Introduction

And we all are some kind of a victim of the system.

The Vision of Modern Education

This is more of a visionary presentation. Some parts of it are things which are real right now. Some parts of it are things which I and a group of experts I work with think are we can actually do quite soon.

Education, AI, and Intelligence Augmentation

So the topic is education, artificial intelligence, and IA means intelligence augmentation. This is the thing we do with the machines. This thing actually helps us, let's say, calculate. So that's intelligence augmentation.

Nobody calculates and multiplies three digit numbers. by themselves in their head, we use a calculator.

And we'll talk about opportunities and dangers in education.

The Need for Educational Reform

So really the question is, why don't we have many Einsteins now that we have so many people globally? And one of the reasons, I'm not saying it's the only reason, but one of the reasons is the education system. because it's quite universal and produces people for factories from the Victorian age. Well, that's what it was designed for.

Historical Success of One-on-One Tutoring

But if you look in history, pretty much everyone worth mentioning had a number of tutors. So the guy in the picture, that's Alexander the Great, and his tutor was Aristotle.

Then we have Marcus Aurelius. He starts his book Meditations by thanking to his tutors. He had 17 tutors.

Einstein had what's probably the best name of a tutor, Max Talmud. He taught him maths and other stuff. So one-on-one tutoring historically has proven to be great.

Origins and Evolution of Universities

Universities, when they were initially created about a thousand years ago, they were created to build, to create well-rounded citizens, well-rounded members of society, good members of society.

This is a fresco from the University of Bologna. This is probably 1,300 something, you know, 700, 800 years old. And you can see very clearly that even then we had the lecture model.

There is a guy delivering a lecture. There is what we call the Zubari in Bulgarian in the front, paying attention, taking notes. There is two guys not paying attention here. They're talking to each other. This guy is sleeping or he's hungover from last night or whatever.

So we haven't invented the hot water within our school terms. There's a thing which is very well known.

Understanding Bloom's 2-Sigma Problem

It's called the Bloom's 2-sigma problem. Basically, we can summarize it like this.

This blue curve here is the performance of a standard classroom. If all of us go into a classroom and are taught a random subject, doesn't matter what it is, this is how our results will be distributed once we are tested. Doesn't matter what the topic is, doesn't matter the age of the people, it could be kids in India, it could be PhDs in Sweden, doesn't really matter.

One thing which is proven, scientifically proven, is that with one-on-one tutoring and coaching we can get 98% higher than the baseline. That's why when you apply for some very important university or a school you go to private lessons, yes? Okay, so

The Current State of Universities

How about the university today? Well, the university today is definitely not one to one.

And especially for technical universities, what we see and we hear this all the time from our customers at Corstot, who are all kinds of training institutions. We sell trainers to them.

We hear this all the time that actually four-year universities or the so-called colleges in the US are going down. So unless you're Harvard, Stanford or in the picture Cambridge, you're pretty much going down.

And the problem, the big problem is that the half-life, the useful time where you can use the skills you're being taught within those institutions, is getting lower, lower and lower.

So you go to school. You study, let's say, programming for five years. And by the time you go out, all the frameworks you have studied are obsolete. So not fun, yes?

Challenges with Traditional Credentials

And at the same time, universities give you some kind of tokens of achievement, which are really only useful when your grandmother wants to see your diploma. I mean, in reality. Nobody cares about your bachelor's diploma or your master's diploma and they're not going to ask you about it. They're going to test you. If you come up well in the test, that's useful.

So the reality is that hard skills are soft. The so-called hard skills, coding and so on, are soft because they're constantly being obsoleted and they're relatively easy and quick to learn.

And then soft skills are hard. Those are the skills which are difficult to build. Those are the skills which are critical and they take a lot of time to obtain. We use them throughout our lives.

And it's not just me saying this. This is the UK Secretary of Education. I would read this to the ones in the back. She said last year, 20th of August 2023, she said, people go to university because they don't know what else to do. That's the Secretary of Education. In Bulgaria, that's the equivalent of the Minister of Education. Think about it.

Embracing Change and Technological Advances

So, happily, we can know that change is the only constant. That's the only guarantee in our lives we have more and more change.

And all the change, or most of the change throughout our lives, apart from a lot of stupid people starting wars, comes from this picture, which any presentation is incomplete without it. This is Moore's law. I guess everyone knows Moore's law.

