.
[Speaker 2]
Um,
[Speaker 1]
Automated course, creation. We're talking Dynamic, learning resources. I understand that. Um but part of the whole concept here is, or what I want to say is that Uh, because we don't need to create static resources anymore because we can create Static resources on the Fly as we need them. And so this was a simple little experiments. I did I prompted chat GPT uh,
[Speaker 2]
To
[Speaker 1]
Please write me a 100 page textbook on logic. That was my prompt. So so, so much for prompt engineering. Um, And this is what it produced for me.
[Speaker 2]
Um,
[Speaker 1]
Now, I could go through each of these topics and say, you know, please write me a chapter on such and such and here's a sample chapter it gave me on propositional logic. Um, now this is stuff. I taught I used to teach this in University dead on accurate. Um, if I spent the whole day, I could produce my 100 page logic textbook, no problem. So, Why would it be necessary for me to create a 100 page logic. Textbook wrap it up as an oer. You know, package it put it in the library, Etc. Whenever whenever when I can create my own logic textbook, localize it for my own needs, put it in the language. I want tailor the examples. Give me a little podcast of each section as we've just seen. And do that in a day. And that's now I could do that in a day now, right? Um, Probably, I could do, you know, give it a bit of time, I can do it in 5-10 minutes and in the future. So, Dynamic content creation. Translation logging and summarization. Uh, we don't have an auto translator going at the moment. Although I have seen them in theory, we could Um, I do know that translations are good enough now that we can get a pretty good translation of say, this talk in real time. Um, on my screen and therefore probably on yours. I actually see the text as I'm saying, it being printed on the screen, so transcription is being generated.
I'm also. Whoops, you can't see that. This is the, the reason I bought this phone which you should be seeing on your screen now. Uh, because it's recording this talk and generating a real-time transcription. This is how I've been generating transcriptions of my talks for the last few years. Um,
All the sorts of stuff and and chat GPT itself. Um, is in AI mode doing all of these things that I described and of course, uh, we can use an application like quillbot or whatever there's dozens of them out there. Now the hard part is finding one that won't rip you off. Um, but you know, we can summarize texts no problem. So I remember the problem. Somebody coming along a year after our project started, Well, they can get caught up pretty quickly and we don't need to have somebody stop. Get all the content together and give them a training session or anything like that. They just use some AI tools and bring themselves up to speed pretty much on their own. So we'd need some scaffolding and support obviously, but basically pretty much on their own. Dynamic Network formation. Uh again, there's a bunch of stuff about that. Um, Up to and including AI? Applications that join in a chat with you people who use Discord or even Reddit know all about AIS jumping into the conversation. Consensus building. Uh, you have AI project management. I created a little project on a service called Monday. Which is, I don't know. And so here, here it is. Oh,
Silly thing. It forgot where I was. So, Timeline. Uh, she's the kanban. About. B, c. And we'll get started. And here's my AI supported project management space. For those of you who've done project management, you know how much of a pain this is. Here we have AIS. Basically going to do pretty much everything. Your project manager who took that whole project management Institute course or set of courses had to learn how to do. We have an AI application doing it for us if you're a project manager. I'm so sorry. Um, Collaborative authoring support. Collaborative intelligence. All of these things are supported by AI. So, um, It's 11 30. So I'm going to stop here, but I just want to point out. I know that there are objections to the AI Uh, does AI improve access to learning? Well, I made a logic text in a day. Uh, how can that not improve access to learning, right? That could print it out on paper if I wanted. But the main thing is, uh, I've created this text is AI, an environmental nightmare if you run the numbers, it actually uses less energy. Our daily cup of coffee. So you have to have a sense of proportion here when you're talking about AI being an environmental nightmare, Does a does generative AI misrepresent members of marginalized communities? Yes, it does. Um, and that tells us that we need better data to train our AI than Twitter posts. Obviously, but the only way to do that is to create and share that data and I've written elsewhere about the need for us to create and share representative data in order to help train AI properly. Does AI steal content? No, it does not. It analyzes content looks at the order of words and uses the order of words in order to predict what the best answer is to a question or prompt. When you create a piece of content, your original expression is protected. Absolutely. And you have copyright over that but you do not have copyright over basic elements of the language. You do not have copyright. That you use. And so there's a very good argument, and I would make that argument AI is not stealing content, not anymore than a person who reads a book and learns from the book and does it violate the spirit of Open Access? Well, look, if it's all run by large corporations for the pure pursuit of profit, creating barriers and making it more difficult for us to communicate with ourselves. Then yeah, it violates the spirit of open access. The companies like open A Have done, especially recently violate the spirit of Open Access. But the response of the community, can't be no, no, no AI is bad. Uh, AI is nothing but mathematics. Right, and end language but mostly mathematics. And, We you know it's mathematics isn't it's something that inherently violates the spirit of Open Access, we have the opportunity In the open, Learning Community to use AI in the way I've described here and make great leaps forward. In Open Access. Uh, it's it's an incredibly powerful tool if we leave it to only the commercial Enterprises, then we'll all pay the price for that 1133. That's my time we have seven minutes for questions. Something like a record for me, so I'll take those questions now.
