mariano sardón

<-- OTHER WORKS bio short bio contact español


"Deep UnLearning" proposes a game between robots and people in which an AI algorithm seeks to learn elementary human gestures by means of a dynamic dance between technology and the participation of the audience.
"Deep UnLearning" is an installation conceived as a mediating learning system in between machine, humans and algorithms that communicates in a nonverbal way.
Robotic arms conducted by neural networks, ask people to interact by mimicking gestures of newborn babies. Interfaces in each robot capture gestures of participating people and images once registered and stored in a database, algorithms use them in a sort of nonverbal communication learning process as they were a baby state of mind.
Inspired by neuroscientist Andrew Meltzoff’s researches about newborn language learning [1.pdf] , [2.pdf]; complex face gestures made by people mimicking elementary gestures of newborns; implies that AI algorithms learn early communication gestures in a kind of a reverse learning process.

The installation is completed by two data visualizations projections which exhibit the algorithmic process in different stances and ways.
One of the visualizations shows real time recent images of gestures performed by participants by using the robotic interfaces that passes through different neural network generated images in an animated sequence.
Spectators can see algorithmic recognizing process of relevant characteristics of gestures like mouth borders, holes of eyes and noses and certain face contours. All of them pretty decipherable features for humans; a kind of metaphor of newborn state of mind


Note that much of the mathematical algorithmic process maintains hidden, it goes too far to our understanding. A tiny space of mathematical procedures in neural networks are recognizable for humans.
A second big projection shows a much more abstract composition made by AI algorithm during its running process. Many of the images in sequence displayed, are extremely abstract and unfamiliar ones; however, information in such an algorithmic visualization is still hidden and undiscovered.
Some parts of the visualization sequence belong to forbidden states of the neural network, a kind of dreaming images not considered along the learning process.
Visual dynamics of the sequence reminds us the resemblance of a brain scanner of thoughts, which in this case, thoughts are achieved by an algorithm.

“Deep UnLearning".
by Mariano Sardón & Mariano Sigman.

Exhibited at Andreani Foundation, Buenos Aires. 2020 - 2021

General Coordination:
Germán Ito.
Techonlogical collabortaion:
Ignacio Fourmentel
Amilcar Malatesta

Panasonic robots, power sources, controllers, projectors, computers, sensors, Artificial Intelligence algorithms.

Patagonia CNC-Machines.