Art-Making Generative AI and Instructional Design Work: An Early Brainstorm
0 likes173 views
The document discusses generative AI in art-making, particularly focusing on generative adversarial networks (GANs) and their applications and controversies. It highlights ethical considerations in using AI-generated content for instructional design, including copyright issues and biases. Additionally, it emphasizes the need for proper training and understanding of these technologies for instructional designers to effectively incorporate them into educational materials.
Art-Making Generative AI and Instructional Design Work: An Early Brainstorm
- 1. Art-Making Generative AI and Instructional Design Work An Early Brainstorm
- 2. 2
- 3. 3
- 4. 4
- 5. 5
- 6. About Art-Making Generative AIs 6
- 7. Generative AIs • Generative AIs are modeled out in different ways: 1. neural with a deep learning model to train / learn on samples from a training dataset and from that learning to generate new ones based on a statistical data distribution (with machine learning / computational data pattern finding across a range of complex dimensions)* 2. symbolic or algorithmic, with a programmer setting the parameters and an autonomous system generating samples within those parameters (Aggarwal & Parikh, 2020) There are combined “neuro-symbolic” approaches and additional approaches outside these categories, too. 7
- 8. Generative Adversarial Networks (GANs) • A common type currently is a Generative Adversarial Network (GAN). The program is trained on (textual/visual/gameplay/music/sound, etc.) data which is curated and often large-scale (human-generated). • The datasets are using of one modality, although cross-modal ones would make sense, too. • The program “learns” about details of the visuals from the training dataset / imageset, often using neural networks, which are a tool in the artificial intelligence (AI) toolkit. 8
- 9. Generative Adversarial Networks(cont.) • The GAN is generally comprised of a generator and a discriminator. The generator (G) creates (textual/visual/gameplay/music/sound, etc.) contents, which are judged by the discriminator (D) against certain standards (established by the training data) for fidelity. • With each work, both the generator and the discriminator become more efficacious (so there is a cooperative aspect to this adversarial- ness…in the sense that competition makes entities stronger). • This is modeled after a two-person zero-sum game…a minimax context…in game theory. 9
- 10. Generative Adversarial Networks(cont.) • A GAN may have additional rules written into it, for issues such as compositing, lighting, and other aesthetic aspects. • A GAN may have additional rules written into it to keep it from generating offensive images, from communicating stereotypes (from the training data), and so on. • Creative Adversarial Networks (CANs) is sometimes used to refer to such AIs that use these technologies to generate text, visuals, games, music, sound…and other dedicated contents. 10
- 11. Machine Art-Making • A subset of generative AIs are designed to make digital visuals, which may be photorealistic, artful (with certain styles), and so on, made up out of whole cloth. • Some of these take inputs from outside (open systems), such as text prompts (and / or with visual prompts) and various parameters to generate the desired image. • Many are available on the Web (based on different terms): Deep Dream Generator, CrAIyon (formerly WALL-E), Midjourney, Stable Diffusion, and others. 11
- 12. Some Controversies about Art-Making Generative AI Tools • Some were built using copyrighted contents (and without the permission of those whose works were captured and used as training data). • Some create likenesses of recognizable people (faces, physiques, voices, and such, separately…and in combination) without their permission. • Some have no guardrails and may show adult content. • Some amplify stereotypes. • Some amplify biases. • Some amplify hate speech. 12
- 13. 13
- 14. 14
- 15. Some Controversies in the Academic Space 15
- 16. Some Controversies in the Academic Space • How can art-making generative AIs be cited when they are used for the creation of a refined visual used for teaching and learning? What if the image is only used partially? What if the image is used only as a reference or a concept or an inspiration? What if the image is used to seed another AI image? • How can an artificial image be described to learners to not misrepresent visual contents? What if such an image is no longer behind an LMS or CMS log-in and “escapes” into the wild? (“LMS” refers to a “learning management system,” and “CMS” refers to a content management system.) 16
