Mohammad Ghassemi is a distinguished AI researcher, educator, and builder who develops tools and systems that combine human and machine intelligence to solve problems. He is the founder of the Ghamut Corporation, an AI strategy and innovation firm, and a professor of computer science at Michigan State University where he leads the Human Augmentation and Artificial Intelligence Laboratory. A computer scientist with a doctorate from MIT, he has earned many national and international distinctions, testified before the U.S. Congress, and authored over 80 peer-reviewed scientific papers.

His company currently supports AI strategy with the Gates Foundation and a group of professional learning organizations supported by the foundation. He brings a unique blend of computer science knowledge and practical experience to help professional learning leaders understand the benefits and challenges of AI and apply AI tools in their work.   

What are the main uses of AI that you are seeing in professional learning, and are they similar to or different from other fields?

Across fields, there are some common places where AI tools can help solve friction points. But within any given field you also have unique opportunities and challenges.

When any organization is interested in AI, including professional learning providers, they usually want to do one of three things. They might want to automate something, such as triaging emails or scheduling coaching time with teachers. Then there’s augmentation, which is having a machine serve as an assistant. This is really what tools like ChatGPT or Gemini do if you use them properly. They don’t replace your thought process, but they might help you with ideation, like when you need to tailor a professional learning program you’ve created to characteristics of different districts.

The third type of use is insight generation, especially when you have a ton of data or information and you want to surface what matters and what doesn’t. Let’s say you receive feedback from teachers across many districts you’re working with and you want to know what you are doing well and where you could improve. You can use AI to turn that data into an insight that you can go share with your team and take action on. 

Where there’s nuance in this field is in operationalizing all the phenomenal experience and knowledge that professional learning organizations have and allowing AI to be an amplifier. A lot of the education-focused tools out there today make the presumption that the primary end user is the student. There are fewer tools that are thinking about the teacher or professional learning provider as the primary end user. But some of those tools do exist and are designed to enable educators to do a better job distributing high quality curriculum or implementing best practices.

What are some of those specialized tools and why are they advantageous?

One category of tools allows you to do monitoring and analytics. This can be very helpful for coaching, which is one of the hardest professional learning components to scale with fidelity. A coach has to observe the teacher’s class, take notes, synthesize those notes into actionable recommendations, and work with the teacher to reflect and improve on the instruction.

AI tools like TeachFX and Sibme have those functions, allowing the human part of the coaching to shine through. They automate the note-taking, augment the generation of actionable recommendations based on the data collected, and try to provide some insight by presenting, for example, what teacher and student talk time looked like before and after the coach’s visits. The coach can use that while working with the teacher.

This doesn’t mean Sibme or TeachFX are always the right tools to use. There are many tools out there that are worth exploring, and choosing one depends on the functions you need and the goals you’re trying to accomplish.

What are the trade-offs of using one of these specialized tools versus a generalized AI tool like ChatGPT or Claude?

A specialized tool will usually cost a little bit more but provides better results for its specialized domain. Here’s an analogy: A mandoline is more expensive than a regular kitchen knife, but it slices vegetables faster because it’s specialized for that task.

A great example in professional learning is Coteach.ai, which is integrated with the Illustrative Mathematics curriculum. In a standard workflow with ChatGPT, you would have to click and drag the specific lesson you want to know about and write out a detailed prompt to help you with lesson planning that’s aligned with the curriculum. Coteach.ai allows you to bypass that, because the information source is already embedded within it. It simplifies the workflow, minimizes errors, and helps overcome user mistrust, which is one of the biggest barriers to AI use in education. You can rely on it to, for example, use the latest version of the curriculum, and you don’t have to spend as much time checking the output. That’s really important, because most educators are pressed for time.

Are those kinds of tools usually embedded in specific curricula?

There are also tools like Playlab that allow you to create something like a specialized version of ChatGPT. You can customize it to support use cases for professional learning. Once you make your module, you can choose to make it accessible to anyone who is interested. A Playlab user could go on and search to see if anyone has created a module for a specific curriculum. This creates an opportunity to contribute to the wider education community. Or if you keep it private, it provides an opportunity to differentiate your organization or system from others.

