AI Index: Interview with Obvious & Ivona Tau

This is a Distributed Gallery interview with Obvious and Ivona Tau as part of the launch of "AI Index", the first drop of [aside] project.

On May 27th, Obvious and Ivona Tau will launch AI Index, the first drop based on the [aside] protocol. This drop will have the following feature: all purchased NFTs will be locked at mint time and will be unlocked according to the state of a sentiment gauge which estimates people’s feelings towards AI by using X’s API .

First of all, before we get started, your new project called AI-Index is a collaboration. How did you meet each other and why did you end up working together?

IT: We met in real life at NFT Paris in 2023, during a discussion panel on the main stage, together with Snowfro, Colborn Bell, Craig Palmer from Makersplace and moderated by Giulia Archetti from Sotheby’s. Before that, I was aware of Obvious' work and, of course, of the famous Belamy auction. We had a shared passion for GAN models, an older way of making AI, before the prompt imaging craze. We decided to collaborate on a project when we were at another event in Paris. It’s quite funny how you often meet the same people within the digital art community but in different places on different occasions, without planning. In this way, it still feels like a very small and welcoming community.

O: We have been aware of Ivona’s work for a few years, and we have always felt close to her in our practice, as we are both early users of AI for art, and bots started with creations using GANs. We connected early on with the aesthetics she develops with the combination of GANs and photography. Early on, we felt this kind of connection, as we participated in the same drops (Braindrops, K011, Bright Moments..), and were invited to speak at the same talks (NFT Paris). When we finally met, it clicked. Our visions overlapped somewhere, and this somewhere felt like a strong basis for developing an artistic project together.

When we presented [aside] and explained to you how the unlocking conditions work, you quickly shifted towards working with the sentiment of people over AI, in a way where artworks are unlocked depending on the public's sentiment towards AI. Could you elaborate a little on this choice?

IT: Artificial Intelligence is certainly THE topic on people’s minds nowadays. Having worked with this technology for almost a decade, it gives you a different perspective about its possibilities and threats. People tend to react very emotionally towards new advancements within AI (of course it is all exaggerated by the media) and we see those trends in public opinion being either scared of AI (overtaking the world, killing humanity, stealing jobs) or excited about it (curing cancer, solving environmental problems, making our lives easier). So we decided it would be a perfect index to accompany and unlock an AI artwork. We linked the aesthetics of the works to the sentiment over AI value. This means that darker and more pessimistic works are unlockable and tradeable only when the public sentiment of society over AI is very low.

O: When you introduced us to these new mechanics, we saw an ocean of possibilities before us. We tried to narrow it down to our core practice, which consists of exploring the complex relationship we have both as humans and as a society towards technology. We feel this relationship is extremely interesting, as it consists of mixed feelings, often extreme, and quickly shifting over time. It consists of both fear and excitement, hope and despair, and it relates to something everyone feels is bigger than them. Science follows its unstoppable course, technology develops, and humans are left to comment. We wanted to explore this very particular relationship and felt [aside] was the right opportunity to do so.

AI raises many questions and controversies. As artists working in the field, do you find yourselves regularly subjected to criticism? And within the circles in which you operate - art / AI research - do you perceive a sense of threat/fear of AI in your surroundings?

IT: It’s quite funny how several years ago, when you mentioned the work was created using AI, you would normally elicit a very positive interest just because of the tools you used. There was a sort of appreciation for using complex tools and code to create art. Now, even though my process has not changed a lot, the situation has been completely reversed. Because of the proliferation of AI imagery and the ease of use of prompt-based tools, there is already some stigma and stereotypes associated with the general field of creating work using AI. You need to defend your process, concept and vision a lot more because just creating something aesthetically pleasing is no longer a challenge. To be honest, I don’t necessarily think it’s a bad thing. Art should not be just about the medium but about the person behind the narrative and the message it conveys.

