Empowering and Untethering Teachers with AI

Satya Nitta, Co-Founder and CEO of Merlyn Mind, an education AI company that allows teachers to automate and voice-activate once clunky digital teaching tasks, as well as the founder of Emergence, which just came out of stealth mode with a raise of a whopping $97.5 million in venture capital, joined me to discuss how the AI technology in Merlyn Mind untethers teachers from their computers, the new learning possibilities unlocked by that change, and the importance of the practical implementation of AI tools.

Michael Horn: 

Welcome to the Future of Education. Where we are dedicated to building a world in which all individuals can build their passions, fulfill their potential, and live a life of purpose. Today, we will discuss transforming our K-12 system and supporting educators globally. I'm delighted to introduce our guest, Satya Nitta, the founder and CEO of Merlyn Mind. Merlyn Mind is one that’s been on my radar for quite a number of years for its approach to artificial intelligence. We’ll hear how their approach is very distinct from a lot of the hype and conversations around AI at the moment.

Satya, thank you so much for being here. It's great to see you. I appreciate you joining us.

Satya Nitta: 

Pleasure to be here, Michael.

Michael Horn: 

You bet. Let's dive in. You founded Merlyn Mind back in 2018, well before the current craze around large language models like ChatGPT. Even back then, adaptive learning was a big topic, and AI was frequently discussed in that context. When you started Merlyn Mind, you made an important decision to focus on serving the teacher first. I’d love to hear about that origin story and why you made that decision. What was the vision behind Merlyn Mind?

Satya’s Journey with AI in Education

Satya Nitta: 

Sure. Before Merlyn Mind, I was at IBM Research for 18 years. In the first half, I was advancing Moor’s Law, working on chip technologies. In the latter half of my time there, I worked on AI. I got into AI around the time Watson won Jeopardy. I was given the keys to the kingdom around 2012 to 2013 when Watson won Jeopardy. That was a seminal moment when a computer seemingly understood language, complex allusions, and puns, and beat the two best players in this complex quiz game. This was similar to Deep Blue beating Kasparov, and both events happened at IBM Research down the hall from where I had an office.

When Watson won Jeopardy, IBM was approached by various companies wanting to use Watson in their industries, including education. In early 2013, I was given the opportunity to explore how to use AI in education. I had no prior experience in education. I was working on either advancing language modeling. Language models predated large language models, which is the whole chat GPT revolution. I was working on conversational systems and speech recognition, and I thought this was a great opportunity to take AI and do something in a particular domain. I concluded that AI works best in deeply domain-specific ways. 

So I spent six months to a year studying cognitive psychology, neuroscience, and learning science. I already had some exposure to neuroscience, especially cognitive neuroscience because of an interest in a branch of computing called neuromorphic computing. So I went back to IBM, and I basically said to them “Look, we can do a number of things with AI and education. We can take the Watson system and build question-answering applications or chatbots across a number of things. Universities can use it to help students who are onboarding get all kinds of answers to their questions. We’re sitting here at IBM Research, one of the places that has really advanced computing, we need to do something foundational with AI and education. When tasked with integrating Watson into education, I drew from the 1957 Dartmouth Conference, where the term "artificial intelligence" was coined. The founders of AI, like Marvin Minsky and Herb Simon, saw teaching machines as a grand challenge. We at IBM Research aimed to build an AI tutor, which was a significant undertaking. I basically said, look, I'm sitting here at IBM research in these hallowed halls where the dram was invented. Moore's law was advanced through Dennard scaling. Watson won Jeopardy. Kasparov was beaten by Deep Blue. Much of modern computing has some footprint in this building. I feel the pressure to do something grand. And we need to go after this grand challenge, build a computer to teach. So build a tutoring system.

And I wasn't just making it up. In fact, that mantle of trying to get a computer to teach was picked up by generations of AI researchers. So we were sitting on top of 30 years of work in academia. Scientists like John Anderson at Carnegie Mellon had spent a lot of time thinking very hard about, how to get a computer to teach. What is an AI tutorial?

And I'm going through this. Sorry, elaborate history because I want to establish the provenance of ideas and I want to land it to where we are in this moment in AI. IBM was thrilled with the vision. We spent about five years with a team of 130 researchers, investing millions of dollars to create this AI tutor. Before the current craze on AI tutoring, we had taken all the work in academia and built the first large-scale industrial tutor. Carnegie Learning is another company that's advanced.

Michael Horn:

I was gonna say Carnegie learning. Newton had another... There had been other attempts at it as well.

Satya Nitta:

Yeah. So we also looked at Carnegie Learning's work and we said, okay, you know, what they did was very interesting. And we wanted to build an even broader approach to tutoring, well beyond something very hierarchical like math and, you know, go into lots of topics. And at the heart of it, what we were attempting to do was to get a computer to build a chatbot that a student can chat within a very natural language. And this is well before chat GPT but with language models of that time. So the chatbot would ask the student a question. The student would respond in natural language. The chatbot would then analyze the response and tell them what they were missing and not give them the response and not give them the answer.

So. And all of all, this is where we published all of this work.

By the end of 2017, I was leaving IBM. I got recruited to go join Amazon, and head an AI effort there. Then I got this once-in-a-lifetime opportunity to start Merlyn Mind. Some of the major backers in this company reached out and said, we heard you're leaving. What's happening to the team? Would you be interested in starting a company? I jumped on the whole idea. I left with a very key realization, which is, we built a tutor. It worked.

