Experts Matter. Find Yours.

AI Art: What Should Fair Compensation Look Like?

New research from Goizueta’s David Schweidel looks at questions of compensation to human artists when images based on their work are generated via artificial intelligence. Artificial intelligence is making art. That is to say, compelling artistic creations based on thousands of years of art production may now be just a few text prompts away. And it’s all thanks to generative AI trained on internet images. You don’t need Picasso’s skillset to create something in his style. You just need an AI-powered image generator like DALL-E 3 (created by OpenAI), Midjourney, or Stable Diffusion. If you haven’t tried one of these programs yet, you really should (free or beta versions make this a low-risk proposal). For example, you might use your phone to snap a photo of your child’s latest masterpiece from school. Then, you might ask DALL-E to render it in the swirling style of Vincent Van Gogh. A color printout of that might jazz up your refrigerator door for the better. Intellectual Property in the Age of AI Now, what if you wanted to sell your AI-generated art on a t-shirt or poster? Or what if you wanted to create a surefire logo for your business? What are the intellectual property (IP) implications at work? Take the case of a 35-year-old Polish artist named Greg Rutkowski. Rutkowski has reportedly been included in more AI-image prompts than Pablo Picasso, Leonardo da Vinci, or Van Gogh. As a professional digital artist, Rutkowski makes his living creating striking images of dragons and battles in his signature fantasy style. That is, unless they are generated by AI, in which case he doesn’t. “They say imitation is the sincerest form of flattery. But what about the case of a working artist? What if someone is potentially not receiving payment because people can easily copy his style with generative AI?” That’s the question David Schweidel, Rebecca Cheney McGreevy Endowed Chair and professor of marketing at Goizueta Business School is asking. Flattery won’t pay the bills. “We realized early on that IP is a huge issue when it comes to all forms of generative AI,” Schweidel says. “We have to resolve such issues to unlock AI’s potential.” Schweidel’s latest working paper is titled “Generative AI and Artists: Consumer Preferences for Style and Fair Compensation.” It is coauthored with professors Jason Bell, Jeff Dotson, and Wen Wang (of University of Oxford, Brigham Young University, and University of Maryland, respectively). In this paper, the four researchers analyze a series of experiments with consumers’ prompts and preferences using Midjourney and Stable Diffusion. The results lead to some practical advice and insights that could benefit artists and AI’s business users alike. Real Compensation for AI Work? In their research, to see if compensating artists for AI creations was a viable option, the coauthors wanted to see if three basic conditions were met: – Are artists’ names frequently used in generative AI prompts? – Do consumers prefer the results of prompts that cite artists’ names? – Are consumers willing to pay more for an AI-generated product that was created citing some artists’ names? Crunching the data, they found the same answer to all three questions: yes. More specifically, the coauthors turned to a dataset that contains millions of “text-to-image” prompts from Stable Diffusion. In this large dataset, the researchers found that living and deceased artists were frequently mentioned by name. (For the curious, the top three mentioned in this database were: Rutkowski, artgerm [another contemporary artist, born in Hong Kong, residing in Singapore] and Alphonse Mucha [a popular Czech Art Nouveau artist who died in 1939].) Given that AI users are likely to use artists’ names in their text prompts, the team also conducted experiments to gauge how the results were perceived. Using deep learning models, they found that including an artist’s name in a prompt systematically improves the output’s aesthetic quality and likeability. The Impact of Artist Compensation on Perceived Worth Next, the researchers studied consumers’ willingness to pay in various circumstances. The researchers used Midjourney with the following dynamic prompt: “Create a picture of ⟨subject⟩ in the style of ⟨artist⟩”. The subjects chosen were the advertising creation known as the Most Interesting Man in the World, the fictional candy tycoon Willy Wonka, and the deceased TV painting instructor Bob Ross (Why not?). The artists cited were Ansel Adams, Frida Kahlo, Alphonse Mucha and Sinichiro Wantabe. The team repeated the experiment with and without artists in various configurations of subjects and styles to find statistically significant patterns. In some, consumers were asked to consider buying t-shirts or wall art. In short, the series of experiments revealed that consumers saw more value in an image when they understood that the artist associated with it would be compensated. Here’s a sample of imagery AI generated using three subjects names “in the style of Alphonse Mucha.” Source: Midjourney cited in http://dx.doi.org/10.2139/ssrn.4428509 “I was honestly a bit surprised that people were willing to pay more for a product if they knew the artist would get compensated,” Schweidel explains. “In short, the pay-per-use model really resonates with consumers.” In fact, consumers preferred pay-per-use over a model in which artists received a flat fee in return for being included in AI training data. That is to say, royalties seem like a fairer way to reward the most popular artists in AI. Of course, there’s still much more work to be done to figure out the right amount to pay in each possible case. What Can We Draw From This? We’re still in the early days of generative AI, and IP issues abound. Notably, the New York Times announced in December that it is suing OpenAI (the creator of ChatGPT) and Microsoft for copyright infringement. Millions of New York Times articles have been used to train generative AI to inform and improve it. “The lawsuit by the New York Times could feasibly result in a ruling that these models were built on tainted data. Where would that leave us?” asks Schweidel. "One thing is clear: we must work to resolve compensation and IP issues. Our research shows that consumers respond positively to fair compensation models. That’s a path for companies to legally leverage these technologies while benefiting creators." David Schweidel To adopt generative AI responsibly in the future, businesses should consider three things. First, they should communicate to consumers when artists’ styles are used. Second, they should compensate contributing artists. And third, they should convey these practices to consumers. “And our research indicates that consumers will feel better about that: it’s ethical.” AI is quickly becoming a topic of regulators, lawmakers and journalists and if you're looking to know more - let us help. David A. Schweidel, Professor of Marketing, Goizueta Business School at Emory University To connect with David to arrange an interview - simply click his icon now.

