Fluxus Travel in the Age of AI

Over the last few years, cultural tastemakers have been revisiting the legacy of Yoko Ono (I would personally recommend Lindsay Ellis’s video essay or Sarah Marshall’s podcast “You’re Wong About”). Yoko was appreciated in her most active years for her contributions as an artist to Fluxus, “an international, interdisciplinary community of artists, composers, designers, and poets during the 1960s and 1970s who engaged in experimental art performances which emphasized the artistic process over the finished product.” Yoko’s best known pieces are arguably her performance pieces, their documentation, and the cultural discussion that surrounded them; but her books Grapefruit and her 22 Instructions for Paintings are today more often generative of new art. Both are excellent examples of a specific type of Fluxus art: the written instruction.

Instead of handing you the finished piece of art to contemplate, the written instruction details the process of bringing art into being. Part of experiencing the piece is contemplating where to put your focus; we might understand the intent from reading the instructions and we might understand something entirely different (but complementary) when we execute the instructions. Or you can find meaning in what is materially created, leftover, or changed when the instructions are finished. It even ventures at times into the meta: in the piece “______, __ _______ circus on ______” by early Fluxus artist John Cage, we are asked to follow instructions to create another set of instructions that then can be performed. By describing a process, the “written instructions” format elicits the same contemplation of artistic action and experience despite the absence (temporal or physical) of the artist, making the written instruction uniquely scalable.

Often, I see the instructional nature of these written pieces and I, a software developer, can’t help but be reminded of code. The written pieces are, by definition, algorithmic¹. I’m not the first person to have this thought: in Christiane Paul and Carol Mancusi-Ungaro’s 2019 Whitney exhibit Programmed: Rules, Codes, and Choreographies in Art, 1965-2018 they draw direct throughlines from the written instructions of Fluxus to digital art. Paul and Mancusi-Ungaro point out not only the similarities in formatting, but also in the way each medium is distributed. Written instructions were experiences distributed via international mail networks of Fluxus creators, foreshadowing the way digital art would later be experienced through the internet. Both code and this format of art are able to capture the creator's intentions and make them uniquely spreadable, accessible, and reproducible. In this way code, like the Fluxus format before it, has proved similarly scalable.

Less than a three hour drive from Mobi’s headquarters, the Massachusetts Museum of Contemporary Art (MASS MoCA) lets you wander multiple warehouse-sized floors of Sol Lewitt pieces. Lewitt’s pieces are written instructions, sometimes mathematical in ways that also remind me of code. The execution of these instructions is unseen or has been deemphasized. This leaves the viewer at MASS MoCA with the resolution of the process (massive, room-sized murals) and the instructions for their creation side by side. The murals have a certain grandiosity, making it easy to forget their size is mostly a byproduct of the space they are housed in rather than inherent to the piece. One could imagine following the same set of instructions in a kitchen or a dollhouse and getting quite a different effect.² The instructional nature of the pieces has made them reproducible, scalable, and accessible while simultaneously enforcing localization and uniqueness.

The Fluxus performance art pieces almost localized too well–Yoko Ono’s Cut Piece 1964 and Bed-In performance pieces directly reflect the current events surrounding the performances. Even more than the Sol Lewitt paintings, the performances' cultural impact make it hard to imagine them somewhere else… in a different cultural milieu, with a different artist or a different war. But as we move further from their date of creation, many Fluxus instructional pieces localize and create new feelings and observations about our here and now. In 1952, when John Cage debuted 4'33" (a written-instruction piece that asks the performer to make no intentional sound for 4 minutes and 33 seconds) in a concert hall in Woodstock, audience members were enraged by how it challenged them to think of ambient sounds as music. In 2024, the piece sparked debate again when a professor at Columbia controversially refused to listen to John Cage’s 4'33" because—at that moment, in that location—the piece would have challenged the students to confront the sound of protests outside. When artists really intentionally craft instructions, they become somehow both for everyone at any time and also hyper-localized in a way that unlocks something more about our world and ourselves.

Excerpt from 4'33" by John Cage

On its face, Mobi.AI is a tech company—but we’re really trying to deal in algorithmic Magic and Art. I remember in my early days at Mobi, it felt like many hires who weren’t field-experts in AI had experienced a travel itinerary that changed the way they thought about travel. That is, a local friend planned out their visit, a professional curated the perfect vacation, or they themselves researched and optimized travel in a way that, when executed, revealed something magical about people and places. We don’t forget when a friend recommends a specific hotel because they remember your daily run can be taken around the nearby park. There’s something delicious about a tour guide building the perfect snack break into the middle of a long museum day. Personally, I got really good at fitting a fencing practice into my schedule anywhere I travelled to meet the locals and break up my indulgent vacation meals with exercise. We, at Mobi, felt really captivated by how intention and localization fused together to create these moments. We’d come to appreciate that a really exceptional tour planner is nothing short of an artist (or a magician).

The problem Mobi was trying to solve then was “how do we preserve the magical qualities of these trips when we try to scale them?” The product and design team started by breaking down phenomenal trip itineraries into their constructive parts. We knew we loved a specific week-long beach trip to Sicily crafted by our in-house tour guide. What could be learned from the structure and the ideas of the trip? We wanted to abstract the trip template enough to apply it to other beach trips in Miami or Thailand and have the template naturally and properly localize to those new destinations. At the same time, if the trip template is abstracted too much, it loses the guide’s original craft. We call these overarching trip templates “trip topographies,” but they are, on an artistic level, Fluxus trip instructions. A trip topography maps out the contours of a specific trip. How do you pace activities, what kinds of activities or meals pair well together, how much time do you spend on transiting vs remaining in different locales?

