Explore the phenomenon: Creative AI
The term AI, or artificial intelligence, has been in use since the 1950s, but only the past 5 to 10 years have seen the development of algorithms and computers that are fast enough to enable the versatile generation of visual material. We are talking about a generative AI that produces new content such as text, images, audio and even moving images, or video. In this context, an algorithm can be interpreted to mean a computer program or a command given to such a program.
Current AI models that can produce visual material have been taught using a vast amount of images. The LAION-B dataset, for example, consists of almost 6 billion images. This dataset or a taught model is at the core of many AI programs that generate images. Another AI that interprets images has reviewed the collected images and identified and classified the objects in them, such as cats, clouds, flowers and landscapes, but also described features like the mood in the image or the style in which the image has been made. Current AI software can be used to create new images by entering keywords relating to the motifs and the style, for example.
The use of this kind of visual material bank involves questions such as who and what is depicted in the images and where the images come from. The images have been collected from the internet, and especially from various image services available, such as the Flickr photo service, ArtStation, DeviantArt, Instagram and Pinterest. The images uploaded in these services have been used as the teaching material for new algorithms. It is possible that the datasets focus on western imagery, culture, landscapes, people and city life. Imagery from other cultures, such as the cultures in Africa or South America, might not be as available in these services. This has an impact on what kind of visual material these AI algorithms generate.
Another problem that has come up is that of copyright. AI can produce images that look like the ones by a certain artist, after the AI has learned the style from the artist’s work published on the internet. This might cause problems, especially to those artists and illustrators that are still alive. If AI can be used to generate images that are very similar to the work of artists and illustrators, it might endanger their livelihood.
Another potential problem is the generation of deepfake material. It is possible to generate new images that seem to depict a famous actor or a politician, based on previously published photos of the person. There is a lot of visual material of people who often appear in the media, and AI algorithms have probably learned what these people look like. An AI can be taught what a person looks and sounds like.
We can see that the use of AI algorithms is moving towards a more ethical direction. In some image sharing services (e.g. ArtStation), people can already deny the use of their images for teaching an AI. People can also delete their images from certain AI algorithms. There is also an ongoing discussion about making the image of humanity more equal in AI models, so that different sexes, ages, ethnical backgrounds and other factors would be better represented.
The next development, of which some preliminary phases have already come about, is to generate video footage in the creative industry. Even though it is already possible to create impressive images, photographs or even paintings, generating video is still demanding. However, there is a lot of positive potential with an image-generating AI. A designer can generate multiple images using different styles, for example, and make different versions to quickly create various examples as a basis for the work. It might be fun to transfer your own image to a comic, watercolour painting or origami.
Currently, you can test making images with different internet services and applications. Some of the services are free of charge, at least to start off with, and you do not necessarily have to use your own images. Money can usually buy better visual material and images are faster to generate.
It is worth bearing in mind that AI is often used in everyday life. For example, it helps us find the best route in a map application when we move from point A to point B, or suggests movies or TV series in Netflix based on our previous choices. AI also helps voice recognition applications, such as Alexa, Cortana and Siri, understand what we are saying. An example of how image editing AI algorithms are becoming commonplace is the Photoshop image processing software: it includes existing AI-based methods, i.e. neural filters, for editing images. It is likely that the use of AI will only increase in applications in the future. AI poses many challenges but also opens up possibilities.
All in all, a generative AI is an interesting and versatile technology that offers countless possibilities for generating images, audio and video. It is still important to understand that its use includes ethical and practical challenges and things to consider, such as data protection, copyright and cultural diversity. We must learn to understand the potential and limitations of AI to use it responsibly and sustainably in the future.
Links:
Timeline for different AI models that create images (and video): https://www.fabianmosele.com/ai-timeline
Google Imagen Video: https://imagen.research.google/video/
CLIP: an AI that interprets images, helps create a connection between the explanatory text and the image: https://www.pinecone.io/learn/zero-shot-object-detection-clip/
Text: Tomi Knuutila
Image: Siru Tirronen
The media landscape of children and young people keeps changing, with new phenomena following each other back-to-back. Providing pupils with tools for understanding and processing these phenomena is important. This learning package is part of Pathways to New Media Phenomena – Information and Exercise Materials Series. The series includes information and exercises for the teacher and the pupils. You can explore new phenomena in a meaningful way with the help of the How to discuss new media literacy phenomena through pedagogical means method.
CC BY 4.0
