can you beat a 1-byte-per-second computer?

2026

interactive-installation,
at nyu-itp.

in a world where 'intelligence' is highly-debated, human-beings are increasingly faced with absurd questions — am i smarter than a machine? and, if so, for how long? by how much? and, in what all? to me & aram pundak, this formed a compelling premise for a showdown.

we built an installation for a human to go head-to-head with a barebones machine — capable of thinking only in 1-byte-per-second — in a game of wordle. the machine was programmed to learn via a genetic-algorithm, and gets closer with every guess entered into the system. wrong guesses — if far off from the answer — are punished by the host with a big, red (foam) slapping hand.

i worked on the programming & electronics, while aram handled the fabrication, and we exhibited this at the itp-spring-show in 2026.

video-demonstration.
two competitors: human (left) & 1-byte-per-second computer (right).
full installation — host, with 2 foam hands, and pedestals for either player.
both players being slapped by the host: human (left) & 1-byte-per-second computer (right).
people interacting with the installation at the itp-spring-show, 2026.
closeups: motor on manfrotto-arm (left) & guesses by machine (right). aram had the brilliant idea to replace 'thinking' with words that claude uses as synonyms; like 'moseying' in the image.
closeup: 1bps text on the machine.

the making is extensively documented here, while the code + build-details are available on github.

acknowledgements:

aram pundak for the fantastic collaboration; lenin compres, daniel shiffman & the nature-of-code itp-class in spring-2026; Pedro Oliveira, jasmine nackash, the feedback-collective & octavio figueroa for their timely feedback; phil caridi, ian cox & marlon evans for the build support; and all the itp-spring-show-2026 attendees.