Social-Media
Twitter aims to improve how picture cropping operates in reaction to racial profiling issues.
Twitter revealed this week that it would improve the way image cropping operates on its website, given that the company's machine-based learning algorithm was racially discriminatory in how images were cropped, especially by preferring white faces over black ones.
In a post dive into the problem, Parag Agrawal, Chief Technical Officer of the firm, and Dantley Davis, Chief Design Officer, clarified how the firm checked the racial or gender bias model before introducing the method. Yet Twitter did not publish how such experiments were conducted at the time, so that an impartial review could be carried out, of what the organization terms "surveillance."
To address this, Twitter is actively undertaking further research to bring more rigor to our study, dedicated to publishing our results, and investigating opportunities to open up our research so that others can help hold us accountable, the pair says.
The organization is now working on more specific updates to Twitter to ensure that the photos are shown as expected by users. Specifically, Twitter is promising to rely less on using machine-based visualization methods and instead build software to show users what the images in their tweets will look like when they write a tweet. The company also notes that it has already begun playing with new options for picture cropping and previews to allow consumers more control.
Going forward, Twitter promises to follow a "what you see is what you get" approach, because whatever photo you append when you write a tweet is what you (and other users) can see when you see it on Twitter.
There are some side cases that Twitter will have to sort out, such as very big or very big pictures, but the aim is to make sure users know from the beginning how their photographs will be viewed on the web.
The company moved back to its new machine learning-driven cropping method in 2018, which uses a neural network to crop image previews driven on "saliency." Unlike the company's previous method, which was programmed to concentrate on images, the new model is designed to concentrate on what the algorithm thinks is "the most fascinating part of the image" by projecting where the average user would first want to see.
There is no confirmation yet as to when these updates to the picture cropping method will come out on Twitter.
