Moonshot Thinking in AI Education: Exploring the Future with Bodo Hoenen of NOLEJ

future of AI and education. And that's not even all these other things to give us a little bit at your life story and how you got to the current moment at knowledge.

superpower. Because I kept thinking, Well, I don't know, if I'm approaching this right, let me ask for help. They may ask for help with this. And that whatever else that gave me a lot of great insight from people all around doing different things. But I think it also highlighted to me that collective intelligence is really powerful. It's collective intelligence, the use of that intelligence that is around you in the network is extremely powerful.

And so even though I would regard myself as never being the smartest person in the room, I often bit was able to create really smart solutions, because I was tapping into the collective intelligence of everyone around me. And so this work progressed, I built a few businesses, I was part of the product team that built this really cool telephone interpreting system for the UK government, which then generated $74 million in revenue every year really successful tool and

product was really cool. After that, I moved to the US and I created a learning management system. Again, with very little funding, we had a whole bunch of interns, and we relied on collective intelligence to build this product. Within three months, we had built the first MVP, within six months, we had totally pivoted, towards enterprise solutions. And then we sold the learning management system to Google. And all of this was really done super fast, using collective intelligence.

So it was there. At that time, I thought, well, let me take a step back, I'm going to start this company called Deaf RX, which is a nonprofit. And let me really experiment on this whole idea of tapping into collective intelligence. And I've been working on that. And suddenly my daughter becomes paralyzed, she within just a couple of hours, within five hours, she loses the use of her arms, her left part of the face starts to droop.

She's struggling to breathe, she can't use her legs are or arms anymore, she can't even speak. And turns out she had this rare polio like illness, and there was really no chance of recovery, there was very little chance of recovery. So we needed to figure out a solution. And so this is this just highlights a bit of how I tackled problems and where we're going with the other projects I'm working on. But we needed to figure out a

solution. We found out that building a brain controlled exoskeleton could help a paralyzed person move again. So it was picking up signals from the brain, converting it into electrical signals, and then using those electrical signals to control the exoskeleton. And this way, even though there was damaged merchant urine somewhere in the spine, we could bypass that damaged part of the circuit by using an exoskeleton. But how do we build that? Well, there's no course for how to build it.

You can't go and find a course for how to build it. But thankfully, there's a whole bunch of people you can you can go to to figure out small parts of this puzzle. Well, it took us thanks to the research and projects I've been working on with Def Rex. We had all these little experiments where We were working on, and we use those experiments. And it took us just seven months to build this brain controlled exoskeleton. And it was, it was such a beautiful

story. Because after building the exoskeleton, which already blew our doctors minds away, they thought it was totally impossible. But we were able to do it. But what was even more exciting was a few months afterwards, my daughter comes up to my wife and I in the kitchen, and we busy cooking, and she says mom and dad, guess what I can do? And she's not wearing the exoskeleton, and she starts to use our arms as if they were

totally fixed. Well, I mean, she had regained about 20 30% of the of the strength enough for her to pick up and move her arm, which I mean, that was, that was crazy. But I think it highlighted that you know what the stuff that I'm working on is actually pretty useful. We could use it to learn things out of the box that maybe someone has never designed a course for, well, maybe this is useful. Maybe we can build a business around this. And then we started to look at building care. What

can we build from this? And yeah, that's the backstory.

a business. And the business is this amazing collective intelligence based AI engine called knowledge that's spelled N O L. E. J, there are a lot of AI education companies right now people are starting them left and right. But knowledge is one of the most high profile ones. It's really interesting what it does, it's very powerful already, and it's out there. It's out there in the world. It's really about course, building, tell us about how knowledge works. What did you make from this experience?

