Google's Teachable Machine and AI Ideas for Educators

Tinkering with Google's Teachable Machine in a live 3-min demo

We actually come into contact with AI far more often than we imagine:

• Microsoft PowerPoint's Design Ideas/ Suggestions (similar implementations in Canva, Adobe Suite etc)

• Email Inbox Spam Filter/ automated tagging of emails

• Autopilot for planes

• Live translation using the phone (Google Translate, Papago, etc)

• ReCaptcha...? Every wondered why you have to do it twice? In this case, the first time confirms you are human. The second, actually, is your effort to help to train future AI models - you can try it next time, any response in your second try should pass the test!(:

In today's session, we went through some use cases:

1 Google's Teachable Machine

Use cases for different teachers:

• Music Teachers: train an audio model to help identify sounds coming from respecitve instruments. As bboy (dancer), I am personally awful at identifying which instrument is responsible for the sounds for layered music and think would find it helpful! Example video here 

• Environmental education and sustainability: you software + hardware to create a simple sorter

• Physical Education Teachers: train a model to identify dance steps or what is considered good or bad form

• My personal takeaway is that it is now easier than ever to train your own computer vision models as evidenced by the demonstration of a model with 500 image samples done in 3mins shown in the top banner image. 4 years ago, this would have taken 4hrs (see Preston's Chicken Rice Classifier, 2018)

Next steps: beyond training, one may consider extensions and deploying the model with hardware such as Arduinos/ Raspberry Pi. For some reason, Google staff is obsessed with showing how you can use Teachable Machine to sort marshmallows and cereal. But just imagine the possibilities! (:

2 Preview of AI and Natural Language Processing (NLP)

As educators, we work with a lot of text data when we prepare learning materials and administer checks for understanding.

Next week, we will talk a little more about Language Models like Minerva which functions well in answering + providing step-by-step solutions to problems across a range of academic disciplines from Mathematics, Physics, Biology and more. On top of that, an added perk that piqued the interest of Math educators today was how the scientific notation was preserved:

You can read a little more here about Google's Minvera below:

Text analysis, NLP, language models are rich and definitely will take more than one session. Some actual applications I use includes a transcriber/ summarizer especially for longer videos that I plan to incorporate in teaching and learning:

We read faster than we listen and that saves some time when preparing lesson material.

Regardless of whether you run it on a Mac, PC or Linux, you can get a nifty transcriber. + summarize up and running in your (personal) computer. You will have restricted use of the command line or terminal on your school computers. Tutorial for this use case of assemblyai here

Parting questions:

• How can such no-code AI powered computer vision tools be used in our T&L?

• Is it fair for us to use AI tools like (copy.ai or hypothenuse or open.ai) to write student termly comments?

• Can we ever guard against AI-assisted written essays?