1 Demystifying AI with Google Teachable Machine: Empowering Students through Machine Learning
Curtis Clements and Jordan ven der Buhs
Abstract
This chapter explores the use of Google’s Teachable Machine (GTM) in teaching foundational concepts of artificial intelligence (AI) to students. By designing models for computer vision sound and pose recognition, students can better understand the role of data in improving AI models and recognize potential limitations in AI systems. GTM helps demystify AI and demonstrates that AI models are only as good as the data that is used to train them. By training machine models, students will be more informed of ethical considerations as well as real-world applications, such as self-driving cars, and uses in medicine, manufacturing, construction, mining, agriculture, among others.
Introduction
The rapid availability and evolution of AI presents both opportunities and challenges for educators in the classroom. One of the instructional design challenges is helping students understand AI concepts in a practical, accessible way without overwhelming them with technical descriptions. Students may perceive AI as an actual person making it difficult to grasp its functions and implications. Without a basic understanding of how AI works, and the increasing availability of these tools, students may become uninformed consumers of this technology.
GTM is an AI tool designed for students and educators that allows students to build simple machine learning models using image recognition, sound recognition, and pose recognition (Google, n.d.). The tool demystifies AI by enabling students to design their own models, and observe how data impacts the performance of their models. The following link provides an explanation of how GTM works, and some practical examples:
https://www.youtube.com/watch?v=T2qQGqZxkD0 (Google, 2019)
After learning how to create their own models, students can then improve their models by reflecting on the quality and quantity of data that is used in training the model. The models can also be exported to be used in specific applications. GTM has been used to create an image model to identify dangerous types of melanoma (Forchhammer et al., 2022). Wong et al. (2022) used GTM in an ecological study that used an image model to detect species of shorebirds and waterbirds in Malaysia, providing a simple and affordable method of doing important conservation work.
Figure 1
Examples of Computer Vision that have Practical Applications
Note. Curtis Clements (2024) generated these images using the Google Gemini (1.5) platform . I dedicate any rights I hold to these images to the public domain via CC0.
GTM is relevant to educators who want to empower students to understand and responsibly use AI and comprehend it’s practical application in real world applications. GTM can fit into a variety of curricula. In high school and middle grades, it may be possible to incorporate this tool into practical applied arts or similar career and technological curriculum. In earlier grades (i.e. grades 4-6) it may be possible to link GTM to a curriculum that explores aspects of communication and society.
By demystifying and effectively communicating the concepts of AI, students will be encouraged to develop their own questions and insights. This will help them recognize current AI applications and explore its potential uses in the future.
Learning Objectives
Learning Objectives
-
Explore how students can use Google Teachable Machine (GTM) to create machine learning models for applications in computer vision, as well as pose and sound recognition.
-
Appreciate how data impacts the accuracy and effectiveness of AI models, including ethical implications such as bias.
-
Identify how this technology is used today and its applications and implications for the future.
Case Study
GTM provides a hands-on experience for students to explore fundamental AI concepts without requiring advanced programming knowledge. Students can easily create image recognition, sound recognition and pose recognition models using their own datasets.
Examples
-
Students can create an image recognition model that can differentiate between different students, or numbers written on a piece of paper.
-
Sound recognition models could differentiate between different songs or voices.
-
The pose recognition model simplifies the human body into line segments and points and can differentiate between standing, crouching or other poses.
This tool allows students to see how data collection, model training, and testing impact AI performance. Students are able to reflect on their models, why they do or do not work and improve them by improving the quality and quantity of the data that is used to train the model. Ng et al. (2024) refers to this as ‘maker-based learning experiences in AI education’ where students utilize design principles or problem solving activities to learn AI concepts.
GTM provides a free and safe environment for students to experiment with AI. No login is required for use and data that is inputted stays ‘on-device’ meaning that no images or sounds leave the student’s computer (Google, n.d.). Students may also upload their model to save it, share it or use it with other applications. The uploaded model is published to Google servers, which allows access to the mathematical model, but not the training samples that were used as data (Google, n.d.).
