Creativity Enhanced By Artificial Intelligence

Creativity Enhanced By Artificial Intelligence: Learn how to use AI to explore new forms of visual expression

Image created by a 5th year pupil from AE Venda do Pinheiro using Artbreeder.

AI technologies are widely used and affect our interactions with the real and digital worlds,  from online recommendation algorithms in social media to image recognition, decision support and other algorithms. AI is widely mythified as machine intelligence and conscience, yet its capabilities and dangers (especially ethical ones) are seldom discussed. With this learning experience, we aim to challenge pupils to discover AI tools within the arts, as well as gaining a deeper understanding of these technologies, using creative projects as starting points.

Lesson Plan:

Estimated duration: 90 minutes per project. Projects can be used as individual or as linked activities.

Age Level: Upper primary-Lower Secondary

Learning Objectives:

• Understanding roles, capabilities and problems related to Artificial Intelligence technologies;
• Understanding new aesthetic frontiers brought by AI-enhanced visual media;
• Appropriate AI tools as visual expression media;
• Enhancing Creativity and Artistic Expression using digital tools;
• Demystifying AI technologies.

Teacher’s role: introducing AI creativity tools; challenge pupils toward a deeper understanding of AI capabilities and problems.

Student’s role: Explore personal creativity using AI tools; reflect, based upon experience, on AI capabilities and problems.

Initial discussion: What is AI? Do you know examples of these technologies in your daily life? Can machines think? Can machines be creative?

Projects

1 - You Sketch, the AI Draws

Use the Autodraw web app to sketch shapes, and choose between the algorithm suggestions to create a drawing.

What is happening? How can the AI understand what you want to draw? This simple app shows us how AI systems understands data, and makes decisions based on inputs. It all comes down to pattern recognition. The algorithm is trained using a broad set of shapes. And the training never ends, it self-reinforces its learning with the user's input. Everytime you choose an image from its suggestions, you are teaching the algorithm how to better understand uncertain shapes.


2 - Dream Like a Machine

Choose a photo, either by taking one yourself with your phone, or by searching the web. Use the Deep Dream Generator app and run your imagem through it's algorithms - deep dream, neuron or valyrian. Iterate the results several times, until you're satisfied with the output.

What is happening? Why is the AI turning a photo into an unreal image? Image recognition algorithms, such as facial recognition or shape recognition, need to be trained with sets of images in order to learn how to analyse, understand and extrapolate a result. These systems are limited in scope by the quality of the datasets they're trained on. Image recognition algorithms, for example, are notorious for its vulnerability to unintentional racial profiling and false positives. But deep dream algos do something a bit different: they're interpreting what they "see" in the image input based on the types of shapes they've been trained upon - essentially, describing not what is in the image, but what they think is in the image.

3 - Photo Surrealism

Access the Art Breeder web app and in Create, choose the General category. Then, in Creation Method, use Compose. The algorithm will generate some random images. You can select different Parent Images, and mix them with Genes to achieve intriguing results. Use the Chaos slider for more unexpected results. All the images you see are generated in real time by a GAN AI.

What is happening? How is the computer creating these images seemingly out of nowhere? GAN neural networks can be trained using extensive sets of images, so that they can later extrapolate new data similar to training. The technique pits two algorithms: one generates new data, based on user input and training sets, and the other judges the output to better match the users intentions. These types of algorithms can be used to generate new images, remix existing images in unexpected ways, or, if  using the work of an artist as training dataset, to understand its aesthetic structure, giving artists and art historians deeper knowledge, or to apply an artist’s personal style to new kinds of imagery (a very cool example are the style transfer apps that turn photos into pictures in the style of major artists).