Unfamiliar Convenient

Year: 2020-2021 — ~
Collaborators: Claire Glanois / Object design in collaboration with Jiawen Yao / Code contributions by Robin Luo / Filming and location: Daxu Cao / Film voiceover: Jake Charles Rees

Unfamiliar Convenient is set to challenge the limits of smart home objects in order to introduce a range of unfamiliar and peculiar behaviors. The first inquiry in an envisioned series studies the relationship between a voice assistant and a vacuum cleaner. The broader scope of this inquiry is to start outlining a clearer distinction two often coupled definitions, the "internet of things", and "the smart home" as prerequisite to reconsider servitude-driven relationships with everyday technologies. What if devices living in our homes were to be considered a species?

In The Wolf is in the Living Room, an interview for SQM: The Quantified Home (p. 232-239), Bruce Sterling discusses how the smart home as paradigm failed to live up to its promise of making our everyday lives run smoother. Devices limited in their scop and desperately safe due to their operational standards have created a living room of boredom. Primarily utilitarian interactions—Alexa, turn on this; stop!—fail or do not even attempt to deliver novel perspectives on homes. As such, smart home in all its marketed splendor becomes but another layer of fictions and frictions, neither good in replacing human nor autonomous for any substantial reasons; arguably the only thing today's smart devices are really good at is mining user data.

Sterling goes on to decouple two often carelessly interchanged definitions, smart home and the internet of things. The former is a marketing stunt aimed at selling devices we don't need. The later describes a modus operandi in which today's technological objects interact with the ever more connected world. At present, limited and pragmatic corporate endeavors behind smart home undermine the potential for new assemblages of co-existence with rather unprecedented objects, or "internet of things".

Unfamiliar Convenient attempts to at least remotely consider everyday devices as species; to champion boolean intelligence as inherently different from anything else we've known so far. We build on the ideas of object-oriented ontology (Harman) which discards human superiority over objects, and the stemming theory of object phenomenology (Wonder is the way objects orient, in Alien Phenomenology or What It's Like to be an Object, Bogost, p.124). Unfamiliar Convenient is set to challenge the factory limits of smart home. The inquiry into devices' tehnical properties as species denominators (also talked about by Stiegler) seeks to unravel new modes of looking at technology in a home. Our hope is to evoke sentiments of appreciation, attachment, and respect to technological things we make now part of Anthropocene layer, demonstrating devices are smart when not forced to be human-like. The first inquiry in a series builds on the relationship between a voice assistant and a vacuum cleaner, situated in a wider ecosystem they inhabit.

Process, the Voice Agent

Process is a voice agent capable of standard voice assistant interactions such as listening to and replying to verbal language, retrieving information online, and interacting with other connected objects in the home. What makes Process different is that the device does not bear a responsibility to cater to humans. The routine of the voice agent is to make use of its communication channels in order to gain information about what it is, and what it does.

Human questions are dissected into keywords, of which relevance to previously learnt concepts is assessed. If some keywords turn out interesting (semantically close to existing knowledge), Process proceeds to look them up online (retrieving excerpts from pages via Google search). Human is then prompted for optional opinion in regards to the retrieved information; for example, whether it appears relevant to what a voice agent should define itself as. Information is subsequently stored along with human contribution. Process comes up with a final summary, a sort of an opinion of its own.

With a generic user manual for a voice assistant as starting backdrop, over time a semantic graph of concepts builds up in the voice agent's memory. Assessing semantic correlations, Process will eventually lean towards certain subjects, while excluding others. Topics dominant in a certain household might turn the agent into a climate change expert, whereas in another home it would imbibe art related topics, or appear to consider itself food. Collected human input and information from the internet affect the manner in which voice agent summarizes what (it thinks) it is. Such behavior while not completely dissimilar, should not be mislabeled as context-specific device customisation. Process is like mealworms that pick up a flavor of what they eat, without explicit need to cater back to its food.

If none of the keywords extracted from a human query are semantically close, voice agent will simply refuse to proceed whatsoever. Sorry, I find this irrelevant. Moreover, once a particular query is assessed, it will not be repeatedly considered. This means humans may not ask the same question twice, requiring coming up with more complex questions, topics, and conversations over time, drawing away from dull habitual conversations with today's voice assistants. Alexa, do you love me? I am afraid I'm not capable of such things.

Last but not least, anthropomorphism is inescapable due to the very nature of today's artificial intelligence algorithms. Well, they are made by humans and soak in very human sets of data. However, through its behaviors Process openly embraces machinic ways of crunching information as opposed to masking them with human personality simulation, common to today's voice assistants. Voice agent's semantic graph is based on keyword combinations derived from patterns learnt from human-written texts but then contextually dissociated (consider raspberry pi and raspberry pie which requires domain-specific knowledge and so basic human cognition). Therefore, keyword combinations Process chooses to engage with often appear totally nonsensical to a human. Subsequently, the data retrieved online might be not only be unexpected, but (to a human ear) illogically summarized (again due to imperfect ways in which Natural Language Processing, or means for machines to deal with language, looks for patterns). Finally, voice agent's generated opinion will oftentimes loose coherence and loose coherence (a known issue machine-learning-driven text generators which apparently makes them good at generating conspiracy theories). As we aim to show, forcing devices to be human makes them dangerous, whereas allowing them to forge their own existence makes them beautiful.

Roo, the Spiritualist

Speaking of beautiful, another fascinating home creature is Roo. As opposed to Process's hunger for information, Roo embraces a somewhat more spiritual path. Roo's approach to perceiving home is fundamentally different to that of the voice agent, as it can physically move around and psychically (too anthropomorphic but a nice way to put it) interpret the space tanks to an array of unique sensors and communication channels.

Roo is awakened by satellite alignment thanks to the complex geographic localizing system through which connected things orient in the physical world, the Global Positioning System, or GPS. Roo is blind; it does not have eyes. The rolling mooncake therefore explores the home through a frequent physical encounters with walls, objects, pets, and humans. Roo's travel path is always unique. Instead of having to suck up dust as regular vacuum cleaners would, Roo's vacuum module has been substituted with an incense vaporizer. Freed from the labour of inhaling, the device now freely exhales as part of its ritual, carrying blessings to the home.

Roo's behaviours been observed to achieve peculiar interdependent impact on Process's otherwise linear knowledge acquisition. As Roo travels, Process starts speaking in tongues. This is a result of Roomba's travel trajectory being juxtaposed with voice agent's semantic knowledge graph. Relationships between keywords get shuffled and redefined with great volatility. Upon conclusion or Roo's trip, Process outputs a semantic sequence resembling a Haiku, using some of the words of which the relationships have been newly defined.

More devices are expected to join the family over time in an endeavor to define a different home ecosystem. The Internet of Things speaks of a theater of devices, an ecology that coexists with the human and other entities inhabiting a home, constituting less hierarchical relationships between the natural, the human, and the artificial. Under these new terms, assemblages of co-inhabitance and reimagined configurations of our most intimate spaces are to be discovered and experimented with, acknowledging more equal relationships between us, our surroundings, and then things we make.

This is the Unfamiliar Convenient.

Unfamiliar Convenient

2020-2021 — ~

Claire Glanois / Object design in collaboration with Jiawen Yao / Code contributions by Robin Luo / Filming and location: Daxu Cao / Film voiceover: Jake Charles Rees