The ease of use and adaptability of this program allows it to be used with a range of grades in a range of applications. Elementary students may use it to extrapolate about the advantages and disadvantages of technology. Secondary students may use it to have more in-depth conversations about bias, stereotypes in data sets, implications for privacy and applications for the future.
Tutorials
There are several tutorials and educational resources that are linked on the site. Webster (2019a) provides user-friendly instructions on creating a model that will determine whether a banana is ripe, under-ripe or over-ripe. Any object can be substituted for the banana and students can get creative, creating models based upon their own ideas.
This tutorial includes extensions such as testing a picture of a banana instead of a real one, changing the background and providing two bananas. Students can use these ideas to test their model to determine its flaws, and determine ways to improve their models.
Webster (2019b) also provides a tutorial on using GTM to create an audio model to differentiate between snap, claps and whistles.
After experimenting, testing and improving image, audio and pose models, students will have a better understanding of the importance of data on AI models, and see connections on some of the technology they use in their lives (i.e. facial recognition on cell phones etc). Applications of this type of technology can also be explored in various sectors of the economy such as health care, manufacturing, construction, agriculture, autonomous vehicles, mining, etc.
Additional Resources
Educators that want to delve into a more thorough examination of AI ethics can utilize the embedded resource “An Ethics of Artificial Intelligence Curriculum for Middle School Students by Blakeley H. Payne with support from the MIT Media Lab Personal Robots Group, directed by Cynthia Breazeal (2019).” This extensive resource provides many ideas on how to engage students in important discussions about AI ethics. AI bingo, a fake images test , an ethical matrix activity, speculative futures activity and resources to create a model that can differentiate between cats and dogs are just some of the activities that educators can use to help students understand the foundations and implications of AI.
Practical Application Vignette
In a middle school science class, students are tasked with designing a machine learning model to classify different plant species. The students gather pictures or leaves of different plants and create a model to identify different species. Students can test each other’s models to find areas of improvement. Students can then modify their dataset by adding more or better-quality images to improve the AI’s classification accuracy. This process emphasizes the importance of data quality and diversity in building effective AI models. Extensions may include, discussing how their models could be improved to work better for people in different locations, and how models such as this could assist the agriculture industry in being more efficient in producing food.
Assessment of Effectiveness
Teachers will need to align the content with their local curricula, however middle and secondary grades may be able to adapt the learning activities to practical applied arts or similar career and technical information curricula. Elementary grades could adapt it to learner outcomes in society, technology, and communication and information topics that are sometimes found in English Language Arts curricula. It is also possible to link it to other subjects, such as science. Martins et al. (2023) provided an in-depth study on a machine learning course that allowed students to use GTM to build a model that would classify recycled materials. This study showed that students perceived it as an enjoyable learning experience with no substantial differences with regard to educational stage, gender or instructional mode (Martins et al., 2023).
Possible Curriculum Applications (Saskatchewan High School Curricula)
|
Course |
Curricular Outcome |
Application |
|
Computer Science 20 |
CSE2.1: Investigate and analyze the impact of technology and technological advancements on society. |
Students could explore AI’s societal impact, training machine learning models using Google Teachable Machine to understand how AI influences decision-making. |
|
CSE2.3: Demonstrate knowledge of basic principles and processes of machine learning (Saskatchewan Ministry of Education, 2016a). |
Students can create projects using Google Teachable Machine to model how machine learning algorithms classify data or objects (OpenAI, 2024). |
|
|
Science 9 |
EC9.2: Analyze scientific knowledge of electricity and describe how it impacts technologies and society (Saskatchewan Ministry of Education, 2009). |
Use Google Teachable Machine to simulate how electrical circuits or sensors can be integrated into AI-driven technologies, demonstrating an understanding of technological applications of electrical science (OpenAI, 2024) |
|
Information Processing 10/20/30 |
IP10.3: Apply technology to solve problems and make decisions. |