challenge? How can we generate curriculum in real time, because that's a huge challenge. So we thought about that. And the way that we are building it is to have these three separate tools. The first tool is knowledge AI, which is now released, and tool two, and three, which I'll briefly explain, those are going to be released at the end, while

middle of 2024. So what are these tools, so tool, one is a knowledge AI, you can think of it as a tool that can met rapidly generate bite sized, elearning material from any source material. So you can upload a video, audio file, PDF, podcasts website, give it a website, give it you know, free text, you can copy and paste free text, if you want, it will then use that as the ground truth, and then generate a whole bunch of interactive educational

content from that. This allows us for example, with my daughter, and I, Hey, there's this web page that teaches us all about 3d printing. Well, it's I mean, it's difficult to read, and it's not really interactive, and all the rest of it, but I can throw this web page edit and suddenly comes out with interactive content that makes it easier for me to learn. That's what it is, knowledge AI at the moment is still version

one. So it's at that embarrassing version one stage, but it's already very powerful. So you just throw any documents at it, video, audio text, and within a few minutes, it'll generate these interactive packages, which are then plug and play ready for any learning management system. So no need to learn any new tools, it just plugs into existing workflow. So I think that's what makes it really powerful. People don't need to learn something new, they can just plug into what it is today.

transcript. So now you have a text transcript of that YouTube video, which is easy to search, you can search for keywords, you can search whatever you want. With that transcript, we then also produce a glossary of terms. What are the key ideas that are being mentioned within this video, let's generate that glossary of terms. But then let's let's look at the source material to see if there's a definition for that term. And let's use that if there's insufficient data in the source

material. Let's generate the definitions to so now you have a glossary of the terms as well as definitions. Okay, now, what's up next? Well, next is let's get some quiz questions. Let's let's generate some quiz questions. So we generate a whole bank of quiz questions, including multiple choice true and false fill in the blank, but even drag the word type of quiz questions, interactive quiz questions. So that's that. Now let's just create a summary too. So we create a summary of what this

content is all about. So we create a summary. But that's not all, because now we're looking at more interactive elements. So let's generate some flashcards to help you study, let's generate some concept cards to help you study, then let's make some gamified elements, like crosswords and find the word type games, then let's create some teaching materials on top of that, right, so let's help the teacher design a lesson plan and create a lesson plan around this content she's just

uploaded. So there's all of these different packages that we create. And that's just the tip of the iceberg, because we also looking at additional interactives teachers are looking for Well, I'd really like a whole packet of information that we can print out because we want to create, you know, an in person type of session. Well, we're doing that too. And then well, some other teachers want well, can you build a slide deck for us? That includes questions and quiz questions and branching logic?

Well, okay, let's do that, too. So we have all these suggestions. And then we say, well, I'm just uploading this PDF, can you auto generate a video from that? Okay, let's do that, too. But that's, that's all few months ahead in the future. But that's, that's where we're going. So imagine the ability, you just upload, whatever source material you

have. And one more point I want to just just to share on this is the approach we're using is quite different to a lot of other AI AI companies in education, where, if you know anything about generative AI, you know about the challenge of hallucinations, it's easy to lose innate incorrect information, we tackle that problem really differently.

Where you upload your source material, that source material becomes ground truth, we can't generate anything that is now contrary to that ground truth, you as an educator are now fully in control of what the AI generates. Because the generating of the ground truth is step one in the process. And every other process in this workflow will always come back to the ground truth, making sure that what is produced is accurate and consistent to what you want your lens to learn. So that's,

need to learn tomorrow. We don't know what problems they need to solve, we don't know what jobs are going to be available. And so us thinking of, let's glue together these Lego bricks, because we know exactly what students need is no longer something that we should be doing. So instead of gluing them together, and these monolithic models, let's keep them separate. But then what do we do with these Lego bricks. So that's where tool two comes in.

Instead of putting them in these monolithic models, let's map them out to a Google Maps for learning where every Lego brick is a concept that can be learned, which has an address on this Google Maps for learning. But this isn't a normal map, which is two or three dimensions. This map is hundreds of 1000s of dimensions D, this allows us to truly represent each Lego brick within the context of all the information around it. What does this allow

us to do? Well, it allows us to do for example, my five year old daughter, and if we were to look to build a brain controlled exoskeleton, there's no cause for it. But if we were to search this map, saying, hey, we want to build a brain controlled exoskeleton, it can now generate a map of our ignorance. Here are all the Lego bricks that lie between our current understanding, which in our case was zero, we knew we knew