I want to note that there are also general purpose AI tools that will allow you to upload any set of documents and get responses grounded in that set of documents. NotebookLM is one of those tools. You just have to keep in mind those tools may or may not be aware of subtleties the way a tailor-made tool would be. For example, they may not read images, which are sometimes a big part of curriculum materials.

Many of the professional learning organizations you work with are focused on curriculum-based professional learning. How are you seeing those organizations use and benefit from AI?

There are several common uses cases, including curriculum unit internalization, lesson planning and pacing, generating student practice activities, ongoing coaching, and synthesis of large volumes of data. I’ll talk about a few of these, starting with lesson planning and implementation.

Planning and pacing are hard. I know this as a teacher at the university level, and it’s definitely true in K-12. For example, many districts are using a curriculum where the lessons are written to be 60 minutes long but the schools’ class periods are only 45 minutes. If you go into a tool like Coteach.ai and say “Modify this lesson to be 45 minutes long, instead of 60 minutes long,” it will draw on all the information it has about the curriculum to suggest ways to do it that are aligned with the curriculum’s goals and materials.

The curriculum designers could then also use that information to improve the curriculum, right?

I’m so delighted you brought that up. This is the thing I find most exciting about AI tools: You can get a flywheel effect. When you have people asking the AI questions about the curriculum, you can look for patterns to see what topics are getting a lot of questions and then use that information to improve the curriculum. Then educators use the improved version and ask additional questions, and you get this wonderful loop where the AI serves as a broker of growth on both sides.

As a result, you can close the feedback loop faster. Right now, curriculum producers are relying on surveys and informal feedback, but the subset of the population that responds to surveys may not be representative of the whole population.

You mentioned coaching as another use case. How can AI be helpful for coaching?

There’s no replacement for a human touch. AI can’t replace what coaches do in the intensive period of offering professional learning. But it can help with administrative responsibilities such as recording, transcribing, and synthesizing your notes. And it might be helpful to sustain teacher learning, especially after the official coaching period has stopped.

Everybody I’ve met in professional learning is extremely sharp, but they’re also extraordinarily busy. Where I think AI can really help is by getting rid of some of the administrative things that would prevent those coaches from doing the deep engagement with the curriculum. So it’s less about helping the coach ask the AI to help them with just-in-time questions about the curriculum, although that’s nice. It’s more about triaging emails, coordinating coaching meetings, and doing some of the other administrative stuff that’s eating up 20% to 30% of the workday. Now that time can be dedicated to the core craft of coaching. The core value proposition of coaches is the ability to not only master the curriculum, but to perform the transmission to the teachers in a way that’s easy for them to receive. And the more complex the curriculum is, the more valuable the coach and the professional learning organization are in the process. AI can free up time for those people to do their craft.

I think allowing the folks in professional learning to spend more time engaging in their primary art will also make them happier, and their “customers” happier. I can relate to this as a computer scientist. When I write software, my software is often more elegant and scales better than what a typical AI output would generate on a first pass. And I enjoy doing it, in the same way I imagine a coach enjoys their trade. That’s ostensibly why they got into it in the first place. I think that’s good for everybody.

What else is important for professional learning providers to know about using AI?

I think of AI use like a pyramid with layers that have interdependencies and ultimately lead to improving student outcomes. The foundation is anything that streamlines your administrative processes and saves time. This ripples up and supports data and analytics as we talked about before. We want to enable teachers to keep their classroom time sacred and use it most effectively. The things that ultimately enable more high-quality interaction between the teacher and the student are going to be the most helpful.

It’s also important to remember that AI is a specific tool that helps in some contexts and is not useful in others. This is true of any tool. Knives are not good or evil; they’re tools in the hands of a chef. If you get a poorly cut salad, that’s not the fault of the knife, or of the vegetables. As in cooking, training and education about AI are essential, because there are ways to use these tools and, equally important, ways not to use them. A mature understanding of what AI is enables people to understand that and get more out of it.

Download pdf here.