Putting a Teaspoon of Programming into Other Subjects (May 2023 Communications of the ACM): About Teaspoon Languages

In May, my students and I published a paper in Communications of the ACM, “Putting a Teaspoon of Programming into Other Subjects” (see link here) about our work with teaspoon languages. (Submitted version, non-paywalled is here.) It’s a short Viewpoint, but we were able to squeeze into our 1800 word limit description of a couple of teaspoon languages, a definition of them, a description of our participatory design process for them, and some of the research questions we’re exploring with them, like what drives teacher adoption of teaspoon languages, use multilingual keywords to engage emerging bilingual students, and identifying challenges to even our simplified notions of programming.

My students helped me to be consistent with our language in this piece, which was so helpful. I’ve been talking about teaspoon languages for awhile, and my language has likely changed over that time. They’re challenging me to be more exact about what I mean.

For example, we use the phrase “teaspoon languages” and not “teaspoon programming languages.” The term “teaspoon” comes from the shorthand “TSP” for “Task-Specific Programming.” So, the “programming” bit is already in there. But in particular, I don’t want to generate the reaction, “But, hey, that doesn’t look like a real programming language…”

Programming languages are used to create software — preferably, software that is reliable, robust, safe, and secure. The programming languages research community works to make programming more effective for people who are using those languages to create software. Programming as an activity can also be used to solve problems and explore domains. We’re building languages for that latter purpose. Much of the programming that scientists and others do to solve problems and explore domains happens to be in programming languages that can also be used to create software (e.g., Python, R, Mathematica, MATLAB). Teaspoon languages (so far) can’t really be used to create software for someone else to execute. They’re not general. I don’t think any of the teaspoon languages that we have created are Turing complete. But teaspoon languages are used to define the behavior of a computational agent. It’s still programming.

Another question we hear quite a bit is “Isn’t this just a domain-specific language?” We tried to answer that in the piece. Yes, teaspoon languages are a kind of domain-specific language, but for a very small domain — a single task. The most critical part of teaspoon languages is “They can be used by students for a task that is useful to a teacher.” DSLs are so much bigger than teaspoon languages. Maybe we can use DSL tools one day to make teaspoon languages, but so-far, we’ve built unique user interfaces and unique languages for each one. The focus is on meeting the need now, and we’ll see if we ever get to generalizability and tools later.

The issues we study in our research with teaspoon languages don’t have much overlap with the programming languages research community. I don’t have good answers to questions like, “How do you support type safety?” or “Why can’t I define a lambda in Pixel Equations?” So, we’ll just call them “teaspoon languages” — and let the “programming” word be silent in there.

How computing instructors plan to adapt to ChatGPT, GitHub Copilot, and other AI coding assistants (ICER 2023 paper): Guest blog post from Philip Guo

Hi everyone! This is Philip Guo from UC San Diego. My Ph.D. student (and soon-to-be faculty colleague) Sam Lau and I are presenting a paper at ICER 2023 on a topic that’s been at the top of many of our colleagues’ minds in recent months:

How are instructors planning to adapt their programming-related courses as more and more students start using AI coding assistance tools such as ChatGPT and GitHub Copilot?

A series of studies throughout 2021 and 2022 showed that GitHub Copilot can solve many kinds of CS1 and CS2 homework problems (see Section 3 of our paper for a summary of these studies). However, Copilot still requires users to install, configure, and activate it within an IDE like Visual Studio Code, which can be hard for novices to do. So when ChatGPT launched at the end of 2022, it made this AI technology far more accessible and easier to use. Now any student can visit the ChatGPT website, copy-paste in their homework assignments and instructor-provided starter code, and watch ChatGPT generate the solutions for them. Given this new reality we’re all now living in, Sam and I wanted to know what computing instructors are planning to do to ensure that their students are still learning well.