Students use Google Teachable Machine to develop simple AI models to solve real-world problems (e.g., classification tasks in image or sound recognition). |
|
IP20.1: Demonstrate understanding of the basic principles of software design (Saskatchewan Ministry of Education, 2016b). |
Using Google Teachable Machine, students can understand the backend principles of AI software design, learning how models are trained and applied in real-world scenarios (OpenAI, 2024). |
|
|
Environmental Science 20 |
ES20.1: Investigate various environmental science technologies (Saskatchewan Ministry of Education, 2017). |
Students can use Teachable Machine to explore how AI could be used in environmental monitoring, such as classifying different sounds in nature or detecting environmental changes through images (OpenAI, 2024). |
|
Practical and Applied Arts (Robotics) |
ROBA3, ROBA4, ROBA6 Understand and demonstrate basic skills, techniques, and procedures related to simple robotics or automation. (Saskatchewan Ministry of Education, 2011f) |
Students work in pairs to train Teachable Machine to recognize common classroom items (e.g., books, markers, erasers) as part of creating a “classroom helper.” They’ll brainstorm tasks the machine could help with, such as sorting supplies. After training the model, each pair presents their “helper” and explains how it works, discussing why certain objects were harder or easier to teach. They can also try using it to “test” objects brought by classmates and analyze how accurately it identifies them (OpenAI, 2024). |
Note. ChatGPT, an AI language model, provided insights into the practical application of Saskatchewan curriculum outcomes with the GTM. These are only a few examples from a variety of different high school Saskatchewan curricula.
Possible Curriculum Applications (Saskatchewan Elementary Curricula)
|
Course |
Curriculum Outcome |
Application |
|
English Language Arts (4/5/6) |
CC4.3 Speak to present and express a range of ideas in formal and informal speaking situations for different audiences and purposes. (Saskatchewan Ministry of Education, 2011a) |
Students create a model in Teachable Machine to recognize two or three objects (e.g., “pencil” vs. “eraser”) and then do a show-and-tell presentation explaining how it works in their own words (OpenAI, 2024). |
|
CC5.3 Speak to express and support ideas in formal and informal settings for particular audiences and purposes. (Saskatchewan Ministry of Education, 2011b) |
Students use Teachable Machine to train it to recognize different colors. Then, they give a short presentation to the class explaining how the machine “learns” colors, supporting their ideas with examples (OpenAI, 2024). |
|
|
CR6.5 Listen purposefully to understand, respond to, and analyze oral information from various texts, including narratives and instructions. (Saskatchewan Ministry of Education, 2011d) |
In pairs, students train Teachable Machine to recognize simple actions like waving or clapping. Each group then presents their model, and the class guesses the actions, listening carefully and responding to what they see (OpenAI, 2024). |
|
|
CC6.5 Use oral language to interact appropriately in pairs or groups, exploring others’ ideas and completing tasks. (Saskatchewan Ministry of Education, 2011c) |
Working in small groups, students teach Teachable Machine to recognize one of their favorite items (e.g., a toy or school supply). Each group takes turns asking questions about other groups’ models, encouraging discussion and curiosity (OpenAI, 2024). |
|
|
CR6.7 Read independently and demonstrate comprehension of various informational texts, including instructional and non-fiction materials. (Saskatchewan Ministry of Education, 2011e) |
Students read a simple text or story about robots and machines. Then, they create a Teachable Machine model for two or three objects they read about and share how it works, connecting their learning from the text to their model (OpenAI, 2024). |
|
|
Social Studies 5 |
RW5.2 Hypothesize about future economic changes in Canada, including the impact on industries, demographics, and resource use. (Saskatchewan Ministry of Education, 2010) |
Students use Teachable Machine to identify different community helpers (e.g., teacher, doctor, farmer) and discuss what kinds of jobs they think will be important in the future, using their model to spark ideas about Canada’s future (OpenAI, 2024). |
Note. ChatGPT, an AI language model, provided insights into the practical application of Saskatchewan curriculum outcomes with the GTM. These are only a few examples from a variety of different Elementary Saskatchewan curricula.
The success of GTM in classrooms can be assessed using rubrics that focus on the students’ ability to build functional models, understand the implications of their data choices, and critically reflect on the biases or limitations of their models. Having students reflect in journals on what they did, the problems they encountered and how they solved those problems may be a useful strategy in assessing this comprehension.
