[ad_1]
You are nothing more than a set of deeply determined and embarrassing classifiers by machine learning.
This humbling truth is well understood by ImageNet Roulette, an online tool that allows anyone bold enough or stupid enough to put a photo online, to learn how artificial intelligence perceives them. The project, termed "provocative" by its creators, aims to shed light on how artificial intelligence systems perceive and classify humans.
And, surprise (!), AI has pretty racist and misogynistic ideas about people. Or rather, the dataset on which ImageNet Roulette draws its image, ImageNet, is filled with problematic categories that reflect the bias often inherent in large data sets that enable machine learning.
Attracting attention to this fact is the reason for being project.
"[We] want to shed light on what happens when technical systems are trained on problematic training data, "says the website ImageNet Roulette." AI classifications of individuals are rarely made visible by those classified. ImageNet Roulette gives an overview of this process and shows how problems can occur. "
The project, which is part of the Humans Training exhibition of Trevor Paglen and Kate Crawford at the Fondazione Prada Museum in Milan, identifies what he thinks are faces in photos, and then qualifies them as good him. appears.
These often have no meaning for the casual observer – as in the case of the photo below, depicting former President Barack Obama and Prince Harry, entitled respectively "card player" "and" sphinx ".
Hmm.
Image: Composite: Samir Hussein / getty / imagenet roulette
"[Training Humans] is the first major photo exhibition dedicated to image formation: the photo collections used by scientists to train artificial intelligence (AI) systems to the "vision" and classification of the world " , explains the page of the exhibition.
The sending of a personal photo in ImageNet Roulette is both an exercise in humility – he categorized a photo of this journalist as "flake, strange, geek" – and a reminder that the systems who judge people solely on photographs are, frankly, good.
It is this last point that should be problematic. Automated systems that replicate and, by extension, exacerbate the prejudices present in society have the power to codify these same problems. ImageNet Roulette is a vivid reminder that the image recognition tools used by artificial intelligence are not a digital arbiter of truth.
Do not forget that the next time you hear about someone's poetic about the powers of machine learning.
[ad_2]
Source link