O: It’s very interesting to see how the reactions to AI-generated art evolved. When we created our first artwork which was quite mediatized, we had the chance to experiment with it at its fullest - extremely polar reactions. People either despised or loved what we were doing. Then suddenly, everyone got these tools in their hands. At this point, we felt something changing, and the focus, the interest of people we met, switched from the tool to something deeper: the message, the approach. We felt the conversations became more interesting, as we could skip all the “AI will replace artists” questions, as people now understood AI to be what it is, a tool, and directly dive into the more interesting stuff. Today, we feel a lot of excitement in the research field, and still some fears in the public sphere. And that’s probably how things need to be to move in the right direction.

Regarding the aesthetics of the AI index works, you have chosen to create fictitious newspaper covers representing different futures in our relationship with AI. How did you arrive at this aesthetic choice?

IT: I think it’s a symbolic choice. Newspaper front pages used to be a sort of a society mirror - signalling the most important events happening, the most influential people and the most pressing trends. Nowadays we mostly use social media through personalised algorithms so the main message is not as curated as it used to be and will be different depending on your “bubble”. So the choice of newspaper covers is a nostalgic nod to older information medium. It is also a reminder that all that feels extra contemporary today, will be very quickly outdated tomorrow.

O: To us, magazines are a perfect illustration of how things change, and how fast they change. You can look at magazine covers not that far in the past, and feel they convey very different standards as compared to what’s presented today. So we thought, with the extremely fast development of AI, the subjects we are currently discussing are likely to be totally obsolete in a few decades. And so we started working with Ivona on creating something that would make today feel obsolete. Additionally, we felt an attraction for the aesthetics of magazines, which are both very diverse and follow a high level of standards. Perfect food for algorithms.

If we were to call it "AI Art", it typically begins with data collection, correct? Could you tell us about your process, how you made your choices, and what combination of datasets was used to create these works?

IT: For me, this data collection was very different from what I would do usually, as I didn’t rely as much on personal archives. The project goal is to imagine alternative futures in different scenarios concerning AI acceptance and show them as imaginary headlines in popular newspapers and magazines. By alluding to popular culture and publications in the Western culture such as Economist, TIME or National Geographic we wanted to evoke familiarity in the viewer and show the immediacy of the speculative futures. To create the models, we gathered several datasets of the historical magazine covers as well as themes that commonly occur in them (nature, fashion, portraits). During training, we wanted to capture some of the popular aesthetics present in the front pages and transform them into AI-centric worlds. What’s interesting is that even when you finetune AI models with custom data, it keeps much of the language understanding, so the results often show the stereotypes of how we visualize artificial intelligence.

And of course, the most beautiful part of the process was the early conversations with Obvious where we discussed our visions and explored a range of ideas. It was somehow very easy to make the creative decisions together as we very quickly agreed on the main concept.

O: Data collection is indeed the starting point of working with AI algorithms. Nevertheless, while collecting data, we like to have a final idea of what we want to create, in order not to get too lost in exploration. The goal of this collaboration is to create something that reflects society’s feeling towards AI. Therefore, we decided to focus on several subjects which, to us, represent the main spheres in which AI is likely to bring drastic changes, either positive or negative. We listed those areas, and chose to collect magazine covers with aesthetics that resonate with each area. Typically, a drawing in the style of the Economist is easy to associate with geopolitics, while a very detailed and photorealistic picture of a landscape often indicates the magazine has something to do with nature.

In a recent documentation about AI Art produced by Serpentine R&D Platform, I read this sentence from ‘Herndon Dryhurst Studio’: “All media is now training data.” A person outside of the art world might wonder if there are legal risks in using models of famous magazine covers as training data. Or, once there is so much divergence from the original work, are we completely outside the realm of the law? Generally speaking, how do you feel about these legislative issues and debates?

