« There are two things really when we're thinking about images and studying images at large scale and using these big models and how they've been trained on massive data set, there's kind of two things to be concerned about. One is data privacy, of course. So we don't really know a lot of times where these images are coming from, what's going on, and this is something that others that have presented about generative AI, this is a real problem that they face. But I'm going to focus a little bit more on this bias in predictions. What we find is that because a lot of these models have been trained using human labeled images, the sort of predictions that we get might convey the inherent biases that might exist among the coders. » · MIARA dossier thématique

« There are two things really when we're thinking about images and studying images at large scale and using these big models and how they've been trained on massive data set, there's kind of two things to be concerned about. One is data privacy, of course. So we don't really know a lot of times where these images are coming from, what's going on, and this is something that others that have presented about generative AI, this is a real problem that they face. But I'm going to focus a little bit more on this bias in predictions. What we find is that because a lot of these models have been trained using human labeled images, the sort of predictions that we get might convey the inherent biases that might exist among the coders. »

Citations: « There are two things really when we're thinking about images and studying images at large scale and using these big models and how they've been trained on massive data set, there's kind of two things to be concerned about. One is data privacy, of course. So we don't really know a lot of times where these images are coming from, what's going on, and this is something that others that have presented about generative AI, this is a real problem that they face. But I'm going to focus a little bit more on this bias in predictions. What we find is that because a lot of these models have been trained using human labeled images, the sort of predictions that we get might convey the inherent biases that might exist among the coders. »
Est cité par: Constantine Boussalis
startTime: 1,197
endTime: 1,245
has top concept: Intelligence artificielle; Biais algorithmique; Prédiction
datasetTimeInterval: 12 October 2023 – 12 October 2023

Filter by resource type and property:

Items with "Citation tirée de la conférence: « There are two things really when we're thinking about images and studying images at large scale and using these big models and how they've been trained on massive data set, there's kind of two things to be concerned about. One is data privacy, of course. So we don't really know a lot of times where these images are coming from, what's going on, and this is something that others that have presented about generative AI, this is a real problem that they face. But I'm going to focus a little bit more on this bias in predictions. What we find is that because a lot of these models have been trained using human labeled images, the sort of predictions that we get might convey the inherent biases that might exist among the coders. »"
Title	Class
The Computational Turn in Visual Political Communication Research	Conference

There are no annotations for this resource.