Checking Twitter’s Vibe And Speculative Sonic Futures With Vibebot
Rowena Chodhowski, Maurice Jones, and Leona Nikolić
In the 1930s, music was introduced into elevators in an effort to reduce passenger discomfort and fear as these apparatus were quite hazardous and unpredictable at the time. This music was not designed as an object for active listening, but rather played in the background as discreetly perceptible, unobtrusive melodies (Ohmura et al., 2018). Such ambient sound became known as ‘elevator music’ for its transformative function as a sensory element—while an elevator may be perilous, it does not have to feel that way. It is here, in the elevator, that we may begin to understand not only music as a legible vocabulary that influences a particular experience of a space, but also to envision the potential for soundscapes to generate speculative futures.
To speculate is to operate within a space of uncertainty and ambiguity. Speculation, in this way, is not analogous to prediction; it is about generating proposals for the future (Dunne & Raby, 2013). Speculative design thus emerges as an imaginative act of interrogation that “actively resists the plausible and the probable” (Marenko, 2018, p. 421) by rejecting the current state of things and insisting on something else. When considering the aesthetic dimensions of speculative design, “music is a very powerful means of denoting the future” (de Oliveira, 2016, p. 43). As sonic fiction, music prioritises the experience of the world rather than the seeking to define it. Such experience of the world is wholly contextual and may be framed as a feeling about the state of things. This conceptually slippery phenomenological condition is one we seek to interpret as ‘vibe.’
By examining the nebulous and trendy notion of ‘vibe’ as a contextual language of both aesthetics and of the political economy (James, 2019, 2021, 2022; Powers, 2019) that is at the theoretical nexus of mood (Heidegger, 2007), affect (Massumi, 1995), ambiance (Spitzer, 1942), atmosphere (Böhme, 2018), and aura (Benjamin, 1935), we conceptualise vibe as a collectively informed and embodied vernacular that is sensory (Garcia, 2020), rhythmic (Henriques, 2010), mathematical (Grietzer, 2017), and programmable (Kender, 2022). As such, we propose a theoretical confluence of the notion of vibe with principles of sonic fiction and practices of speculative design.
Translating this theoretical framework into algorithmic software, we developed VibeBot, a Twitter bot that performs a ‘vibe check’ of the social media platform. Since it was founded in 2006, Twitter has earned a reputation as one of the angriest and most unpleasant social media platforms (Gayomali, 2014; Leopold, 2013; Suciu, 2022). Inspired by the relaxing effects of elevator music, our intention with VibeBot is to induce a conceptual “vibe shift” (Davis, 2022) for the social media platform, as an effort to reimagine the hostile and “toxic” (Kender, 2022) vibe of the platform through playful soundscapes. In other words, this project investigates the means by which we can create alternate futures on Twitter by transforming the platform’s current vibes into speculative sounds.
After several iterations of different concepts, we produced a functioning albeit non-autonomous prototype of the VibeBot. To check Twitter’s vibe, the VibeBot extracts a large sample of public tweets corresponding to a trending hashtag and analyses them as an assemblage using the distilBERT-uncased-emotion sentiment analysis model to determine the hashtag’s vibe in one of six pre-determined labels (sadness, joy, love, anger, fear, surprise). The label ascribed to the hashtag is then processed through MUBERT, a text-to-music AI model which generates an audio file that represents the hashtag’s vibe. VibeBot finally produces a tweet consisting of an audio file and its corresponding hashtag. Thus, both followers of VibeBot and those viewing the trending hashtag are exposed to VibeBot’s vibe check. With VibeBot, we explore how to mobilise vibe as a speculative vision of the future that proposes what Twitter could be rather than what it is. How can tweets, as a textual present, become a sonic future through the VibeBot’s speculative vibe modelling? Moreover, how can vibe engage with possibility, potentiality, and fantasy? And how can we mobilise a non-sonic vibe as a sonic experience?
In our attempt to create a bot which would be capable of performing a vibe check on Twitter, it became apparent that, despite the improvements in natural language processing techniques and the increase in powerful natural language processing models like BERT and GPT 3.5 (Chat GPT), there remain significant impediments to sentiment analysis beyond positive-negative and polarity analysis. It also raised questions about whether or not sentiment analysis, based solely on textual information, could correctly discern a poster’s vibe.
In addition, VibeBot uncovered many of the challenges surrounding the ‘magic’ of AI. First of all, there is the immense amount of unrecognised human labour and resources that goes into the development and implementation of seemingly automated AI systems, be it the extraction and labelling of data sets or the time and resources spent in and on Colab notebooks. Second, this hidden, human labour shows how there are many critical places of possible intervention in the making and deployment of AI systems. In this project, questions of which sentiment analysis and AI tools to use, what constitutes a vibe, how to map a vibe to a sentiment to a sonic output, are all points of intervention. Third, a general shift from an initial free and open-source paradigm towards increasing commercialisation became apparent. While models are in trial mode and not developed enough to deliver satisfying results they are open source and free for people to explore and use. In turn these interactions are used to feed data sets and improve algorithms until the point where the models are ripe for commercialisation.
In the future, it may be less straightforward to create a vibe-checking bot using open-source tools and accessible models available to the general public. Increasingly complex models and large data sets may demand increasingly costly and power-intensive computing equipment, while affordable access to model application programming interfaces may become more difficult for hobbyist and non-commercial users. However, the current shift in the internet—driven by the role of AI across a variety of industries—may make it more important than ever to check the ‘vibe’ of our rapidly evolving postmodernity.