Datasets as Imagination

Datasets aren’t simply raw materials to feed algorithms, but are political interventions. As such, much of the discussion around 'bias' in AI systems misses the mark: there is no 'neutral,' 'natural,' or 'apolitical' vantage point that training data can be built upon. There is no easy technical 'fix' by shifting demographics, deleting offensive ter... See more

here are two basic approaches to creating AI datasets. The first one, which is typical of the case we have been studying, a pool of open works is purposefully chosen to ensure license compliance. The second approach creates the dataset by scraping the “raw internet” and relying on copyright exceptions. LAION , a dataset of 400 million image-text pa... See more

As Kristoffer Ørum, artist and self-proclaimed ‘misuser of technology,’ pointed out in a RADAR interview, as LLMs become “very good at drawing things that look like something,” humans have the opportunity to push in the opposite direction, reviving more absurdist and abstract forms of art — much like how the expressionists thrived after the advent ... See more

Many of these projects are saving time by training on small, highly curated datasets. This suggests there is some flexibility in data scaling laws. The existence of such datasets follows from the line of thinking in Data Doesn't Do What You Think, and they are rapidly becoming the standard way to do training outside Google

Beyond collecting and storing information, new tools for thought allow users to resurface, connect, and generate knowledge. Whether visual or textual, this new breed of tools encourage users to link their ideas together, find interesting patterns, and create original content based on their research.

To guard against these issues while reaping the potential benefits of image generators, we provide recommendations such as regulation that forces organizations to disclose their training data, and tools that help artists prevent using their content as training data without their consent.