Ce diaporama a bien été signalé.
Le téléchargement de votre SlideShare est en cours. ×

Human-Machine Collaboration: Using art-making AI (CrAIyon) as cited work, or inspiration, reference, base


Consultez-les par la suite

1 sur 56 Publicité

Human-Machine Collaboration: Using art-making AI (CrAIyon) as cited work, or inspiration, reference, base

Télécharger pour lire hors ligne

It is early days for generative art AIs. What are some ways to use these to complement one's work while staying legal (legal-ish)?

It is early days for generative art AIs. What are some ways to use these to complement one's work while staying legal (legal-ish)?


Plus De Contenu Connexe

Similaire à Human-Machine Collaboration: Using art-making AI (CrAIyon) as cited work, or inspiration, reference, base (20)

Plus par Shalin Hai-Jew (20)


Plus récents (20)

Human-Machine Collaboration: Using art-making AI (CrAIyon) as cited work, or inspiration, reference, base

  1. 1. Human-Machine Collaboration Using art-making AI (CrAIyon) as cited work, or inspiration, reference, base Shalin Hai-Jew Kansas State University
  2. 2. 2
  3. 3. 3
  4. 4. 4
  5. 5. 5
  6. 6. 6
  7. 7. Presentation Overview Generative AI can create natural language text in various formats and voices and perspectives (ChatGPT) and emulative style-transferred visual images (CrAIyon, DALL-E, MidJourney, and others). Controversies are swirling around various aspects of generative AI. 7
  8. 8. Presentation Overview (cont.) How the generative AI tools are made and run: • uses of copyrighted seeding texts and visuals in databases to train the AIs (without the permission of the original authors, in some cases), • guardrails around generative contents (such as those against x-rated content, against hate speech, against various dimensioned stereotypes, etc.) vs. those without any guardrails, 8
  9. 9. Presentation Overview (cont.) How the generative AI tools are used: • academic honesty, citations, • commercial applications, • authorship and crediting (and rewards and liabilities) 9
  10. 10. Presentation Overview(cont.) • A typical sequence in using a text-seeded generative AI that creates digital visuals is to use seeding phrases to describe the desired visual (often multiple iterations)…selecting the image…downloading the image as a .webp (“weppy”) format (neither raster nor vector), and directly using the image with citation…or using the image as an inspiration, reference, or base. (One creates a derived image, by borrowing some visual concepts from the generative AI.) “Photorealistic” asks the AI for create an image that looks like an actual photo. “after Picasso” or “in the style of Georgia Totto O’Keeffe” asks for a style transfer from the known works of the artist into a different context. 10
  11. 11. Presentation Overview (cont.) • [Please see “Co-Creating Common Art with the CrAIyon AI” on SlideShare: https://www.slideshare.net/ShalinHaiJew/cocreating- common-art-with-the-craiyon-ai for a clearer visual gist of this phenomenon.] • The presenter wrote an article titled “CrAIyon: Putting an art-making AI through its paces” on the C2C Digital Magazine at (https://scalar.usc.edu/works/c2c-digital-magazine-fall-2022---winter- 2023/craiyon-paces). As a former college faculty member and current instructional designer, the presenter introduces the topic and throws a conversation around the complexities. 11
  12. 12. Giving CrAIyon a Roll 12
  13. 13. What are some text prompts that you might use with CrAIyon? • Remember that this tool enables a wide range of languages and symbols and combinations. • This tool can engage in “style transfer,” so if you have a particular visual artist you like, you can evoke that artist’s name. • You can evoke various materials that you want the generative AI to produce. • Go a few rounds. Iterate from your original text prompt. Or try some highly variant prompts. • Any visuals worth downloading (.webp)? 13
  14. 14. Impressions? (first or otherwise) • What are you noticing about the machine-generated visuals? • Did anyone try “photorealistic” as one of their prompts? • Did anyone try “after” a particular artist style? • Did anyone try a particular material effect? 14
  15. 15. Impressions? (first or otherwise) (cont.) • Do you feel like you could export something practically useful? Impractically useful? • What do you like? Why? • What do you dislike? Why? • What are some practical applications for you? 15
  16. 16. Impressions? (first or otherwise) (cont.) • Are you starting to notice some visual tropes that the AI has? • Whare are some strengths of the generative AI (based on impressions)? What are some weaknesses? • If you could change the tool capabilities, what would you do, and why? 16
  17. 17. Debate around Human-Machine “Collaboration”…and Credit In terms of generative AI 17
  18. 18. 18
  19. 19. Generative AI… • …refers to computational programs that emulate humans by generating various content (text, audio, imagery, motion visuals, video, and other elements, including combines audiovisual ones) • The AI programs learn through large samples of big data, so it can emulate nuances of understandings…although such programs are in early years (some compare them to “teenagers”). 19
  20. 20. Generative AI (cont.) • Generative adversarial networks (GANs) do not only generate contents to particular objectives, but they have a built-in test of quality / veracity to desired outcomes against which the generative aspect strives to improve. • Generative AI can generate content into ∞ and beyond, but there may not be enough human attention (attention economy) to enjoy the various works. The works need to be meaningful to humans. 20
