Vancouver animation professor says artists are fighting against rising threats of AI
Lorelei Pepi at Emily Carr University says AI threatens artists' identities and livelihoods
A Vancouver animation professor says the impact of artificial intelligence (AI) on students and local artists is terrifying and can't be ignored.
Lorelei Pepi says her students and colleagues at Emily Carr University, like artists around the world, feel the increasing use of AI is putting their identities and livelihoods at risk.
Pepi spoke to CBC's Dan Burritt about why artists are worried, how AI can reinforce harmful stereotypes and the ways artists are fighting back.
This conversation has been edited for length and clarity.
Dan Burritt: This is a big conversation, particularly in your field of study with students. What are you hearing from them about the impact of AI on their field?
Lorelei Pepi: They feel that there is a threat that is at their front door. They're learning skills that take decades to develop and refine in the process of seeing oneself as an artist. The threat of what artificial intelligence does is the idea that they can somehow be supplanted, that their work is somehow capable of being modified and commodified beyond their control — that artists themselves may be threatened in terms of their livelihoods.
A new Netflix animated short film, The Dog & the Boy, was created partially by AI. Why is this so controversial?
This particular work represents animation that was done in combination with artificial intelligence … and in the context of how they presented it, they are preparing us for this inevitability of AI stepping in. It's an example of the degradation that takes place, though I think they intended for it to be a positive.
We used AI to create multiple images of women by searching woman plus different artistic styles and descriptors. How does that actually work, and why are there concerns about it using and reinforcing stereotypes?
Imagine just an enormous library made of millions of images and millions of words. When you put in that data tagging request to create an image, what you're getting is a result that the AI assimilates and assumes to be the best possible match.
To create that data set they did what's called scraping with bots — imagine a fishing trawler with a net and they just grab everything they can. But there is a slant in terms of the attitudes toward certain ethnicities and races, and towards women in the data sets, which are mostly from North American, Caucasian-centric sources.
We have an example of an image where CBC News put in the search terms to create an "exotic woman." What do you make of this?
The term exotic has always been used to be able to refer to those who are other and essentially non-white. That othering is a way to be able to distance, to be able to separate and create a different category and that categorization is a denigration. With this ethnic slant, with this racist attitude that comes through, what we see revealed is a data set that is not representative of the world and that is highly sexualized in some cases.
There are particularly problematic data sets with serious problems around pornography, rape, and children, including one housed in Germany. There was a huge outcry when these [images] started to emerge and so the companies started to pull back and started to use more intentional filtering.
But the problem is they had released these data sets open source and anybody and everybody could get them. So regardless of putting up these gates to prevent people from accessing these kinds of harmful representations, there are plenty of people who know how to get around them, and they have.
How can artists protect themselves and is anyone helping them?
For many of us who have had their work online, that horse is gone, it's gone way down, it's past the street, it's in another country. It's almost impossible to be able to extract your work. Some artists are using legal recourse, including one artist who is suing to be able to have their content specifically removed from the database. At a higher level, there are people who are looking at being able to challenge artificial generation and its relationship to human generated images and copyright.
There's also a software that's being brought out called Nightshade and this is a system of cloaking images in an invisible language that corrupts the data and makes it impossible for AI to function if it gets scraped. I think that this sabotage level is a response to people feeling such deep threat and powerlessness that there's a willingness to be able to push and force these companies to be responsible and to recognize the debt that's owed to having made quite a bit of money off of artists.