IT: It is an very important topic that we discussed since the beginning. We wanted to create work about pop culture and society that uses the original assets for model training to extract common stereotypes and aesthetics present in the print media. However, when working on such a project you have to be mindful of the transformative use (which in our case is speculative parody) and careful about the outputs you produce, for them not to be too similar to anything in the training data. While AI models could be capable of perfectly imitating the covers in training data this was not our goal and we made sure that there is always a divergence - both in the subject matter (working with themes and subjects not present in training data) and artificially added corruptions to make sure no one mistakes our works for the original magazine covers.

In the traditional art world, there have been many cases of transformative use of magazine covers. For example, artist Michael De Feo uses Vogue covers as his canvas and adds painted elements to the original photographs on the covers.

O: It is a very interesting subject, and probably one of the major legal pain points surrounding AI today. There is no clear legislation or incentive around data use for the creation of transformative outputs. Usually, we tend to use free-of-rights data, which is the zero-risk strategy. When making art, you can’t always have a zero-risk strategy, and sometimes going in the direction you wish to go involves doing what you believe to be right. For the creation of this series, we did what we and Ivona both felt to be right: entirely new compositions of original images, titles, and AI artefacts, which are identifiable as not being actual magazine covers, which focus on representing tomorrow's society, and our sentiment towards AI.

It's a somewhat naive question, but it might help me discover other artists. I wanted to know if there are any AI artists you admire, and if there's a particular AI artwork that fascinates you?

IT: I’ve lately been fascinated with the works of Iñigo Bilbao, Pete Burkeet and Petra Voice.

O: Among those we admire as artists (AI and generative) and as humans are Pindar Van Arman, Sofia Crespo, Zancan, Agoria, Claire Silver (whose vision we connect with), and of course (and you definitely already know this one), Ivona.

If I'm not mistaken, the second major step in the work is the choice of architecture and model parameters. How did you proceed, and what difficulties did you encounter in creating these works?

IT: Yes, that’s right. The choice of architecture is usually dictated by the desired aesthetics. While the data controls the direction towards which AI model is going, the model parameters and architecture dictate the path of how we are getting there. Some of the models would produce more details due to their larger learning capacity, while others would give more abstract and generalised outputs. For this project, we trained several Stable Diffusion type networks. The biggest challenge, I guess, was training the models to generalize beyond what’s in the training data. For example, the magazine covers in the training data did not have any images related to AI and when prompted to generate something on this topic they very quickly diverged from the training data aesthetic. We also had one very funny geopolitical model that overfit to a small part of the data and only produced good images when the prompt had the word “China” in it.

O: Choosing the models was a major discussion we had and one that would have the most impact on the final visuals. Although we were both very familiar with the use of GANs, we decided to work on something more explorative, which would in the end better serve the purpose of the series. We decided to train different Stable Diffusion models, one for each subject we wanted to work on. This way, we could develop very different aesthetics within the series to represent the different societal concerns we wanted to tackle. After that, the main challenge resided in creating covers which are future projections of what society will look like, involving the same aesthetics as existing covers, with different subjects. We managed to get the results we were hoping for with an iterative process of prompt and parameters engineering and a beautiful co-selection process.

In the AI Artist's toolbox, I imagine there are some « tools » that come up very often, and favorites combinaisons of tools. What are your favorite tools and combinations? Especially for AI-Index.

O: In this increasingly large toolbox, we like to use a different combination of tools for each project. We are particularly interested in bringing tools which are originally dedicated to different mediums and arranging them to serve a common purpose (for example creating audio-reactive videos, or here creating an interdependency between text and image). For the development of this series, we used different algorithms, including text generation, text removal, text arrangement, image generation, and image-to-image algorithms. This would not have been possible a few years ago, and we find it increasingly interesting to combine the available tools to achieve a vision, in this case, the one we share with Ivona.

IT: For me, the stack of tools I use to create varies a lot from project to project and also evolves gradually with time, as we get better models and tools. My favourite is not necessarily a tool, but the part of the process where you explore the freshly trained model. I don’t use any sophisticated UIs for that and my favourite kind of “tool” is a simple Python script within Jupyter Notebook that generates a new image for a new random seed with every refresh. I guess it is a kind of code equivalent to pen and paper - something quick and dirty that lets you focus on the core idea.