Example Rubric for Environmental Science Project Using Google Teachable Machine
|
Criteria |
Exemplary (4) |
Proficient (3) |
Developing (2) |
Beginning (1) |
Score |
|---|---|---|---|---|---|
|
Understanding of Environmental Concepts |
Demonstrates a thorough understanding of key environmental concepts relevant to the project. Clearly connects concepts to the application of Teachable Machine. |
Shows a solid understanding of environmental concepts with clear connections to the application of Teachable Machine. |
Demonstrates a basic understanding of environmental concepts, but connections to the project are vague or unclear. |
Shows little to no understanding of relevant environmental concepts. Connections to the project are missing or incorrect. |
/4 |
|
Application of Teachable Machine |
Uses Teachable Machine creatively to address an environmental issue. The model is well-trained, effectively classifying data or images with high accuracy. |
Applies Teachable Machine to address an environmental issue with a functional model that demonstrates clear understanding. |
Basic use of Teachable Machine with a working model, but the application is simplistic or has limited relevance to environmental issues. |
Limited or unsuccessful use of Teachable Machine, resulting in a poorly functioning or incomplete model that fails to address an environmental issue. |
/4 |
|
Problem-Solving and Innovation |
The project shows original thinking and innovative solutions to environmental monitoring problems, going beyond basic expectations. |
Shows some creative problem-solving and innovation, applying technology to address an environmental issue or scenario. |
Attempts to solve a problem using technology but lacks creativity or innovation. The solution is straightforward. |
Little to no evidence of creative problem-solving. The project does not effectively address the environmental issue. |
/4 |
|
Technical Skills and Model Training |
Demonstrates excellent technical skills in training the model, showing clear understanding of how inputs impact the outputs. Model accuracy is high. |
Shows good technical skills in model training, with some understanding of input/output relationships. The model is mostly accurate. |
Demonstrates basic technical skills, but the model is only moderately accurate or effective. |
Lacks technical skills or understanding of how to train the model. The model is inaccurate or ineffective. |
/4 |
|
Presentation and Communication |
Presents information clearly and confidently. The project is well-organized, and the student effectively communicates their understanding of environmental issues and the technology used. |
Presents information clearly, with a fairly organized project and good communication of key points about environmental issues and technology. |
Presentation is somewhat disorganized, and communication of understanding is unclear or incomplete in some areas. |
Presentation is disorganized or incomplete. The student struggles to communicate their understanding of environmental issues or technology. |
/4 |
|
Connection to Curriculum Outcomes |
The project thoroughly aligns with the curriculum outcomes for Environmental Science 20, demonstrating deep engagement with course material and issues. |
The project aligns with curriculum outcomes and shows understanding of environmental science material. |
Limited connection to curriculum outcomes, with some gaps in understanding of environmental issues. |
Little to no connection to curriculum outcomes. The project does not demonstrate understanding of relevant course material. |
/4 |
Note. The assessment rubric for the Environmental Science project using Google Teachable Machine was adapted from a framework generated with the assistance of ChatGPT (OpenAI, 2024).
Responsible use of AI
As mentioned previously, GTM operates entirely ‘on-device’, meaning that webcam or microphone data does not leave the user’s computer. This setup makes it ideal for discussions about data privacy, as students can explore how some technologies collect, sell and transmit personal information. It also initiates a conversation about ethical concerns, such as how data collection could lead to privacy violations or potential biases in AI models that are trained on limited or biased data.
Future Research and Innovation
AI and machine learning will continue to evolve, with future advancements likely to further simplify model building for educators and students alike. Emerging trends such as explainable AI may become more common. Explainable AI makes AI decisions more transparent by providing information of where the AI references material and why AI makes certain decisions (Bhat & Long, 2024).
Summary
Acknowledgements
This chapter was completed with the assistance of various AI tools. The tools were used ethically and responsibly during this learning process.
- Images: Google Gemini was used to create images. This program uses AI to generate an image from text prompts.
- Writing & Ideas: Google Gemini was used to provide feedback, generate ideas, and check for grammatical errors.