nothing about exoskeletons. And here's the goal you want to get to well, in between those two points, here's your map of ignorance, all the Lego bricks of all the concepts you need to learn between there and your goal. So that's tool to tool three is, well, now you have a map of ignorance, which can be dynamically generated on the fly for any learning goal. How do

you go through that? Well, we don't want to give you just a recommended path saying, You know what, here's the recommended path through this, we want to facilitate you to explore that map, tapping into your intrinsic motivation, tapping into serendipity, because one of the big challenges and again, this comes up to my daughter and I, today, we're busily working on x, this was a plan right? X, we need to work on X, however, serendipity comes and life throws you a curveball. And suddenly we're in

the position. Oh my goodness, we now have the opportunity to speak to this person who happens to be an expert at y. Now, in the linear plan. We weren't supposed to tackle problem y until seven months from now, but we have this opportunity today. This allows you to tap into those those serendipitous opportunities, but also so we think of two or three as kind of like a GPS to guide you through that map of ignorance. But the GPS does something quite unique.

It not only guides you to those Lego bricks, that content, we also recognize that learning or that content is king content can only get you 20% through the learning experience. A remainder of the learning experience is your interactions with other experts, with learners with teachers with mentors with serendipity All right. And so the GPS guides you through to those connections as well. So I think that gives a quick overview of what that is,

pretty old idea. A lot of people have tried it, a lot of people who've done it, myself, I've done various different prototypes of this, the floor in many people's thinking was to, well, let's structure it the way education is structured today, in these flat two dimensional taxonomies can or let's be creative, let's think of it in three dimensions. Wow, you know, three dimensions that have to, that's cool, too. I mean, visually, it's, it's fine for

us. However, even in three dimensions, it's impossible to really structure the data in a way that truly represents knowledge, I think of it has, you know, you're busy learning about ancient Egyptian history, and you're busy learning about the pyramids, well, that doesn't just have related information about history that also comes into philosophy and to mathematics, and to geometry, and all these different, even astrophysics, and the alignment of stars know all of these

different concepts. You can't structure that on a two dimensional taxonomy, you need a more suitable data structure. And so machine learning, the way that large language models work is really a good analogy for how these maps look and work, you map out concepts in 1000s, hundreds of 1000s of dimensions deep, and that suddenly gives you the ability to structure, education, or learning and information, the way is truly structured. So what I'm

learn about that thing. But when I'm done with that thing, it says, You just finished learning about that. There are a whole lot of other things connected to this at various levels of connection weights, you know? And yes, what do you want to learn next? Where do you want to go with this? Why are you learning this? Are you learning it to something new? Are you learning it to master a subject, and it'll recommend or show a lot of options, so where to go next,

LMS. And yes, one thing that's been really notable about what you've been doing, especially compared to some of the other AI companies that are just sort of getting off the ground, you already have secured partnerships with open AI, you already are working within Google Classroom, you're integrated there. I noticed you're there in instructors AI app platform, which they're just, you know, launching now that partners are there. So this

is really interesting. You're obviously a collective knowledge, you know, dizziest, tell us about these partnerships and how you've working to integrate knowledge into other edtech and make it more findable.

platforms. And that what is beautiful about that is, let's say Canvas at the moment, if someone would need to build a course on Canvas, it's gonna take them 2030 hours to build all of the interactive manually and do all this work. Well, now imagine they have a button that says automate. And they click that button, they upload their documents, they go grab a cup of coffee, and they come back and

it's all done for them. That's the type of value proposition we are offering to these companies saying, you know, what, you keep the platform, you just help you automate and structure that far better. And I think that's a huge value proposition and works really well. And so yeah, we are still early stage startup. But we have noticed and recognize that early on. This is the way to do it. These big platforms have tremendous distribution. They are trusted, which is great. So let's leverage that.