  21. 21. Human in the Loop • As it stands, the generative AIs may be prompted in a mix of ways: • Textual prompts to help the AI know what is desirable in terms of visuals, so it can output various visuals for consideration • Visual prompts to help the AI have examples of the visual gist and perhaps content that may be desirable, so it can output various visuals for consideration • Combined inputs (textual and visual) • And others 21
  22. 22. 22
  23. 23. Debate around Human-Machine “Collaboration” • How much is the human responsible for the artwork vs. the computing machine? • Does the computing machine have an agentic role given its design as “artificial intelligence”? • Does it matter if the work is run unedited directly from the generative AI download? • Does it matter if the work is edited and changed, superficially? • Does it matter if the work is edited and changed, down to the pixel level? 23
  24. 24. Debate around Human-Machine “Collaboration” (cont.) • Who has the first idea (remember that the machine can turn out something serendipitous and unconceptualized by the person who inputs the prompt)? • What if the content (text or visual) is created sui generis (unique, one-of-a- kind) by the computing machine? • What if the content (text or visual) is created ex nihilo (from nothing, specifically without inputs by people’s works or by people’s prompts)? • Who is responsible for the aesthetics? The composition? The social angle? And why? • Who is the “animating agent” and why? 24
  25. 25. …And Credit Human • Traditionally, artists are credited for original single-artist and team-created fine artworks. • The artist’s name and personality and history serve as an artist brand. Machine • Can AI be credited for (fine and other) artworks? • What are the implications for renown? For money? For payment? For prizes? • Should the programmers behind an art-making AI be credited for the artworks? 25
  26. 26. …And Reward…and Liability Human • In tradition, the artist bears responsibility for the work to some degree…and then the owner thereafter… Machine • In tradition, computing machines do not generally bear reward or liability. Rather, their owners do. 26
  27. 27. Reapportionment of Credit? Rewards? Liabilities? 27 • Should credit be reapportioned? Rewards? Liabilities?
  28. 28. Drawing Lines…Or Not… • How should schools address generative AI? • How should professions? (How can people show their value especially while in competition with generative AI works?) • How should publishers? • Or is this too soon to start drawing lines? 28
  29. 29. What Would You Do, and Why? 29
  30. 30. 30
  31. 31. 31
  32. 32. Q1: How would you use generative AI in your respective contexts? • And then, do you feel the need to disclose that you use an AI for assistance or some other reason? • Do you count neural filters (in digital image editing) as AI (after all, various neural networks were used to tune the digital image editing, to transfer styles, to recolor photos, to change hue / saturation, to change the depth-of-field, to enable deep zoom, to enable facial editing, and others)? 32
  33. 33. Q2: If you teach or edit, what would your policies be about generative AI? Why? • Explain why your policy is timely…but fair…and pro-learner / pro- author (or researcher). • How would you differentiate something that is created from generative AI vs. not? • What are ways to manage this issue of generative AI in teaching and learning? 33
  34. 34. Q3: What can learners learn from generative AI if it is harnessed as a learning tool? • Given the state of the world, with advanced technologies coming to the fore in the 4IR (Fourth Industrial Revolution), how can people adapt and better position themselves for these fast-arriving changes? • How would you as a learner harness generative AI? • Are there ways to make generative AI more friendly for learners? • Much of generative AI has been democratized, with free web-facing portals and apps. Are there other ways to extend the benefits to the world? 34
  35. 35. Q4: What are some ways to position for the impact of generative AI on human jobs? • Generative AI is thought to squelch some white-collar jobs while creating other jobs. What do you see as the possible impact of generative AI on your particular work and professional circumstance? Why? • If you have funds to retrain, how would you use those funds? • What are some other AIs and technologies on the horizon that may well have a big impact on the near-future? The mid-term future? 35
  36. 36. Q5: Any questions for the group that have not yet been asked? 36
  37. 37. My Own Usage of CrAIyon 37
  38. 38. 38
  39. 39. 39
  40. 40. Some Real-Life Applications Practical: Visuals in an Open-Source, Open-Access Magazine • An open-source and open-access publication needs some light visuals to break up the gray text. • I have gone through social imagery datasets and not found anything that the author likes. • I go to generative AI to output a few ideas, and then I create a visual image to use with the generated image as an inspiration or a reference image (used for its lines which I capture via trace). Impractical: Amusement, Entertainment • I want to be amused. I can go to a generative AI with various text prompts to see what it will output. Some can be highly surprising and funny. (Generative AIs can have a strange sense of physics and bodies and hands and eyes.) • A generative AI can offer high entertainment value. 40