- 17. Some Controversies in the Academic Space (cont.) • How can AI-generated works not be confused with the authentic thing in publishing? In student work? • Current forensics do not yet enable accurate identification of such works as machine-generated at-scale (with speed and accuracy). (There are systems to authenticate “real” works…and reverse image searches and other tools keep a memory of online images…but “deep fakes” cannot be fully identified as-yet. “Shallow fakes” either.) • What are the implications of AI-generated art on professional human artists? On potential human de-skilling? 17
- 18. Some Current Professional Ethics • No plagiarism • No fraud • No misrepresentations • No defamation of others (egos or entities) • No untruths • No slander • No defamation of others • No threats • No crying fire in a theatre 18
- 19. Some Current Professional Ethics(cont.) • Control against negative learning (or misapprehensions) • Stringent accuracy • Inclusiveness in the teaching and learning • Respect for others in the teaching and learning (and just being a person in the world) • Accessibility of all learning contents (digital and analog) • Usability for all learners (such as based on “universal design”) • And others… 19
- 20. 20
- 21. About Instructional Design Work 21
- 22. Instructional Design Work • Instructional design work involves a complex skillset and plenty of collaboration on cross-functional teams. • The skills include the following: research, writing, grant-writing, data analysis, analytics, learner modeling, learning design, learning development, learning deployment, adhering to legal and technical standards, intercommunications, documentation, coding / scripting, photography, alpha testing, beta testing, and others. 22
- 23. ID Preferences for Content • 1) Self-created contents or those from known entities are preferred because of the clear sourcing and provenance. There are the raw files…and the legal releases…and other contents, with clear documentation. • 2) Free contents from the U.S. government, with legal releases, clear provenance, professional archival, and long-term availability. • 3) Free contents from open-source repositories…but with the limitation of unclear provenance. (Non-stock-image-sharing corporations can have very restrictive releases on content…which require constant renewal.) • 4) Stock images (free and commercial) for decorative applications. 23
- 24. 24
- 25. Some Professional Ethics in Using Art-Making Generative AIs in Instructional Design Work Early Thinking 25
- 26. Citations When? • The presenter created a continuum between “Ephemeral / Immaterial” on one pole and “Realized / Material” on the other in terms of the visuals. • If a generated image is only used conceptually (1), stylistically (2), as rough drafts (3), as references (4), perhaps the citation is not needed, since the world is full of visuals and inspirations. • If an AI-generated work is “partially finished” (5) or “refined finished” (6), attribution is necessary. 26
- 27. Ephemerality to Materiality Continuum for Harnessing Art-Making Generative AI for Instructional Design Ephemeral / Immaterial Realized / Material Conceptual (1) Style (2) Roughs (3) Reference (4) Partially Finished (5) Refined / Finished (6) Inspiration Ideas Concepts Brainstorming Motivation Exemplars Style extraction Style application / style transfer Draft Wireframe Digital outline Storyboard Reference visual Trace from Used as a seeding visual (to create other AI-based visuals) Usable in part Usable as a part (requires extraction; requires compositing) Layout Refined work Publishable Usable (without much editing or revision or transcoding) 27
- 28. Training Instructional Designers • Any training should help instructional designers understand appropriate uses of art-making generative AIs. • It should include knowledge of how these systems were created and how they function. 28
- 29. 29
- 30. In a Time of Change 30
- 31. Changing Technological Affordances • This work is a first draft. • The technologies are changing in real time. • Many other art-making generative AIs have come to the fore. • It looks like multimodal generative AIs are forthcoming soon, if they haven’t already emerged. (Those that create videos are multimodal machine creatives.) 31
- 32. References • Aggarwal, G., & Parikh, D. (2020). Neuro-symbolic generative art: a preliminary study. arXiv preprint arXiv:2007.02171. 32
- 33. Communications and Contact • Dr. Shalin Hai-Jew • ITS • Kansas State University • 785-532-5262 • shalin@ksu.edu • This work is based on a draft chapter titled “Professionally Ethical Ways to Harness Art-Making Generative AI to Support Innovative Instructional Design Work.” This work is being considered for publication at the moment of this sharing. • All the visuals in this work were created using an art-making generative AI but were also edited to look like illustrations (down to the pixel level). 33