Let's help them make their lives easier and make their products better. And we'll just be in the background.

into a course. And one aspect of this generative AI that has been really fascinating to me is that you can take knowledge that's been buried in academic journals for decades, and actually find it and pull it out and use it and connect it to other things. I'm curious how you see, you know, obviously, an individual could put a PDF in knowledge and get an output, which is incredible. But is there a potential to put you know, 10,000 PDFs that are the entire history of the nature journal?

into it? Yeah. Tell us about that.

on this educational material. I think it's a win win situation. So we aren't doing that yet. But that is the intention is, let's be create a tool that can map out that educational content, creating these Lego bricks that anybody can use.

business. In many cases, they're really good business ideas. So that was interesting. And I wonder if that is another incredible corpus of material that you might want to dig through for various reasons. If you're looking to do something amazing and new that nobody's ever done before? Well,

doing and other patterns. While this is an exercise that could easily take me weeks and weeks and weeks, because I need to search for the patterns, and then look at the similarities and differences. Instead, I go to my digital twin saying and I download some relevant patterns, I upload them and say, Hey, can you list out all the similarities and differences between these patterns and boom, you know, two minutes later, or 20 seconds later, I should say,

there, it's done. And so what would have taken me easily, five, six weeks was done in 20 seconds. And that's crazy.

themselves as an AI bot. They'll learn about the difference between Long and short term memory, they'll learn about human feedback within the system, learn about how AI is Think through problems. So I thought, you know, this would be a cool project. And we had three goals that we kicked off with. One was can we teach an AI to answer questions that we provide any information about, for example, Hey, what's my favorite book? What's my favorite pets? Who are my best friends? So

basic information? So that was goal one, can we do that? We were able to achieve that within the first two weeks. So that was a pretty easy one. But that was goal one. Goal two is can we ask the AI questions we haven't trained it on, but have it answered the questions as if it was us. So it incorporates the information it knows about us and then makes a suggestion. So for one example is it knows the

books that my son likes? And we asked okay, you know what books I like right recommend some other books, which you think I would like. So now it needs to analyze the books that my son has liked. figure out, Okay, what does it like about these books? The characters that story progression? What level of book? Is it? Right? He's a fifth grader. Okay, so let me look at fifth grade content. And then let me recommend some other books. Turns out that the recommendation is brilliant.

Right? So that was a goal to goal three was, throughout this process? Could we learn more about ourselves? Could we learn about who we are as individuals? This was more of a subjective goal. But could we? And the answer is hell yes. Like we've learned really deeply about ourselves. There's this one video where we explore the difference, how of emotions and how emotions can change how we answer things and how we perceive the world. So one example, we have these AIs,

right? They have the exact same information, but we give them different emotions, with one. In one test, we gave them a really positive emotion. And then in the very next test, we gave them very negative emotion. We asked them the very same question, and they provide us two separate types of answers. Why is that? This started to help us explore these type of questions with the kids. Well, anyways, let me end the story. So we build these

digital twins. And now as a final experiment, we say, okay, can these digital twins be good tutors, to my kids? You know, they're about to go back to school? Can they help them answer questions and learn about the things they need to learn and do their homework and help them through these problems sets and problem questions? Oh, my goodness, are these digital twins able to help so much better than just a generic Chatbot? Why? Because the twins

know everything about my son. So sorry, I must emphasize that all of this is completely private, right? We're not sharing this private information with anybody. So that was one of the early requirements is that we want this to be totally private. Anyway, so So yeah, we're super excited about it. I mean,

yeah, exactly. It just really gets closer to that sort of that blooms two sigma idea. And then we've sort of ever heard, I mean, you know, it's not just a tutor that knows you, it's a tutor that knows a lot about you. That's really, that you know, everything about you ideally, that idea of the emotional filters, they learning social and emotional intelligence by Yeah, absolutely. Yeah. So, so many

things in there. You know, one thing that I love about what you're doing there, this is just on an almost like a publicity level is that I've been playing with these AI tools since they started, they are mind blowing. And it's been so sad to me that I think that one of the prevailing narratives in schools is cheating. It's cheating tool.