  41. 41. Some Real-Life Applications(cont.) Practical: Learning about Digital Image Editing • I am practicing digital image editing. I want to learn new functions in some very complex digital image editing software. • I create some unusual prompts, and then I use the imagery to apply new learning. • I also want to learn more about AI…and how it interacts with language. Impractical: Open-Sharing of Social Imagery • I want to create visuals for open- sharing on the Social Web. I conceptualize an idea…and generate some images…and then use those images to serve as references for an original visual. • Anything used more directly is credited as from AI. 41
  42. 42. Some Real-Life Applications(cont.) Practical: Learning about Visual Thinking • In other times, I want to learn how to create moods and other visual effects. The generative AI can offer fresh insights. • Ditto with various visual communications messages…some banal…and some more creative… • Sometimes, I want to understand word definitions in a visual sense. Impractical: Wasting Time • In between work, sometimes, it helps to just put in a text prompt and have the generative AI doodle something (in under 2 minutes). • Sometimes, there is the additional benefit of learning…but a distraction is sometimes very welcome. 42
  43. 43. Some Initial Personal Observations • There is not a straight-line to the visual I usually want, which tends to be somewhat artsy and strange. • Iterating from an initial prompt with additional prompts may not get me much closer. • Word order only seems to matter sometimes. • It does help to know the lingo for digital image editing and analog image making (sketching, drawing, painting, mixed media, and others). • The generative AI is too often literal and less symbolic and less figurative. It is almost never poetic. • Occasionally, there is a fluke that can result in something very aesthetically pleasing and unexpected. 43
  44. 44. Some Initial Personal Observations (cont.) • The generative AI, CrAIyon, IMHO, has its own quirks and signatures (based on my half-year of experiences with it). • It draws famous (and non-famous people badly, with weird eyes and bad hands when it is trying to be “photorealistic”. • It has a hard time drawing human proportions, too. • It draws animals poorly, without a real-world sense of anatomy. • It blurs much of a small image. (Blurring reads as non-committal.) • It seems to like spirals to stand in for something precise. • It does not do clean gridlines. • It does not seem to do gradients. 44
  45. 45. Some Initial Personal Observations (cont.) • It seems to have some bias towards spooky things. • It sometimes goes full-bore stereotypical. • It sometimes pulls people in its own quirky universe instead of aligning with the real world. • Between figurative and abstract, it seems to do the middle ground fairly well but not those on the extreme ends. • It does not handle text (probably there is some throttling back against the uses of any language…so it has a doodle-y version of its own). 45
  46. 46. Some Initial Personal Observations (cont.) Some of my purposeful and unpurposeful misadventures: • When I typed in “fairytales for men,” the AI generated 9 visuals of men in various fashion walking down catwalks. • When I typed in “wild,” the generative AI came up with a variety of apparent mammals that do not exist except in its own imagination. These are patchwork animals, with patches of skin, some horns, different hooves, and so on. 46
  47. 47. Some Initial Personal Observations (cont.) • Truth to tell, I am mostly generally impressed and appreciative. 47
  48. 48. 48
  49. 49. The Upsides • So to end on an upside, I like CrAIyon because of the following: • It is always game. Whatever prompt I put in, it always kicks up something. • It outputs visuals I can practically use in most cases. • It sometimes outdoes itself and offers me something that is easy to make look artful. [But this is after years of my work in the digital image editing space.] • It is surprising now and again. 49
  50. 50. The Upsides (cont.) • While it has challenges with some shapes, it composites well. • It handles values (lights and darks) well. • I think it’s working on its color sensibilities. Often, it can be quite muted. It does not take risks in its color choices. 50
  51. 51. The Upsides (cont.) • I for one will keep right on learning…because it helps to be ready. And this saves so much on time and analog art materials. Well, I do have some time sinks where I spend an hour digitally editing a work until I am satisfied. • We should test limits… • My uses of generative AI (currently) are for “artwork” with a lower- case “a,” nothing aspiring to high art. My uses are for common art. 51
  52. 52. A Caveat • Generative AI really does make my life somewhat easier when I have to chase images, but the jury (general public, decision makers, and others) is still out about the uses of generative AI-supported visuals…so perhaps we need to see how the world shapes out in terms of its approaches. 52
  53. 53. 53
  54. 54. 54
  55. 55. Conclusion and Contact • Dr. Shalin Hai-Jew • Instructional Design • ITS • Kansas State University • 785-532-5262 • shalin@ksu.edu • Note: All the visuals in the slideshow (save the cross-functional diagram) were seeded with CrAIyon visuals but edited pretty heavily in Adobe Illustrator and Adobe Photoshop. Some could have run unedited, but I wanted to change these up from the originals. 55
  56. 56. 56