About the event
We are very sad not to be able to bring you the Artificial Intelligence workshop at Edinburgh Science Festival this year. But there is a huge amount of information and activities available online if you want to investigate yourselves! These exercises are perfect if you want to improve your computer skills, you will need to be fairly comfortable with experimenting to enjoy them.
If you are new to coding why not start with the Micro:Bit simulator – here!
We have been creating content that introduces people to coding and digital creativity for a long time. A couple of years ago we partnered with Google to open an Innovation Hub in a university in UAE. (Please note that these cources were written some time ago now. There are up-to-date tutorials on how to use Google's machine learning software for yourself here.)
Once you’ve got an overview of machine learning and how it is useful, why not try building your own? We have been working on a model that can recognise different types of dog. This is an example of supervised learning – where a machine is trained on a data set of labelled examples and can then recognise new examples from the same categories.
Train it on the Stanford Dogs Dataset – a huge collection of sorted images of different breeds of dogs. Download the 700MB Images file. You’ll need software that can decompress a tarball – such as www.7-zip.org.
Or try teaching your browser to play the old mobile game Snake. This is an example of reinforcement learning – where the AI is given input (the state of the game) and a goal (get the high score). It’s similar to how Google trained AlphaGo – possibly the best player of any game ever in the world today.
If teaching things makes you feel too much like being at school, why not explore the creative side of Artificial Intelligence? By using Generative Adversarial Networks (GANs), researchers are able to pit two neural networks against each other – one creates and the other tries to detect the fakes. This way people have created software that can create images and text. There’s room for improvement on the image side, but some of the text can be convincing (and hilarious)
GAN Paint lets you draw with buildings, clouds, windows
There are a range of projects at Deep Angel. You need to sign up but access is free. My favourite is the cat eliminator, which can remove cats from photographs. Why would you want this? I don’t know.
Talk to Transformer is the most sophisticated text GAN on the internet today. It was trained on a huge amount of data from across the internet and knows the difference between Harry Potter and Star Wars, between a news story and a recipe.
Give it a prompt: “Harry walked into the room” or “Rey turned around and smiled”
Find out the news: “The Prime Minister announced today”
Of course it’s trained on the internet so has inherited some of our biases and prejudices. Do you get different results from “She had always wanted” and “He had always wanted”?
Developed with support from