These ideas are the basis on which the itinerary-building aspects of our “Agent for Agent” platform were developed. A travel guide can create a trip template that describes the “DNA” of the vacation. Using a calendar/personal planner-like interface, they can define the important features or themes of the trip and describe how activities are intended to relate and synergize with each other. Then, when an agent selects a template trip for a given date and location, Mobi’s AI tools fill in the localization details. Each calendar block in the template can be populated with timely, local, bespoke recommendations for restaurants, hotels, and activities. The system automates and optimizes routing and transportation. In short, we capture the travel agent’s artistry and insight as “Fluxus trip instructions” and use technology to facilitate the instructions’ localization.³

For more tangible examples of what localization looks like: maybe a travel agent normally runs a trip Monday-Wednesday, but a client is traveling Thursday-Saturday. The original trip cannot be copy-pasted without modification because activities in the original trip like museums, restaurants, and trains might be closed or have modified hours. Shifting the trip by hand requires travel agents to find these conflicts, make transportation modifications, and use the original “intent” behind a cancelled activity to make a smart substitution. Shifting the trip with our system automates this work. With a template describing the trip topography, we can shift the Monday-Wednesday trip to Thursday-Saturday and use our AI and routing tools to seamlessly validate and adjust the trip to the new time frame while relying on the original travel topography to preserve what was special about the original itinerary. Similarly, a trusted travel agent who regularly uses a specific trip template for their beach trips to Sicily, Miami, and Thailand could reuse this same template to generate a new, localized vacation in an unfamiliar location (say Sydney, Australia).

This effective localization is enabled by Mobi’s amazing travel datasets, data-processing capabilities, and data-retrieval systems. If our beach vacation trip instructions tell us to “have a dinner with an amazing view of the sunset over the water, no more than a 25 minute drive from the previous activity”, we have the ability to find that restaurant in Sicily, Miami, Thailand, and even on Sydney’s eastern-facing coast! All while assuring it compliments the rest of your itinerary. We do this by pulling data about the time of the sunset, the location of potential restaurants and the water, each restaurant’s travel time from your previous activity, and review/photo aggregation, so we can reserve a table at just the right time and place to create a magical sunset moment.⁴

This example also hints at how we don’t just use instructions to generate entire itineraries. We can also use Fluxus-like recipes to improve recommendations for one-off activities, hotels or modes of transportation. With this “recommendation” task, the human artist we are amplifying with our technology is not necessarily a tour guide, but a data-savvy hobbyist⁵. If a diver wants an excellent localized snorkeling recommendation, for example, they might combine a map, water clarity forecast, temperature forecast, and a guide book to select a good dive spot for their next trip. A geospatial data scientist can scale this process, writing instructions to automatically generate candidate scuba locations over different timeframes and trip-destinations using the same data sources (we explain this and more in greater technical detail in our previous explainer “How Does Mobi Make Great Place-Based Recommendations?”).

James tenney and Alison Knowles’ Fluxus piece “House of Dust (1967)” combines a computer and fluxus-instructions written in FORTRAN to recombine four variables about a house: “material”, “situations”, “lighting” and “inhabitants”. This early pieces uses computers to explore the wide range of ways a house could be, possibly allowing us to discover combinations we wouldn’t have otherwise considered.

This allows us not only to hunt down the local gems that you’d normally need an inside-connection to find, but sometimes results in unexpected, serendipitous discoveries. Mobi has, for example, used the science behind predicting rainbows to identify algorithmically where to stand to see one! This means we could, as part of an itinerary, reroute someone to a street corner at the perfect moment to encounter a rainbow. The idea of a “magical rainbow moment” and how to construct it both come from a human, but it is the computational mediation that enables the moment to be found in the real world.

Discovery of this sort was not a major theme in the Fluxus art movement of the 1960s and 1970s. Even when Fluxus art pushed us to reflect on our relationship with technology, Fluxus pieces often reframed familiar contexts and relationships more than they provided guides for exploring the unknown and unfamiliar.

In the age of extensive datasets and AI, would a renaissance of Fluxus art have more to say about computationally-mediated discovery? We can imagine how a Fluxus written instruction piece might leverage today’s technology to shift an experience from a familiar context into an entirely different culture or location. This written instruction would then serve as a structure to compare, contrast, and explore the new context—creating a starting point for connection and lowering the barrier to new experiences. In a travel context, this might look like going on a more intentional and serendipitous trip enabled by smartly leveraging data about the world. Empowering people to experience the world calls back to the Fluxus movement’s ethos of experiencing art through participation. Looking at the Fluxus art movement helps us understand why the “written instruction,” code, and AI systems can help us scale ideas, localize our experiences, and create magical moments. This is why Mobi can look to design, art history, and social analogy to imagine the future of human-AI collaboration.

_{1. “Informally speaking, an algorithm is a collection of simple instructions for carrying out some task. Commonplace in everyday life, algorithms sometimes are called procedures or recipes. Algorithms also play an important role in mathematics. Ancient mathematical literature contains descriptions of algorithms for a variety of tasks, such as finding prime numbers and greatest common divisors. In contemporary mathematics, algorithms abound.” Theory of Computation, Michael Sipser.}

_{2. One could even imagine asking ChatGPT to generate imitations.}

_{3. This piece was directly inspired by the idea of using AI to amplify the artistic and compositional genius of real people who have been at Mobi. I would like to specifically call out Shelby Dziwulski’s amazing insight into trip topography and its contribution to this concept in this piece.}

_{4. “There are one thousand suns arising every day. We only see one of them because of our fixation on monistic thinking”—Yoko Ono, Grapefruit: A Book of Instructions and Drawings.}

_{5. As in footnote 2, I would like to specifically call out Elina Oikonomaki’s for her amazing data visualizations and prototypes that used geospatial datasets to show we could find these magical moments. Her thoughts on how to understand a place through its data were essential to producing this piece.}