To gather a diverse sample of perspectives, we interviewed 20 university CS1/CS2 instructors across 9 countries (Australia, Botswana, Canada, Chile, China, Rwanda, Spain, Switzerland, United States) spanning all 6 populated continents. To our knowledge, our paper is the first empirical study to gather instructor perspectives about these AI coding tools that more and more students will likely have access to in the future.

Here’s a summary of our findings:

Short-term, many planned to take immediate measures to discourage AI-assisted cheating by weighing exams more, trying to ban these tools, or showing students their limitations. Then opinions diverged sharply about what to do longer-term, with one side wanting to resist AI tools by creating more AI-resistant assignments and exams, and the other side wanting to embrace these tools by integrating them deeply into introductory programming courses.

Our study findings capture a rare snapshot in time in early 2023 as computing instructors are just starting to form opinions about this fast-growing phenomenon but have not yet converged to any consensus about best practices. Using these findings as inspiration, we synthesized a diverse set of open research questions regarding how to develop, deploy, and evaluate AI coding tools for computing education. For instance, what mental models do novices form both about the code that AI generates and about how the AI works to produce that code? And how do those novice mental models compare to experts’ mental models of AI code generation? See Section 7 of our paper for more examples.

We hope that these findings, along with our open research questions, can spur conversations in our community about how to work with these tools in effective, equitable, and ethical ways. 

Check out our paper here and email us if you’d like to discuss anything related to it!

Sam Lau and Philip J. Guo. From “Ban It Till We Understand It” to “Resistance is Futile”: How University Programming Instructors Plan to Adapt as More Students Use AI Code Generation and Explanation Tools such as ChatGPT and GitHub Copilot. ACM Conference on International Computing Education Research (ICER), 2023.

Broadening Participation in Computing by Moving Away from Computer Science: Information, Arts, Humanities, and Sciences offer better models for #CSforAll

 In April, I gave a talk at Carnegie Mellon University’s Software and Societal Systems Department (S3D) “Broadening Participation in Computing by Moving Away from Computer Science” — slides available here, and video available here.

The argument I’m making is that computer science as a field has become more narrow over time. I wrote a CACM Blog Post last month where I provided several definitions of computer science: “Education is always changing: We need to define CS to keep the good stuff.” The earliest definitions of computer science described it as something to be taught to everyone, a critical literacy for 21st century citizenry, and touching on many different aspects of modern life. (I tell the story of those early definitions in this blog post.) More modern computer science definitions are much more narrow.

Computer science departments perform a critical function. They produce software professionals. Our society needs those. But that’s not the only societal need for computing education. CS departments also perform a gatekeeping function so that they can certify their graduates as ready for professional programming. If we want “Computing Education for All” with alternative endpoints, we need less of that.

During the pandemic, I worked with a Computing Education Task Force at my University to discover what kinds of computing education was currently available (see that story here and our final report here) across the campus. There’s a lot, even outside of CS. Arts teaches wonderful courses on expression with computing. Sciences teach how to discover with computing. Humanities reflects on the role of computing in society and critiques our digital and computational systems. Our School of Information teaches similar introductory computing courses to what we offer in CS, but with a focus on data science, user experience, and impact on society.

I suggest in my talk that new initiatives like our Program in Computing for the Arts and Sciences in University of Michigan’s College of Literature, Science, and the Arts (LSA) is more likely to invent computing education for everyone than will CS departments. The first half of my talk is about the history of computing education and about the narrowing of definitions (for which I use the U-M CSE standard slide templates), and the second half (for which I use the U-M LSA standard slide templates) is about PCAS, our new courses, and how this serves to broaden access to and participation in computing education.

Let’s think about how we could broaden our goals beyond CS departments and CS majors in K-12 education. Advanced Placement CS A and CS Principles are tightly tied to CS curriculum. What would it look like to create Advanced Placement exams for other needs for computing education? For example, what would an AP in Computational Science look like? What would it mean to value alternative endpoints in “Computing Education for All”?