A.I. and Higher Education: The Rise of ChatGPT

ChatGPT. Maybe you’ve heard of it. Colleges and universities certainly have. It’s the chatbot that uses artificial intelligence (A.I.) technology to generate sentences based only on a brief prompt, writing anything from college-level papers to fanfiction. And as one might expect, the popular chatbot is taking the academic world by storm, raising questions about trust, academic integrity and even the future of college admissions. We turned to Seth Matthew Fishman, PhD, Assistant Dean of Curriculum and Assessment and associate teaching professor in the Department of Education and Counseling at Villanova University, to get his thoughts. Q: What makes ChatGPT different and why is it causing such a stir? Dr. Fishman: The use of chatbots is not a new debate in higher education. But ChatGPT and other similar free software certainly add a complex layer that we are only just now starting to have conversations about. There will be an ongoing debate about trust—Who wrote the material we are reading? To what extent if any, will it impact faculty members? There are also A.I. digital images, graphics, and design—To what extent do these programs impact our creative arts and design programs? I think these fields will mostly embrace A.I., though I can see issues of copyright infringement and artist control/attribution. Q: How are other chatbots being used in academic settings? DF: A.I. use already impacts higher education. If you ask any faculty member teaching a foreign language that requires a translation, they will have tales of work submitted by students who use online translation software. But benefits do exist for students and faculty regardless—we’re able to interact a bit more with others, reducing some language barriers. I expect we will see hundreds of articles about ChatGPT’s impact on education; there are likely several dissertations underway, and I expect to see ChatGPT and similar software cited in papers and likely even in authorship groups. Q: What will the impact of ChatGPT be on the college application and admissions process? DF: I think we’ll see conversations from college admissions professionals on the impact of ChatGPT on higher education admissions. For example, key components of college applications such as essays and writing samples may be impacted. And ChatGPT may also be used to write some rather good letters of recommendation. Q: What does the future hold? Will ChatGPT and similar A.I. programs maintain popularity? DF: I’m curious if A.I. will be used to generate employment cover letters. Additionally, many corporations already use A.I. to sift through candidate applications to narrow down their applicant pools. It may continue to transcend academia. I also expect to hear more from our philosophy and ethics experts to help us better understand the societal and educational implications of using A.I. in these ways. And these kinds of conversations will be had with our students to engage them as partners in the learning experience. We will probably generate new ideas and different perspectives from doing just that.

