###### tags: `CDA`
# Reading Responses (Set 2)

### Apr 03 Fri - Artificial intelligence
The rise of generative artificial intelligence has made it possible for virtually anyone to produce striking images with nothing more than a text prompt. While impressive, it comes at a steep cost to artists and writers who have spent years perfecting their craft. 

As Newhauser explains, diffusion models can now generate "high-quality, photo realistic images that are becoming increasingly difficult for humans to distinguish from non-generated images." Thus, when someone posts a piece of art online, instead of praise comment sections can be full of accusation of the artwork being AI-generated. It is clear to see that the traditionally valued aspects of art such as composition and artistic talent are the same qualities that trigger doubt since AI has gotten so good at replicating these traits. As Tyler Gold observes, large language models are essentially "small, thin veneers of smiling faces" layered over a distorted reflection of all the information humanity has ever put online. Addition to the increased scrutiny for real artists, AI constantly makes mistakes and hallucinate at a rate of 50% according to researchers from MIT and Stanford.  

#### Additional Reference:
MIT: https://arxiv.org/abs/2602.19141
Stanford: https://www.science.org/doi/10.1126/science.aec8352#sec-3

### Apr 07 Tue - Algorithmic bias
Fundamentally, algorithms reflect the data they're trained on. As researcher Johanna Burai noted, society dictates Google's search results and the stereotype visible in them is a reflection of structural racism rather than programming. 

One of the core issues is that algorithms cannot understand context. Searches for "unprofessional hair" overwhelmingly return with images of black women, but many of those images actually came from an article protesting the same issue. This demonstrates how algorithms match keywords without context and ended up reinforcing the same stereotype the source material was trying to combat. Similarly, gaps in training data can also lead to disaster, as demonstrated when Google's photo app mislabeled black users as gorillas. 

However, bias in algorithms can also manifest through human choice. The National Review documented how ChatGPT would refuse to produced a story of then President Joe Biden's corruption but when asked to write about Donald Trump's corruption it was more than happy to. Though this article seemed very bias and right leaning, it is still important to note how the researcher's at OpenAI encoded their values into ChatGPT, intentionally or not, which inevitably  carry political weight.

### Apr 14 Tue - Digital language and generations
Gretchen McCulloch frames the internet language as this dynamical system shaped by its users rather than as a derivative of "proper" writing. She defines groups of people as old, full, semi, pre, and post internet people based on the conditions under which they went online. She shows examples such as the semantic shift of "lol" from literal laughter to insinuate irony, the usage of post style writing for many older internet users, and strategies of teens to get around "context collapse", to argues that language innovation happens mostly through digital language and writing. McCulloch's interview with NPR summarizes this distinction between old and new rules as old rules enforce the correctness of language whereas new rules create connections between people.

McCulloch's point against the digital native vs digital immigrant framework that many people believe was very interesting. She showed that technological skill and online social skills are not the same. For instance, many Post Internet People may know how to tell the subtle difference between different meanings of 'lol' based on the sentence that surrounds it, but they might not be able to organize files into folders. 

### Apr 17 Fri - Pushback
When Logan Lane recounted standing by the Gowanus Canal in Brooklyn and deciding her iPhone belonged with the trash floating in the water, she was enacting one of Morrison and Gomez's motivation for pushback behavior. Logan's acted in a way to take back control from the "evertime" of constant online connectivity. Morrison and Gomez's article discuss why users revolt and sorts them along a spectrum from changing small personal habits to full disconnection. Valukul's article on the Luddite club showed how Lane and her peers have swapped their smartphones to flip phones, gathered in parks to read, and have begun to spread their messages to nearby colleges.

Something that Morrison and Gomez missed was how pushback became performative and communal rather than private. Their categorization defined pushback as actions that individuals would take. But the Luddite club flips that notion around and showed rejecting technology can lead to subgroups forming. Which leads me to question if pushback as a collective strengthens it, or does it reintroduce the same dynamic of 'in groups' that drove people off social media to begin with.