Participatory Design to Support University to High School Curricular Transition/Translation in FIE 2022

Here’s my second blog post on papers we presented during the first year of PCAS. Emma Dodoo is an Engineering Education Research PhD student working with me and co-advised with Lisa Lattuca. When she first started working with me, she wanted a project that supported STEM learning in high school. We happened upon this fascinating project which eventually led to an FIE 2022 paper (see link here).

The University of Michigan Marsal Family School of Education has a collaboration with the School at Marygrove in the Detroit Public Schools – Community District. The School at Marygrove requires all high school students to take a course in Engineering every year. The school is new, so when we came into the story, they were just starting to build an 11th grade Engineering curriculum. Where does an innovative K-12 school find curriculum for not-often-taught subjects like Engineering? It seemed natural to look to the partner university.

The University of Michigan has recently established a Robotics Department with an innovative undergraduate curriculum. The leadership in the U-M School of Education and the School at Marygrove decided to use some projects from the undergraduate curriculum for the 11th grade Engineering curriculum. Emma and I came in to run participatory design sessions to help with supporting the high school in adopting the university curriculum. We focused on one project in particular, where students would input data from a LIDAR sensor on a robot, then visualize the results. What we wanted to know was: What are the issues that come up when using a university curriculum to inform an innovative high school curriculum?

We started out with a set of interviews. We talked to undergraduates who had been in the Robotics curriculum and asked them: What was hard about the robotics projects? What were things that you wished you knew before you started? They gave us a list of issues, like the realization that equations on a plane could be used to define regions of a picture, that colors could be mapped to numbers and equations, and that pixels could be queried for their colors.

We talked to high school educators about what they wanted students to learn from the project. They were pretty worried about the mathematics required for the project, especially after the students had spent all of 10th grade on-line during the pandemic.

Emma took these objectives and concerns, and generated a set of possible activities to be used in the class. She used Desmos, Geogebra, and our new Pixel Equations teaspoon language. (I mentioned back in this blog post that we were using Pixel Equations in participatory design sessions — that’s when this study was happening.)

Pixel Equations was developed explicitly for the concerns that the undergraduates were raising and that the math educators cared about. Users specify the pixels they want to manipulate (leftmost column) by providing an equation on a plane or an equation based on the RGB channels in the color in the pixel. They specify the desired color changes in terms of the red, green, and blue channels (three columns on the right). The syntax for the boolean expressions and the equations for calculating colors is the same as what students would see in Java, C, or JavaScript. But there are no explicit loops, conditionals, or data — it’s a teaspoon language.

Emma ran participatory design sessions with stakeholders, including teachers from the School at Marygrove. Her goal was to identify the features that the stakeholders would find valuable, and in particular, to identify the concerns of the high school educators that may not be addressed in the university curriculum. She identified four sets of issues that were important to the stakeholders when transferring curriculum from the university to the high school:

• Prior Knowledge: The students knew Desmos. Using a tool they used before would help a lot when dealing with the new robotics concepts.
• Priming: The computing must be introduced so that there is time for the students to become familiar with it.
• Motivational Play: High school students need more opportunities to see the fun in an activity than undergraduate students.
• Self-efficacy: It’s important for students to feel that they can succeed at the activities.

That’s where the paper ends

All of this design and development happened during the pandemic. We didn’t hear much about what was going on in the high school, and we couldn’t visit. When we talked to our contacts at the School of Education, we found that they didn’t have much news either. It wasn’t until much later that we saw this news item. The 11th grade Engineering class actually didn’t do any of the mathematics activities that we’d helped with. Instead, the school got a grant for a bunch of robots, and the classes focused on directly programming the robots instead. It’s disappointing that they didn’t use any of the things that we worked on, but as I’ve been mentioning in this blog, we find that adoption is really hard. Other factors (like grants and the wow factor of programming a robot) can change priorities.

The paper is interesting for investigating stakeholder issues when transferring activities from university to high schools. Those are useful issues to know about, but even if you address all the issues, you still might not get adoption.