The AI Team that Brought Back Andy Warhol

In February 2019, Zohaib Ahmed founded Resemble AI – a voice generator and cloning service. The results of their programs are often startling and eerie. Earlier this year, Netflix employed Resemble to re-create the voice of Andy Warhol for their documentary series, The Andy Warhol Diaries (2022). The choice to have a machine resurrect Warhol likely would have pleased the artist, who famously told curator Gene Swenson, in the November 1963 issue of Art News, that ‘everybody should be a machine’. But its application has raised numerous ethical questions among artists and policy-makers. Frieze editor-in-chief Andrew Durbin spoke to Ahmed about how AI might change the art world in 2023 and what guardrails are in place to protect artists from theft and forgery.

Andrew Durbin: For many of us, this was the year artificial intelligence leapt into our lives – from Dall-E to your team at Resemble AI re-creating Warhol’s voice for Netflix. What can we expect in 2023?

Zohaib Ahmed: Next year we will see innovative, generative AI flourish and find its way into all aspects of content creation. We will see more AI-generated voices in movies, television shows and video games, as well as new applications for personalized content creation at scale.

AD: Artists are still grappling with how to use Generative AI. What are some of the best applications you’ve seen so far?

ZA: One of the key differentiators of Generative AI over other technologies is that it truly offers the ability to create applications that weren’t possible before. A couple of examples of this are Avatar AI and Ask My Book, where language- and image-generation models are used for small yet creative tasks that require human-like skills.

AD: I can imagine a world in which an exhibition of art created by AI is reviewed by an AI critic for a publication edited by AI. How do we best utilize this technology while still prioritizing the human touch?

ZA: You can look at historical uses of AI to see how we can retain the human touch. You may be familiar with Adobe’s Magic Wand, which was introduced into various tools like the Creative Suite. It helps you quickly select and remove objects from photos. However, professional photographers still need to draw on their human expertise to choose the right settings and to understand the context of the photo to make sure the AI tool is used in the most effective way. Similarly, I believe AI can be used as a tool to help us create art, but human expertise and understanding will still be required to achieve the best results. We believe this to be true with AI Voices as well, which is why we introduced a feature called Resemble Fill that uses AI inpainting techniques to add or replace words and remove mistakes from existing audio clips.

AD: So, you think the human touch still matters?

ZA: It absolutely does. At the end of the day, generative AI is uniquely created from datasets that are devised by humans. As such, a key differentiator will be the ability to infuse human creative touchpoints throughout the process to help inform, guide and shape the results to achieve the best possible outcome.

AD: Artists are concerned about the ethics of AI – whether it might appropriate and recast their work without their consent. Have these concerns fed into your team’s approach?

ZA: We are very aware of the concerns around the ethics of AI, and we take them very seriously. We have a dedicated team that is responsible for ensuring that our technology is used ethically and responsibly. We have also put in place a set of policies and guidelines that we require our customers to adhere to. These policies and guidelines are designed to ensure that our technology is respectful of people’s privacy and data rights, and that it is not used in a way that might cause harm. We also believe that it’s our responsibility to educate and work together with others in the industry to define guidelines. That’s why we open-sourced Resemblyzer, a python package that ships with a pretrained model to analyze voices with deep learning.

AD: There’s a lot of anxiety, especially among policy-makers, around AI’s disruptive potential for altering public debate with misinformation. It’s not unthinkable that AI might someday play a role in art forgery, too. Are there guardrails that you believe should be in place for AI’s use?

ZA: There are guardrails – particularly around the issue of misinformation – that should be in place for the use of AI. We believe it’s important not only for AI applications to be transparent about their algorithms and how they generate results, but that they should be designed in a way that allows humans to easily understand and verify those results. For example, in the case of art forgery, we believe that it’s important for there to be a way to easily identify when a work of art has been created by AI.

AD: Where should frieze readers look for the latest developments in AI?

ZA: There are many individual creators doing amazing things with AI tools. Following them on Twitter is a great way to keep a pulse on the latest developments. I would suggest levelsio, Danny Postma, SALT and Itay Schiff.

_{Main image: The Andy Warhol Diaries, 2022. Courtesy: Netflix}