- Summary Video: Photoleap AI was used to create images. This program also uses AI to generate images from text prompts. ChatGPT was also utilized to organize a script from the information created by the authors in this chapter.
Open Researcher and Contributor ID (ORCID)
Curtis Clements https://orcid.org/0009-0004-0688-9864
Curtis Clements holds a B.Sc (Biology), a B.Eng and a B.Ed and is currently working on his Masters in Education Technology and Design (M.ETAD). He has been teaching secondary science, design and robotics for 18 years.
Jordan ven der Buhs https://orcid.org/0009-0003-7622-1133
Jordan ven der Buhs holds a Bachelors of Science in Biology (B.Sc.) with a minor in Chemistry as well as a Bachelors of Education (B.Ed). He is currently working on his Masters in Education Technology and Design (M.ETAD). He has 8 years experience teaching every grade in the school system with a focus in secondary science and outdoor education.
References
Bhat, M., & Long, D. (2024, July). Designing interactive explainable AI tools for algorithmic literacy and transparency. In DIS ’24: Designing Interactive Systems Conference. IT University of Copenhagen, Denmark. https://doi.org/10.1145/3643834.3660722
Forchhammer, S., Abu-Ghazaleh, A., Metzler, G., Garbe, C., & Eigentler, T. (2022). Development of an image analysis-based prognosis score using Google’s Teachable Machine in Melanoma. Cancers, 14(9), 2243. https://doi.org/10.3390/cancers14092243
Google. (2024). Google Gemini (Sept 25 version) [Large Language Model]. https://gemini.google.com/app
Google (n.d.).Google Teachable machine [Web-based tool]. https://teachablemachine.withgoogle.com
Google. (2019, November 7). Teachable Machine 2.0: Making AI easier for everyone [Video]. YouTube. https://www.youtube.com/watch?v=T2qQGqZxkD0
Martins, R. M., von Wangenheim, C. G., Rauber, M. F., & Hauck, J. C. (2024). Machine learning for all!—Introducing machine learning in middle and high school. International Journal of Artificial Intelligence in Education, 34(2), 185–223. https://doi.org/10.1007/s40593-022-00325-y
Ng, D. T. K., Su, J., & Chu, S. K. W. (2024). Fostering secondary school students’ AI literacy through making AI-driven recycling bins. Education and Information Technologies, 29(8), 9715–9746. https://doi.org/10.1007/s10639-023-12183-9
OpenAI. (2024). ChatGPT (October 2024 version) [Large language model]. https://chat.openai.com/
Payne, B. H. (2019). An ethics of artificial intelligence curriculum for middle school students (Supervised by C. Breazeal). MIT Media Lab. https://doi.org/10.5281/zenodo.3751050
Saskatchewan Ministry of Education. (2009). Science 9 curriculum. https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2010). Social Studies Grade 5 curriculum. Retrieved from https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2011a). English Language Arts Grade 4 curriculum. Retrieved from https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2011b). English Language Arts Grade 5 curriculum. Retrieved from https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2011c). English Language Arts Grade 6 curriculum. Use oral language to interact appropriately in pairs or groups, exploring others’ ideas and completing tasks. Retrieved from https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2011d). English Language Arts Grade 6 curriculum. Retrieved from https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2011e). English Language Arts Grade 6 curriculum. Retrieved from https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2011f). Practical and Applied Arts: Robotics and Automation curriculum. Retrieved from https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2016a). Computer science 20 curriculum. https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2016b). Information processing 10/20/30 curriculum. https://www.curriculum.gov.sk.ca
Saskatchewan Ministry of Education. (2017). Environmental science 20 curriculum. https://www.curriculum.gov.sk.ca
Webster, B. (2019, Nov 7a). Teachable Machine tutorial: Bananameter Barron Webster. Medium. https://medium.com/@warronbebster/teachable-machine-tutorial-bananameter-4bfffa765866
Webster, B. (2019, Nov 7b). Teachable Machine Tutorial: Snap, clap, whistle – Barron Webster. Medium. https://medium.com/@warronbebster/teachable-machine-tutorial-snap-clap-whistle-421