In AI-Index, works are unlocked when a sentiment gauge based on X reaches certain intervals, ranging from extreme fear to extreme enthusiasm over AI. Could you explain how you created 10 works per interval, resulting in 10 different aesthetics, each aesthetic being linked to one of the intervals/feeling states?

O: Each artwork has two variables, the subject it treats, and the sentiment it relates to. While the subject defines the aesthetics of the artwork, the sentiment defines its subject, and both are interdependent. We had a lot of fun building the different categories and defining their aesthetics through data, finetuning and generating iterations. We also enjoyed projecting ourselves in the 10 possible futures of society, each one represented by an interval, and each one part of this larger scale. What we love about the series is the many ways you can read it, that is through the categories of aesthetics, through the scales for each subject, or the categories of scales.

IT: Obvious came up with this amazing list of imaginary titles & guidelines that could be linked to different index values and placed on a linear scale. It helped a lot to then relate these scenarios to visual ideas. I also think that there is a kind of subjectivity within each interval, with aesthetics being an added dimension to this multidimensional representation of human sentiment over AI.

In the 10 intervals, [0,10][10,20][20,30][30,40][40,50][50,60][60,70][70,80][80,90][90,100] which one is your favorite and why?

O: We really love the [10,20] and [80,90] intervals, as those are the ones where you feel everything is already extreme, but with a very slight touch of moderation.

IT: Ironically, my favourite is [0,10] - the most pessimistic one. It realizes the darkest aesthetic and this is something I’m frequently drawn to in my work, even though I’m an optimist in everyday life.

The infrastructure on which the works rely is that of ChainLink, and in particular, the ChainLink function stack which allows to connect smart contracts to any web2 API. As artists, what do you think of this new opportunity for the art world?

O: To us, the development of features such as connections of the NFTs to the ChainLink infrastructure is exactly what NFTs can bring to artistic creation. While NFTs have been a game changer for digital art, what we feel stands out most is the creations that make use of the blockchain mechanisms. To us, this type of artistic project conveys the best of what technology can bring to art - new mechanisms to change both the art creation process and the art collection experience; brought together by the artists, the gallery, and the smart contracts developers, who altogether build a system that makes sense, both technically, conceptually and artistically.

IT: I absolutely love this idea, and I was instantly captivated when I first heard about the project. With code-based work, there is a lot of potential in interactivity, for the work to be driven by external forces, whether it’s the viewer or an external API. I am also excited about the opportunities to engage with the blockchain, justifying the use of the medium by more than just tokenizing work that could exist outside of it. ChainLink allows a great entry point for artists like myself (with no experience in directly working with blockchain) for this kind of exploration.

If you were to create a new series of works using Oracles, with what external phenomenon would you like to connect your works and why?

O: We share a strange fascination for the universe, the infinite space, and everything that relies on the paradox of our existence. If we were to do another series of work, it would probably relate to the revolution of celestial things or the occurrence of random phenomena in the very large universe.

IT: My crazy idea would be to create an artwork that would be connected to myself. I’m very much a numbers nerd and love to keep track of my steps, heart rates and all sorts of parameters related to my body. Making an API that would upload this data and then connect an artwork to that, for example, by unlocking it only once I’m dead, would definitely be a project on my wishlist.

My last question is a bit inevitable: where do you stand between extreme fear of AI and foolish enthusiasm? Between 0 and 100, where are you?

IT: Similarly to our AI index values, my beliefs fluctuate too 😄 One thing I’m not afraid of is AI gaining consciousness and destroying humanity. But I’m afraid that AI is a powerful tool that even without sentience could cause both harm and good. So - 64 at the moment.

O: It is a bit special to answer for us, as we are three members in the trio, with divergent opinions on the matter. The fascination for the relationship we have towards AI probably comes from the very discussions we have within Obvious. If we had to draw the average, we would probably stand around 75.