It says that price performance of anything silicon-based doubles every X amount of time. And people say, OK, it doesn't double over 24 months. It doubles over 18 months. Doesn't matter. What matters is it doubles. It's exponential.

And the reality is I've been using this picture in presentations since 2015. And the reason it is in this presentation in 2024, nine years later, is because it has this thing. It surpasses the brain power of a human in 2023.

What did we have in November 2022? End of November 2022. A product. Everybody uses it. Come on. Good. Yes.

So can it speak in a more competent manner than you in, let's say, things you don't know? Yeah, probably does. So the reality for many of us is that we are here. Human performance is basically linear. Over time, we can get better in many things, but not too much better. And we're here for, in many cases, the computer performance is getting very close or better than humans.

Future Schooling: AI-Assisted or AI-Resistant?

So what about schools? What kind of schools will we have in the future? My guess is we'll have two types of schools.

We are going to have AI assistance schools, which As the AI built the picture, I asked the AI to build me a picture for this. There will be those shiny places where kids go and learn nice things, and they have a prosperous future.

And we'll have the AI resistance schools. So they'll go to the factory.

Immediate Solutions with AI in Education

So those are the things which can be fixed right now by AI and software. I haven't invented them. I'm not taking credit for this research. It comes from a guy called Sanjay Sarma from MIT. He is the lead of their research lab for education. That's basically a summary of a couple of his books.

It doesn't exist in this form anywhere. But I've been using it in presentations.

So the only thing we have today, really, is lectures. That's what we have in the classrooms. And we're doing it in quite a poor manner, various class sizes, every one size fits, so on and so on and so on. And we're doing it in person.

We have none of these things. We don't have the retrieval effect. We don't have spaced repetition, interleaving, and so on and so on and so on.

Basically, the whole thing is, you can go ahead and read it, but I'll give you an example. You are more likely to remember something And that's what the retrieval effect is. You're more likely to remember something if you're asked to explain the concept to someone exactly in the time when you're starting to forget it. Then you'll remember it for much longer. Or if we mix the subjects. and have interdisciplinary lines throughout all subjects, we're more likely to understand and remember the concepts. Those are things we don't have generally in the education system right now. And of course, you know, pushing curiosity, coaching, hands-on learning, those are in many cases just wishful thinking.

Leveraging AI to Transform Education

So what we can do is we can use computers for what they're good at, and we can use humans for what they're uniquely good at as well. At the same time, we need to take the mega trends into consideration.

Those are the megatrends listed by Pearson's Future of Skills report. Pearson, if you don't know, is this multi-billion monster which rules big, vast amounts of the educational world. If you've gone for a professional certification, most likely they have made the exam for this certification. If you've gone in university, they've probably printed the books, and so on and so on. So they say that those are the megatrends.

Technological change, obvious. globalization or the lack of the reverse of globalization, which is right now happening. We also have the demographic change. You know, Bulgaria, for example, we're talking about the kids and the K-12 education all the time, but the reality is we do not have a demographic pyramid. There's no pyramid. I mean, there's less kids than grown-ups. That's clear. So why are we not talking about the upskilling of adults? And then we have environmental sustainability or urbanization. Clearly, all the statistics are pointing this out. Increasing inequality, by the way, not the least pushed ahead by the technological change and political uncertainty, which I think Vlado spoke to you about quite a lot just recently. So here's a list.

Steps Towards Personalized Learning

Here's a list of a couple of steps which we can do now and then I'll go down and drill down into some of them. But basically the initial thing we can do with available tools right now is we can identify a person's potential which is a combination of interests and natural skills. There are products which exist right now which can do that.

Based on that, we can make and use co-pilots for learning. So we can flip the classroom. Flipping the classroom means instead of giving lectures out to people, record the lectures, provide asynchronous learning materials so they can read or learn online whenever they want.

And then they come to the classroom and ask questions to the teacher. However, when they're asking questions to the teacher, first, initially, the questions go through a co-pilot. And the co-pilot answers, let's say, 90% of the repeatable questions. Because people tend to get stuck in the same places.

So if you repeat the same lecture several times, you'll see that people ask you the same questions phrased in different ways. So you can use a co-pilot with that. It's quite simple. I'll show you one in a second.

