We're excited to tell you about our AI4GA project, where we're co-designing and piloting an artificial intelligence elective with Georgia middle school teachers. This 9-week course covers such topics as self-driving cars and autonomous robots, recommender systems (Netflix, Tiktok, Amazon, Google), intelligent agents (Siri, Alexa), machine learning, language understanding, societal impacts of AI, and career opportunities in AI and robotics.

We are investigating:

• How can we make AI relevant and exciting to different populations of students, specifically African American, Latino, and rural students?
• How can we help students develop personal identities as STEM creators preparing for future AI-powered careers?
• What kinds of support do Georgia middle school CS and CTE teachers need in order to feel comfortable teaching AI?

This is year 1 of a three year project. In future years we will develop an online teacher professional development course and learning community to scale up the implementation across Georgia. Our project team includes investigators from Carnegie Mellon, University of Florida, Boston College, Georgia Department of Education, and The Findings Group.

We invite you to ask us about our project, including:

• What prior computing knowledge does the project assume?
• What have we learned from our middle school teacher co-designers?
• What software tools and online demonstrations does the curriculum make use of?

Thanks for sharing this fascinating work! I'm excited to hear your team is using co-design to engaging stakeholders across GA in developing tools and curricula that connects with the experiences and needs of the various communities where you work. The video does a great job of highlighting some examples of the how AI might connect with the lives of your students. How are these insights impacting the design of your PD and curricula? Being aware of AI is a vital first step (for example thinking about the sensors in the Roomba or tractor). What comes next? How is the program supporting students in seeing themselves as "STEM creators"?

This is so great - thanks for sharing!! I love the conversations you had with the teachers to help insure that the designs you develop are relevant for their students. One thing that has always struck me in the literature on CRP and Funds of Knowledge / Identity is how hard it is for many teachers to actually get to know their students / communities. I wonder how you helped the teachers navigate that and / or whether you also spoke with students? Thanks for sharing! 

Several of our teachers in the first cohort polled their students and got feedback on the curriculum that they shared with us as well. 

It comes as no surprise that the students enjoy and want more hands-on activities. They love the demos and the other components of the curriculum that compel them to explore the different aspects of AI via some immersive experience. Dave has designed some really cool activities and the teachers have been amazing at translating them into their instructional frameworks. 

Wow - what an all-star team working on truly important topics.

I really appreciate your focus on culturally responsive & sustaining pedagogy. It's no surprise that you're doing a very good job of it, but I want to appreciate two things you mention in particular: 1) recognizing that the groups we assign learners to are not monoliths, and 2) taking an asset-based approach to what learners bring to the learning. It's a good reminder for me to build this into practices & projects. Thank you!

Working with middle schoolers (youth and teachers) always unearths new things when I have the opportunity. What has surprised you so far as you have built-out this project, and how have you adapted to take that into account?

Have you noticed any difference in identity-making between young women and young men? Has there been any difference in interest in the topics/examples that they bring up or that you supply between genders?

Thanks for sharing this with the world!

Demo of the day: SpeechDemo

Although we're not ready to release our curriculum yet, we do have some publicly accessible demos that we make use of in our curriculum. One of them is SpeechDemo, which uses the Google Speech API (Google Chrome only) to demonstrate speech-to-text transcription in several languages. What's unusual about SpeechDemo is that it not only shows you the highest scoring transcription, it also shows you several runners-up so you can appreciate the ambiguity in the raw speech signal and wonder at Google's cleverness at resolving this ambiguity. That's where the AI comes in.

The demo includes a list of suggested experiments involving things like homophones (there/their/they're or which/witch), nonwords, and famous quotations. These highlight the different kinds of knowledge involved in resolving ambiguities, and also point out places where current systems fall short. (People are much better than current speech recognition systems at transcribing phonetically well-formed non-words.)

Direct link: https://www.cs.cmu.edu/~dst/SpeechDemo

This is such an important project and amazing research team! I really appreciated reading through all the thoughtful comments, questions and responses here as well. As someone who is working in GA, I'd love to learn a bit more about particular types of challenges teachers have shared to implementing or integrating AI into different disciplines and in different school settings? (maybe a year 3 question!). Again, really wonderful to learn about this project.

Josh, we're just beginning to look at student exit surveys from cohort 1. We'll pay attention to the evolving perspectives of all the demographic groups, attending to intersectionality where possible, and include those reflections in our annual report. 

One thing I didn't realize in middle school is the approach students and teachers take to the elective courses. In Georgia, most middle school CS courses last 9 weeks and students expect them to be decidedly different from their core classes: minimal lectures, hands-on activities, exposure focused rather than deep skill development. This all matched well with the goals of our curriculum but when we strayed too close to the traditional learning environment, they made sure we were aware. Conversely, several students were so engaged that they looked for ways to dig deeper into the content. Some requested to take the elective again in the following 9-week rotation while others asked for afterschool clubs and opportunities related to the topic. We had some amazing teachers in cohort 1, so I'm sure that accounted for some of the enthusiasm. I'm curious to see how the students in cohort 2 respond. 

Demo of the day: FaceDemo

This is an online demo of face detection using the TinyYOLOV2 neural network. It uses the computer's webcam and draws a bounding box around any detected face in the image. What's unusual about the demo is that it's a "glass box" demo, not a "black box" demo: it lets you look at the internal representations of the neural net. One set of displays show the kernels of the first convolutional layer, which are sensitive to various types of edges and are also color-sensitive. You can see both the kernel weights and the responses. A second set of displays shows responses of units in the 4th max-pooling layer. These units are about mid-way through the deep neural net; they compute more complex features, such as locating eyes in faces.

The demo is accompanied by some tutorial material explaining the network architecture and the organization of the display, and a set of suggested experiments that can be used to probe the behaviors of the different kernels. Middle school students can perform these experiments by holding one of the pre-supplied stimulus patterns up to the camera and observing the responses.

Direct link: https://www.cs.cmu.edu/~dst/FaceDemo

Demo of the day: Word Embedding Demo

Word embeddings are a feature vector representation of words that are an essential part of neural network natural language processing. Neural network technology for natural language processing is used today in many AI applications such as web search, machine translation, and intelligent conversational agents.

The demo offers a 3D spatial representation of words as points in a semantic feature space, and provides for experiments such as finding the 10 closest words to a given word, or performing vector arithmetic analogies like the famous "king - man + woman = queen" that are the hallmark of semantic feature representations.

We have two ways to introduce middle school students to this material. The first is to have them play Google's Semantris game. After playing the game, we ask students to speculate about how the game knows that two words are related. Feature vectors are the answer. The second approach is to have students physically model a 3D semantic space in the classroom. This approach was developed by one of our teacher/co-designers and subsequently adopted and further refined by other teachers.

Direct link to the demo: https://www.cs.cmu.edu/~dst/WordEmbeddingDemo