Squid Game: why you shouldn’t be too hard on translators

By David Orrego-Carmona Squid Game has recently become Netflix’s biggest debut ever, but the show has sparked controversy due to its English subtitles. This occurred after a Korean-speaking viewer took to Twitter and TikTok to criticise the subtitles for providing a “botched” translation, claiming: “If you don’t understand Korean you didn’t really watch the same show.” Only this year, Squid Game, Lupin, and Money Heist – all non-English originals – have consistently been at the top of Netflix’s most-watched shows globally. This growing popularity of productions in languages other than English and streaming platforms investing more in them has led to an increase in the visibility of the work of translators. When it comes to translating films and series, subtitling and dubbing are the most common forms of translation. Subtitles show the dialogue translated into text displayed at the bottom of the screen; while in dubbing, the original voices of the characters are replaced with voices in a new language. Translation is not new to viewers, but the instant, almost frictionless access to different language versions of the same film or show definitely is. Streaming platforms allow viewers to swiftly change from watching a film with subtitles to listening to the dubbed version or the original. This creates an opportunity for viewers to compare the different versions. Why do originals and translations differ? Just because the translation doesn’t say exactly the same as the original, it doesn’t mean it’s wrong. Films and TV series are packed with cultural references, wordplay and jokes that require changes and adaptation to make sure what’s said and seen on screen makes sense across languages. Making allowances and adapting what’s said are common practices in translation because, otherwise, the translators would need to include detailed notes to explain cultural differences. Consider the representations of washoku (traditional Japanese cuisine) which are so beautifully embedded in Studio Ghibli films. While additional explanations about the significance of harmony, kinship and care represented in the bowls of ramen in Ponyo or the soft steaming red bean buns in Spirited Away could be interesting, they might get in the way of a viewer who just wants to enjoy the production. Professional translators analyse the source content, understand the context, and consider the needs of the variety of viewers who will be watching. They then look for translation solutions that create an immersive experience for viewers who cannot fully access the original. Translators, similarly to screenwriters and filmmakers, need to make sure they provide good, engaging storytelling; sometimes that implies compromises. For instance, some original dialogue from season two of Money Heist uses the expression “somanta de hostias”. Literally, “hostia” means host – as in the sacramental bread which is taken during communion at a church service. But it is also Spanish religious slang used as an expletive. Original: Alberto, como baje del coche, te voy a dar una somanta de hostias que no te vas ni a mantener en pie. Literal translation: Alberto, if I get out the car, I’m going to give you such a hell (hostia) of a beating that you won’t be able to stay on your feet. Dubbed version: If I have to get out of the car, I’m gonna beat you so hard you don’t know what day it is. Subtitles: Alberto, if I get out of the car, I’ll beat you senseless. The dubbed version of the dialogue adopts the English expression “to beat someone”. The subtitled version uses the same expression but offers a shorter sentence. The difference between the two renderings reflects the constraints of each form of translation. In dubbing, if the lip movements don’t match the sound, viewers often feel disconnected from the content. Equally, if subtitles are too wordy or poorly timed, viewers could become frustrated when reading them. Dubbing needs to match the duration of the original dialogue, follow the same delivery to fit the gesticulations of the characters, and adjust to the lip movements of the actors on the screen. Subtitles, on the other hand, need to be read quickly to keep up with the pace of the film. We talk faster than we can read, so subtitles rarely include all the spoken words. The longer the subtitle, the longer the viewer will take to read it and the less time they will have to watch. According to Netflix policies, for example, subtitles can’t have more than two lines and 42 characters, and shouldn’t stay on the screen for longer than seven seconds. Additionally, in the above example, the translations do not reflect the reference to religious slang, typical of Spanish culture. Rather than fixating on this reference and assuming it is an essential part of the dialogue, a good translator would consider what an English-speaking character would say in this context and find a suitable alternative that will sound natural and make sense to the viewer. New rules of engagement It is encouraging to see that some viewers are so devoted to the content they watch: foreign films and TV shows help promote cultural understanding and empathy. But not all viewers act in the same way and the solutions provided by the translators need to cater to everyone who decides to watch the show. This leads to different viewing experiences, but it only reflects the reality of watching any culturally charged product, even in our own languages. In English, for instance, consider all the references and nuances that a British viewer could miss when watching an English-language film produced in South Africa, Jamaica or Pakistan. Translators do not blindly look for literal translations. On the contrary, in the translation profession, hints of literal translation often signal low-quality work. Translators focus on meaning and, in the case of films and series, will endeavour to provide viewers with a product that will create a similar experience to the original. The case of Squid Game has been instrumental in bringing discussions about translation to the fore. Of course there are good and bad translations, but the main gain here is the opportunity to debate what determines this. Through such discussions, viewers are becoming more aware of the role and complexities of translation.

Soundgarden, Chris Cornell's Widow Spar Over Royalties and Recordings

It's been a "Black Hole Sun" set of months for the members of Soundgarden, who are in a legal fight with the widow of the band's former lead singer, Chris Cornell, who died in 2017.  Cornell's wife, Vicky, filed suit against the band in December 2019, claiming that her husband's estate was owed "hundreds of thousands of dollars" in royalties for unreleased recordings prior to his death. Now, the band has filed a motion to dismiss the lawsuit. Intellectual property expert and law professor Michael Risch says it's an interesting decision on the band's part.  "The allegations in the lawsuit are heavily fact-bound, as are the defenses," Prof. Risch says. "A motion to dismiss assumes that all the facts in the complaint are true, but the band's motion asserts that the facts are false." He says that band members usually always work things out. When you see lawsuits for copyright or other reasons, it's usually always heirs that are involved.