Then we can analyze the user information. So we can build a baseline of facial microexpressions, of tone of voice, of vocabulary, and so on, and look for deviations from it. From that and from the personal profile of everyone, we can build their profiles, we can build their personas, and we can find you in the learning path based on this information for them.

And then we can use AI agents in the end to create new curriculum. Right now, the problem is not that we don't have enough learning materials. The problem is that we're drowning in learning materials and that the learning materials which are printed are obsolete by the time they're printed. So what we can do is we can use AI agents using our profile, using the copilots, using materials which we have access to and or are public, to mix them in the optimal way for us. And then we have our own personalized learning path.

In terms of integration, I've expanded a little bit in the next slides. There's four phases, basically, the way I see them. Again, non-inclusive doesn't include everything. But we have four phases.

Phases of Integrating AI into Education

We have the introduction and integration of AI tools in all kinds of, let's say, teaching institutions. Then we have enhanced personalization, interactive learning, AI as collaborative educational partner, and teachers as mentors and coaches. Their role is changing. The role of the teacher is changing.

So in the first phase, I know wall of text won't read it to you, but it's the introduction and integration. And the key thing is we need to train the teachers. Doesn't matter if it's the teacher in the kindergarten, doesn't matter if it's the teacher in high school, doesn't matter if it's the university professor, we need to train the teachers. It's the same way as if you have a team of developers, they need to use the latest tools. If we don't use the latest tools which are available, we are lacking behind everyone else.

The transition criteria to move to the next stage is we move to the next stage when integration is seamless, teachers are comfortable with technology and students engagement shows improvement.

I have three kids. I'm drawing pictures with the 4-year-old. He happily talks English to the phone and then DALL-E 3 draws pictures with him.

The 10-year-old uses ChatGPT and our own tool EtherMind. That's a new company we're building. to write her homework. She summarizes textbooks and uses it.

And then the 14 year old just recently recorded her history professor in class, passed everything to Whisper, which is a speech text, then put everything into ChatGPT. And then told ChatGPT, OK, write me the referata. I don't know the word in English, the 10-page thingy you need to write for the assessment. And that's it. That's the way you get a 10-page assessment. Or you can waste your night to do that.

Phase two is enhanced personalization and interactive learning. But the key thing here is personalized learning paths. That's a thing which we've been talking about in the education space since forever. If you have a look, there's a guy called Sugata Mitra. His tech talk in 2012 was about personalized learning paths. And we're still nowhere.

We have a one size fits all educational system. So the thing here is to use AI to create individual learning paths, adjusting pace and content based on student progress and interests. No, we don't have to put everyone through the same tests. All we need to check is progress. We need to know the delta, not the common performance of everyone. So transition here is when personalized learning paths show positive results in student performance and engagement.

The Potential Impact of AI on Learning Materials

Phase three, AI is a collaborative educational partner. The key thing here is that we can use AI to adapt learning materials and strategies on the fly, real time. while having the feedback from the students or from pupils. And we move forward when there's evidence of successful human-AI collaboration.

And phase four, which is, by the way, an all-encompassing phase, is that we need to move the teachers, tutors, university professors, and everyone else into the role of a mentor and a coach.

Shifting Roles: Teachers as Mentors and Coaches

so the human does not stand in front of people and deliver a lecture but rather is the person who's helping them who's guiding them into their choices life choices career choices monitoring areas like emotional intelligence helping them with ethical reasoning what's good what isn't what is a fake media and so on those are the things where humans can really help with the unique human features so in reality what we see over the past year and a half since chat GPT was released, everyone's like, you know, magic is happening, AI is coming either to fix everything or to kill us, depends on which side of the line you're in. But, you know, we really can do things which we were not able to do with, let's say, with GPT-3, because we had access to GPT-3 and you really couldn't do the things you can do right now with the modern transformers.

It doesn't... have to be GPT-4. It can be Cloud, it can be Lama 2 or something else.

Mistrial works well quite. This is a screenshot of our latest company, our CTO is over there, Ivan. And that's a company called Ethermind.

Emerging Educational Tools and Platforms

We use licensed training materials from training companies. We have like hundreds of courses in it and we're building co-pilots which We're launching in the US in a couple of weeks, so this is a pre-release screenshot.

