Covering a talk that PhD student Kevin Dunnell gave as part of an "Imagination in Action" event last month, John Werner, a contributing writer for Forbes, highlighted Dunnell and his work for an article on AI, text, and multimodal data.  The article reads in part:

"Sometimes it seems to me like a lot of our research goes into how to take one form of data and make it into another. To frame it a different way, it's about interdisciplinary research and identifying use cases.

We’re seeing a lot of demonstrations and presentations that put this front and center. For example, Kevin Dunnell, a PhD student in Viral Communications at the MIT Media Lab, is working on a project called Latent Lab, a tool that uses AI to help reimagine the user experience of our file systems and how we explore the digital content we consume, create, and communicate.

'LLMs have provided a huge benefit for accessing information and generating text through a chat-based interface,' he said in a recent talk about expanding the interface and re-envisioning what’s possible in research.

Ten years ago, he said, Bret Victor gave a talk on 'Thinking the Unthinkable,' and it inspired him to make progress in 'upgrading the channels of thinking' for users. As a concrete example, Bret provided the double helix structure of DNA, explaining that James Watson and Francis Crick conceptualized it by interacting with physical metal models—a higher channel of thinking than the symbolic methods typical of biochemists at the time.

Latent Lab, he said, aims to help 'upgrade the user’s channel of thinking' by visually and semantically organizing unstructured digital documents. He referred to 'embeddings,' numerical representations of text, a compression interface with interactive visualization, and categorical color coding. 'You can see how intermingled they are,' he said, referring to Media Lab research projects processed by his system and portrayed in the visual graph, 'and it really shows how interdisciplinary we are here at the Media Lab… once you have data in the (system) we can do a few interesting things.' In aid of explaining the utility, Dunnell talked about system features like ‘progressive disclosure’ for transparency of individual documents that collectively suggest high-level organizational themes and ‘knowledge evolution’ to help illustrate how themes evolve over time."