Excerpts from an interview with Dr. Porsche Fisher
How did the Educational Statistics and AI degree start? How did you get involved?
I was hired as a curriculum designer for the program. Bill Schmidt, who passed away recently, created the Measurement and Quantitative Methods (MQM) Ph.D. program. He later worked with the College of Education to create this new Master’s degree in Educational Statistics and AI.
He received a grant for my salary for two years as a curriculum developer. My salary was 100% curriculum development, which we later modified to 95% curriculum development and 5% service so that I could sit on a couple of committees. I'm co-chair of the Academic Specialist Advisory Committee. We created a new program that gives foundations in AI and the ethics of using AI with educational statistics, designed for people who are working in K-12 education as administrators or educators. Or those who are working in the government sector, in governmental institutions, or nonprofit organizations. Or just educational organizations who have large amounts of educational data from large scale educational data sets.
A large scale dataset might be the SAT, the ACT, or the M-STEP here in Michigan. There are also national and international large scale data sets such as the PISA, the TIMSS, and the NAEP. People get their scores and they don't necessarily know what to do with them, scores from their district, their classroom, their school. What does my data say about how we're doing? What are we doing well? What does that say about our teachers? Where is their room for improvement? How do we identify students that could use additional assistance? How do we identify students who could be an AP placement? How do we do things like that? And so we're creating classes that will help them answer those questions and then make changes to curriculum and policies, create programs, or modify things. We want to help them do this efficiently and quickly, to save time, money and labor by using AI applications. Not necessarily only using coding in Python, R, or SQL, but visually using drag and drop dashboards and workflows.
If you are building a predictive model using the traditional method, you are working in a coding space where you are only using characters, typing everything out in code, using syntax, whereas if you're using a low code you're building it visually. For example: You're pulling in your data set, you can just say, "Oh, look, there's my CSV file!" and you pull it onto the dashboard. The dashboard will have nodes on it, and you can say, "Oh, I wonder if I can run this process on it". You drag it, and then it'll give you a red light, and it'll say, "Nope. Can't happen." Or it'll give you a red green light and say, "Yep, you can run this." And then you can take this further and say, "Oh, do I have any missing values?" And it'll give you a green light or a red light.
You can build this whole workflow: cleaning the data, addressing missing values, splitting the data, running a linear regression, identifying high or low risk students, giving you findings. Then working up a visual– something that you can share with your stakeholders– and writing up your findings. You can do it all without using coding and have recommendations that are beautiful and accessible to stakeholders at every level; whether it's teachers, CEOs, board members, or even parents and students in your classroom. It's accessible for everyone at every level, without you needing to be someone who can write high-level syntax. Being the expert in the educational environment that you're in will help you make decisions with this visual data flow using this low code/no code environment. We are able to teach them how to do that in 10 classes, nine of them are traditional courses, and then the final is a capstone. It’s a project similar to a thesis, but it is a practical application of the skills that they have learned. They employ the skills they've learned in a real-world project that they can share with an employer, in a job interview, or for a promotion at their current place of business.
What was your background before working on the MA in Educational Statistics and AI? Had you worked in education previously?
Prior to this program, I was working for the STEM UP program at Michigan State. I was the program evaluator. My doctorate is in Human Services with a specialization in data, analytics, and program evaluation. I was the program evaluator on a grant to provide job training to differently abled individuals. This grant was from the Rehabilitation Services Administration of the United States, given to the Michigan Department of Rehabilitative Services, then to Michigan State University. When that ended the principal investigator on that project referred me to Bill Schmidt.
Prior to the STEM UP program, I had worked at Wayne State as the manager of Education at the Center for Health and Community Impact. I ran an educational program on a grant there that they had received from the Health Resources and Services Administration (HRSA). Before that, I was the education and registry manager at a nonprofit organization called the Michigan Community Health Worker Alliance. I ran their education program when I was finishing my doctoral research. So I have been in education for quite some time.
The MA in Educational Statistics and AI is preparing for the first cohort of students. Do you have an idea of what a typical student would look like, or what their background would be?
The requirements include having a bachelor's degree, but we would like to have a diverse set of people as far as ages and cultural backgrounds go. We're looking for educators who are passionate about making changes to improve educational equity and quality for their students, schools, and districts. That's the goal. I would love to have a cohort with older, more experienced teachers alongside younger teachers so that they can share the wisdom of their experience, and the younger students can share insight from their deeper technological understanding.