But basically, the idea is that you go for a class. You listen to the instructor, be it recorded or via WebEx or Zoom or whatever. And whenever you have a question, you don't have to ask the trainer. You don't have to wait in line to ask the trainer.

The machine already knows all the content from this class and from all the other classes from concentric circles of knowledge. And it works like magic. So far so good.

And then there is this tool called MindStone. We are right now on one of the MindStone events. It's a multinational event. The reason we're speaking English is because it's a part of a series of multinational events.

This one is being recorded over there. Everything is being transcribed. What MindStone does is it then uses this talk and all the other talks and all the other learning materials in there. It transcribes them, cuts them into separate pieces of knowledge.

And once you go there, those ones of you who have profiles on MindStone, once you view several materials, the machine actually knows and presumes what kinds of competencies you have based on what you've watched automatically. You don't have to do anything. It can then generate tests. ask you questions and so on.

Of course, promotion time. They have an AI mastery course. I have no idea how much it costs, but they gave me a code for discount, so you'll get the presentation afterwards.

Then there's a number of other tools which are popping up all the time. Things like Study Hall. So they have an AI guide librarian. Things like this one, Third Space Learning. They have one-on-one teaching for kids, automatic assessment assignment of one-on-one teachers.

And those of you who have kids, Photomath, you must have seen it. Photomath is crazy. They raised $26 million a couple of years ago, Series B. They solve equations. So if you were like me, I struggled with maths in high school, I could have used that.

Then we have Microsoft. Obviously, Microsoft is expanding their co-pilot, which includes everything Microsoft-ish, including all the office suit. So imagine you can go to PowerPoint and tell it, OK, make me a presentation about the internal combustion engine, because that's what we're studying today. You get the presentation immediately.

Assembling the Pieces: A Unified Educational Ecosystem

You don't have to do anything. So, really right now it looks like it's a matter of assembly, right? All the pieces are coming together, various actors are building various pieces, so it's a matter of assembling everything into one place, yes? Or is it?

Regulatory Considerations for AI in Education

We have this thing called the AI Act, which is going to be adopted, I guess, very soon. And a couple of notes there are very important to take into consideration, especially about education.

First one is that there is an area called unacceptable risk. And in there, there is the real time remote biometric identification systems, such as facial recognition.

So in order to have personalized learning, We need to recognize the people in some way. We need to analyze their emotional state. This can be done, of course, in a browser.

There's a library by the Fraunhofer Institute that runs on the browser completely anonymously. But it has to be done.

That's the only way. I mean, if you have a private teacher who teaches you, let's say, sales skills or maths, they know who you are. So we have to identify the people. That's an acceptable risk right now, as defined in the AI Act.

And then in the high risk category, which is quite tricky to comply with, it says that high risk systems are AI systems which are intended to determine access, admission, or assign natural persons, not AI persons, to educational and vocational training institutions or programs at all levels. This can be read again as we are going to have a one size fits all system and there's going to be quotas for everything. The reality is this needs to be changed or adapted or in some way not very strictly enforced.

The thing is who reads that and are we trying to be ultimately very compliant. Then AI systems intended to evaluate learning outcomes are also rated as high risk. And that's the future in the EU. So we have to be very careful.

We have 36 months to comply if this thing is accepted as it is. And then we'll see how it goes.

Global Educational Advancements

The question is, will they wait? Will the rest of the world wait? We're not the center of the universe in the EU. And who has an answer?

Does anybody have an answer? Will the guys in India wait? Yeah, okay. So India is adopting their educational system like crazy.

They're going AI first. Obviously, they see this as a unique opportunity. And it's not just India.

It's many, many countries. But think of it this way. Let's take Bulgaria. If our educational system was 20 years behind everyone else up till last year,

Now we are on the same page. Everyone is one year behind just because nobody has implemented the AI tools in their educational system. So now is the time for our unique opportunity. Now is the time where we can leapfrog.

And if we don't do it, Bulgaria or the EU, the rest of the world will. This is a picture of a classroom in China. And those bands scan the brainwaves of the kids, just to make sure they're paying attention.

They're not training any AI models on them. It's just to make sure they're paying attention. So I'll close with this. The future is already here. It's just the thing is not evenly distributed. Thank you very much.

Conclusion and Q&A

And we're open for questions.