So what do we do about that, it's like, seems like the most closed minded thing you can imagine the type of project that you just said, a six week project, create a functional chatbot, that understands that. It's like, that's what AI enables in education that your hand up project that should be happening in schools, and there just aren't enough exemplars and example, the positive, you know, blow your mind in a positive

way. Visions, people have these stories about how AI might take over the world, or they're worried about these crises. But there are these incredible positive futures that can be seen you obviously are right on this path. And I've been this project, I think, just really is an amazing example of that.

environments. And they know the struggles they have within these environments. And so they they came up to me and said, You know what, we now know how to build these twins. We want to teach all kids everywhere, how to build that, how to build one themselves. And I think this is the beauty of it. They're not interested in in the financial rewards. This is all going to be nonprofit, open source type

projects. But imagine a world where in a year from now, anyone on the planet can build their own digital twin, a tutor that is super powerful. And it's the narrative is like, now we have an AI company built by kids for kids that is completely private. No information is being shared to any big tech companies. Like this is a beautiful narrative. It's a beautiful story, especially in comparison With all the negative AI stories out there,

synthetic voice of my kids. And then we used various different methods to create a digital avatar of them, and then to animate that. And there's lots of tools out there that helps you do that. But it's great. It gives that additional layer of realism to the conversation.

really interesting idea. So I want to ask you about sort of speculative, you're obviously already living it, but about the speculative future of this AI world. The two funny things that are running through my mind, as I hear you talk about some of these projects. One is this digital twin project reminds me so much of the famous Calvin and Hobbes sequence where he builds a doppelganger machine, right, he builds all runs of himself, and he's like, great, these clones will do my homework for

me. And then the cons are like, I don't do homework. I'm you. And it's, it's, it's amazing. But it also is interesting, the idea of, you know, you can build this digital twin for yourself, you could also build a digital version of yourself twin and then have it tutor other people, right? You could have Stephen Hawking could have made a digital twin that for the next 100 years, you can ask Stephen Hawking to tell you what's going on, there's so much to unpack

other is going to say. And I don't even know what they're going to say. But I just witnessed where these conversations would go. And it's ended up speaking about what they discovered and how they discovered it. So that is great. So I recorded that. And then I took that video, I pushed it through knowledge, AI, and now it generated a whole course around their conversation. So then I thought, Okay, let me take this experiment a little

further. Let me ask Einstein to now present a short lecture to high schoolers about his work about the theory of general relativity. And I just asked him to do that. He provides a lecture, really solid lecture. And I record that I again, push it through to knowledge, AI, knowledge I creates a whole lesson from it includes the transcript, the quiz questions, all of this. And it's, it was fascinating. So absolutely, yes, the digital twins can be applied

to anything. But also think of it as a being applied to your company, you can create a digital twin of your company, you can create a digital twin of a product, you can create a digital twin, and this is what my kids are doing. They are creating a digital twin of the eighth grade and in fifth grade curriculum. Now they can ask that curricula, hey, I'm learning about this. Can you provide me a lecture? Can you help teach this concept to me because I didn't understand it

at school. Now, their AI can communicate with this other AI and create a lesson from that. So it's

year in less than a year. Yeah, the generative AI part of it, of course, tell us a little bit about what you think is coming in the next, you know, three 610 What is coming in this world? It's so interesting.

carriages lighter? The whole industry was focused on this part. It'll a problem. And yet there was one or two individuals, Henry Ford is a good example, who thought, well, maybe we don't need half faster horses, maybe we don't need to worry about a horse manure in New York City. Let's think about the problem a little differently. Let's build something different. I think

we're at that stage. At the moment, if I look at the industry, 99.99% of the educational industry is looking to apply AI to somehow optimize the existing processes. They tried to figure out what to do with horse manure. What I'm suggesting is we don't need to worry about the horse manure. We have this educational system, which was designed 150 years ago, it's now time to let that go. And we have an opportunity to think radically different

about education. Imagine providing a totally personalized learning experience to every student. That wasn't possible 11 months ago, it is possible today. Imagine providing a real time curriculum as they go. That's not wasn't possible a few months ago, it is possible today, we can rethink education from the ground up. We can also think about education from the perspective of the learner, what is intrinsically motivating for the learner to learn. It's not literacy and numeracy outside of

any context. It's, hey, I want to build a business, I want to become a bicycle mechanic, I want to solve problem X, I want to solve problem why I want to do something. Now suddenly, we can provide contextual education that taps into your learners intrinsic motivation, that again, was not possible before.

What does that mean for educational systems, that means we no longer need to think about traditional assessments and traditional courses and think of grouping kids in random eight limited cohorts, all of this can go out of the window. So yeah, that's what I think.