Can you talk a little more about the curriculum you've designed for this program and what you've wanted to focus on?
The curriculum that I've designed has really focused on the AI components. The three courses that I'll be teaching are Intro to AI for Education, Ethics for AI in Education, and Data Collection and Analysis for Education. They really focus on artificial intelligence: How do we understand it? How do we use it? When should we use it? How does it affect the people that we're using it with and for?
The other classes students will take are going to be blended PhD/Masters classes. They will take the first 2/3rds of the class with the PhD students, then they will apply what they've learned using AI tools. Instead of doing advanced theory in the final third of the class, like the PhD students do, they'll use AI tools.
You mentioned teaching a course about AI ethics. Can you talk a little more about how AI ethics are discussed in your program, how you're incorporating them, and how you're teaching them?
There's a lot to unpack there. So, what a lot of people think we're talking about is how you use AI in the classroom. That's really not what we do, we're talking about how to use AI to analyze educational data. When you're dealing with these large scale educational data sets– it could come from the TIMSS, PISA M-STEP, SAT, ACT, local classroom, school district, or statewide data– you've got a lot of confidential information that could apply to individual students, teachers, and families. That data needs to be handled according to FERPA and COPPA guidelines, these are federal guidelines for how we deal with privileged student and family data. There are entire classes that need to be taken about how to handle different types of data. The AI ethics class teaches students how to handle those types of data, how to use AI, which types of AI to use for which types of data, how to clean data, and how to not compromise your student's data when you are trying to analyze it, and how to share the results with your stakeholders.
Also, there is another concept, which is not the same, but it's called algorithmic bias. When you're dealing with generative AI it's trained on training data. The model is only as good as the data and data is based on what has been created by humans. So, the training data is only as good as what the humans have created, and humans can't create anything without bias. What we try to do is minimize that algorithmic bias when we're analyzing data. What you can do now is use multiple different models and compare them against each other in order to reduce algorithmic bias. There are many different ways to do that, we can teach students about how to verify their data to reduce bias and increase validity. And there are a ton of different methods to do that in models, we can teach students how to do that as well.
Have you had issues with AI bias in the past?
I think people talk about it a lot more than it is actually an issue. Over the last few years, there have been so many ways to minimize it and mitigate it that it's becoming less and less of a problem. But I think that when AI first came out, people weren't prepared for the issues that would come up and now that people have started to mitigate it, the old problems are still talked about. I think it's worth talking about, but I think there are solutions. When people created the first few versions of Claude and ChatGPT there were lots of problems with them, but as the new versions of these AI's are coming out they keep updating them. The old AIs are rewriting the new ones. It's very reflexive and in this reiterative process. It is able to improve itself so much than humans ever could in this recursive process, because humans are not infallible. It's changing so quickly that, I don't think it'll be as big of an issue as it was at the beginning.
Since AI is evolving so rapidly, how are you gonna keep up and equip students for after they graduate?
It's changed just since we started the course. The way that that class will be taught has completely changed. I've had to change the syllabi for the courses… I can't even tell you how many times. It's built into the syllabi due to the transient nature of this content being ever-evolving. We will teach you and give you the skills that you need to know to adapt.
The whole foundation of this program is built on a theory that was developed by one of the theorists at MSU named Dr. Rand Spiro, called Cognitive Flexibility Theory. Instead of teaching students a concept that they memorize and spit back out at you, you teach them how to take apart the concept and put it back together in different ways. They are able to adapt to changes as they come by knowing how to take things apart, add different pieces, then reassemble it. Because it's changing so quickly if you just have them memorize it, it probably won't even last till the end of the semester.
Like, we filled out these IT readiness forms and got approval [for tools used in the classes]. One of the classes that we had built, the tool that was the foundation of the class, is already outdated. We had to switch to an entirely different tool. So we’ve built that into the program, due to the transient and ever-evolving nature of this content, these things will be constantly changing. We also do that with the electives. Those will be updated constantly. And we have a floating class which will be called something like ‘Emerging Topics’ so that we always have a placeholder, in case something crazy comes out. It was built into the program creation that we fully understand that we are prepared to adapt as needed.
The Master’s in Educational Statistics and AI will launch with its first cohort in Spring 2027. For upcoming virtual open houses check our Events page. To learn more about the program, you can request additional below.