Observations on Dumb Robots


(A lengthier—better—version of this was published in POOL Issue 2: Rules)

Observation I: The majority of devices that claim to be “intelligent” or “smart” in the realm of architectural design—devices for the home, personal computing, and other non-industrial devices—are in fact dumb. They constitute a realm of artificial intelligence referred to as weak AI or narrow intelligence.

Observation II: A common counterpoint to the plausibility of true artificial general intelligence is the argument that we simply do not have the computational power to process all the rules necessary for such an algorithm to work. Since weak AI is already permeating new urban infrastructures, and tools for manufacturing, designers are beginning to address larger cultural implications of these ubiquitous processes (beyond the technical).

e.g. How do big data algorithms affect how we interact with our own disciplinary knowledge in curatorial or historical practices?

Observation III: Twitter is a new site for a variety of information exchange. In a social climate where political figures are expected to communicate through short, one-hundred and forty character snippets, the relevance of lengthy discourse is waning.

Observation IV: Twitter Bots are an emerging type of weak AI that participate in discourse along with real users online. Twitter Bots have a variety of abilities. They can hack webcams (@FFD8FFDB), generate images (@archmixes), assemble coherent sentences (@LatourBot), even go back in time (@wayback_exe).


Description I: As a means of articulating this phenomenon diagrammatically, I made a dumb robot that says smart things. Theory Machina (@theorymachina) is a simple twitter script that “reads” K. Michael Hays’ seminal anthology, Architectural Theory Since 1968. The script parses through a unicode text file and filters out complete sentences matching a set of rules based on twitter’s character limits.

Description II: The conditional statement used is what is referred to in computer science as a regular expression,“regex,” component; a syntax originating in 1956 that searches a sequence of characters and extrapolates user-defined patterns. Regex is a core component of any search engine algorithm. Our bot ignores specific characters, page numbers, footnotes, etc. In plain language the script states, “find me the next instance in which fifty or more characters, but less than one-hundred and forty, contain a capital letter and end with a period, exclamation point, or question mark. Ignore numbers followed by periods.”


Reflection I: The result is clunky at times, and depends largely on the text that is being processed. Some tweets are declarative with manifesto overtones while others are more fragmented musings. Curiously, there is an immediate contextual distinction between a Manfredo Tafuri tweet and a Denise Scott Brown tweet based solely on sentence structure. Twitter also has its own algorithms which make some tweets stand out from others, giving the feed itself a pseudo-personality. Due to the imperfections in the regular expression conditional, and the text file’s raw structure, some errors do occur which create semantic breaks in the flow of the feed. These range from abrupt change in tone, topic, or sentence structure to unforeseen breaks in abbreviations and other notational devices.