Oh, and look at the chat, he's just filled with questions.
It's fine with me.
I talk to them about it by talking about how I use Ai and how AI gives me capacities that I didn't have before. Um, I'm writing a little application. Um, and uh, Here it is. All right, and Uh, this code, I'm writing this code, the AI isn't writing this code but the AI is teaching me how to write this code. Oh, I want to share it and nobody saw it. It's silly. So yeah, here here, here's my application that I'm writing 10, JavaScript. I'm not really good at JavaScript and it's it's a massive a language to work with. Um, I'm writing the application. I'm designing it. The AI is helping me. Its suggestions are often wrong. But its suggestions are right enough that I can look at it and figure out. The right way to do it. So AI helps me a lot. Uh, if they still believe AI is bad, then that's their belief and I'm fine with it. Uh, but now I have a tool that they don't. And and that's all I can say about it.
Uh, in the one that I just produced. No. Uh it was dead on accurate um and the reason for that is there's a lot of material in the data that the AI sampled that would be saying exactly the same thing. You know it's it's not that there are many ways of doing propositional logic. There's basic, it's propositional logic. You basically one way of doing it, That chapter was accurate. But in general. Um, You know right now generative AI produces a lot of factual errors. And the reason for that is, is that it is not an encyclopedia tool. Uh, it is a language interface. So it speaks language really well as many people have pointed out it has no idea what it's talking about has no base of knowledge. You know, you you can create, you can pre-program it with some content, using a process called retrieval augmented generation or rag, where you give it a body of text and say use this body of text. When you're generating your answers and this can be very large bodies of text and then it will be accurate to that text, go beyond that, and it won't be accurate. Yeah, there are issues with accuracy because of the way. The tool is designed. It's not The problem with AI, it's just people expecting the tool to do something. It wasn't designed to do.
Beverly Gibson says to everyone, we are nothing more than the sum of the information. We've consumed Great comment. I love that. Except I would say uh we are nothing more than some of the experiences that we've had. There's a difference between information and experiences. And the experiences are the raw data. Information is an interpretation of that raw data. Um, But I still really like the. I I like the sentiment. How are you acknowledging the use of tools? In what you're creating. Everything I've created with AI. Um, I've written a post and said, look what I created with AI. Uh, basically. Um, And and, you know, that that's it, right? You say, you know, this was created with the assistance of, or I used AI. Here's the prompt that I used. Uh, I've always posted the prompt for the software them developing. I've got a complete log of all of the interactions that I've had with chatg Pete, chat GPT, and unless something breaks dramatically my intention is to make that chat log available as well. But I want to be careful. Harper Friedman asks, do you think the term A.I gives a false understanding of what A.I actually is? Um, Yes, and no. Um, I want to be careful here because there were some subtleties. So the first simple answer is. Okay, we do not yet, have artificial intelligence, I think everybody understands that, right? Like I said earlier, it's all mathematics. And the the sum of intelligence isn't encompassed by mathematics. I think there may be some other stuff going on. Who knows. Right. Um, More to the point. Uh when the term AI was first coined in the 50s or whenever we were nowhere close to having artificial intelligence, right? The term AI represents an aspiration. Not an accomplishment. We would like to use our tools to get to artificial intelligence You know, so it's sort of like, you know, the space program. All right, the space program was called that even before they ever shot their first rocket. All right, the moon mission was called that before Neil Armstrong ever set foot on the moon and nobody had a problem with that. Nobody said, well, you can't call it the moon missing because nobody's ever actually stepped on the moon. Well, no, But we intended to and we did. So, And that that's basically. My answer to that. But also I want to say, you know, intelligence is one of those Concepts that's really tricky. And the same sort of processes are used by the computer as are used by human brains as massively, demonstrated by the fruit fly connectome. So it's not just aspirational. It's also saying we're trying to do it in kind of the same way human intelligence works as well. So acts 11 40, sorry, go ahead.
Yeah, I'm gonna cancel or I'm gonna log off because I want to see the next session too. Um, Thank you, everyone.
Today: 7 Total: 224 [Share]