Challenges with Uptake in AI Tutoring

Satya Nitta:

It did something very complex and profound from a technological perspective, much more than anything today's tutors do.

We had elaborate student models and knowledge models. We could score student responses, and we ran into a fundamental roadblock, which is that students weren't using it. We built this for higher ed, and we couldn't get them to use it, despite us leaning in on topics like multiple representations, which are multiple ways to teach a student a concept and put the student in charge. We spent a lot of time thinking about the user experience, but we just couldn't solve the last mile, which is the motivation problem. We couldn't get a student motivated enough to use the chatbot.

And by the way, that fundamental question remains today. Okay, what most people don't answer, don't ask. When you see all these flashy demos of GPT four-based tutoring, are people using it? What's the monthly average use? What's the daily average use? How long are they using it for? Are they sticking with it? I mean, how much have they used it initially? And how much did they use over six months time.

And nobody asked them the hard questions about, is this thing solving the problem of teaching these kids something. Are we seeing an improvement in learning outcomes? So all these things became major questions in our heads, and we finally concluded that the major problem here isn't a technology problem. It was something much more profound, which is, students learn from people best. That the teacher becomes the central fundamental role model, who delivers kind of wisdom and knowledge and serves as a human example of learning for students, okay? And they're motivating the kids. They're giving them examples. They know the kid. They're situating learning within their background. And so we learned the hard way what generations of educators had already known, and which is that the teacher is the central and most important figure and factor in improving learning outcomes. So when we started Merlyn Mind, we said, I don't think we really want to do something impactful. It wasn't about doing something flashy and raising a bunch of money and being in the news.

It was about making a real change. We said the best thing we can do is to empower the teacher, okay? Use AI to reduce the friction, give them time back, give them cognitive space back, allow them to be with their students, and we're far more likely to help improve education than by attempting to replace the teacher.

Which we learned from hard Noah, through hard experience is an incredibly complex problem.

Empowering Teachers with AI 

Michael Horn: 

So let me pause you there. What you're describing is interesting. There have been efficacy studies on programs like Khan Academy and IXL, showing they work if used enough, but only a small percentage of students actually use them. Your point is that while IBM built a working tutor, it wasn't used. So, you shifted focus to the teacher. How is Merlyn Mind helping teachers today, especially with the recent explosion of interest in AI?

Satya Nitta: 

Before COVID, teachers were already using numerous applications in their classrooms. They spend a lot of time at their desks, switching between different educational tools, which keeps them from walking around and engaging with students. We aimed to solve this by allowing teachers to control their computers with voice commands, letting them move freely around the classroom. We developed a system where teachers can use a small push-to-talk mic to control their computers, launch new tabs, play videos, share snapshots, and answer questions without being tethered to their desks. 

One automation we developed allows teachers to share links with their class through a simple voice command. The AI takes care of copying the link, opening the email tool, populating the student email list, and sending the link, saving teachers several steps. This untethers teachers, saves them time, and reduces their cognitive load, allowing them to focus on teaching.

How the AI Agent Works

Michael Horn: 

So, the AI agent can navigate different apps, bring up lesson plans, and handle various tasks? 

Satya Nitta: 

Exactly. Our system controls the browser, which is where most educational tools reside. The AI operates the browser like a human, navigating tabs, clicking hyperlinks, launching videos, and more. This technology combines large language models with automation, allowing the AI to perform complex tasks based on voice commands. 

Michael Horn: 

Can you give an example of how a teacher might use this in a high school geometry class?

Satya Nitta: 

Sure. A teacher could say, "Send this video to my period three class," and the AI will handle the rest. It can differentiate between classes and even send resources to specific groups within a class. We're developing the ability to customize content distribution further, but the core functionality already supports significant time and cognitive load savings.

Michael Horn: 

Let's discuss the technology behind Merlyn Mind. How does it differ from large language models like OpenAI's GPT-4 or Google's Gemini?

Satya Nitta: 

Our system does include large language models, but it also incorporates additional technologies. This emerging field of AI is called AI agents. Our system, which has been in development for a decade, combines voice computing, language modeling, and AI agents. These agents can control browsers, perform tasks, and automate complex workflows. While we use our own large language models for privacy and security reasons, the system's uniqueness lies in its ability to perform multi-step tasks that generalist models like GPT-4 cannot.

Growing AI Awareness and Future Plans

Michael Horn: 

How has the increased interest in AI, especially with ChatGPT, impacted Merlyn Mind?

Satya Nitta: 

The rise of ChatGPT has been beneficial for us. It has made people more familiar with AI, reducing the need for us to educate the market. Now, people understand AI's potential and are more open to seeing how our tools can benefit them. This has helped us gain traction and interest.

Michael Horn: 

Where do you see Merlyn Mind in the next two to three years? 

Satya Nitta: 

We aim to continue improving the teacher assistant and eventually extend our tools to help students. However, we won't replace teachers. Instead, we might offer review tools that package lessons for students to study. Privacy, safety, and security are paramount. We ensure that no data is monetized, sold, or used to train our models. We're compliant with regulations like COPPA, FERPA, and GDPR. 

We plan to deepen our large language model capabilities and allow others to build with our models. Our models are designed to be faster, cheaper, and safer than generalist models. We'll continue to empower teachers and eventually assist students, always prioritizing privacy and security.

Michael Horn: 

Fascinating. We'll stay tuned to see how Merlyn Mind evolves and continues to support educators. Satya, thanks so much for joining us.