top of page
Search
Writer's picturejames girouard

Matmulling Around in my Mind

Do you think about Artificial Intelligence? What even is it? How can inert, non-human machines make 'intelligent' decisions for and about us flesh and blood emotional soulbags? These questions swarm my mind.


In looking for answers, my research led me to the nitty gritty of programming ai. What does it look like to code a hierarchy of choices and options that are theoretically 'intelligent'? Within the philosophy of intelligence we can assume that decisions that have some kind of experience, or thoughtfulness, or informed history behind them can inform better decisions. So again, what might this actually look like?


If we imagine information as a shape, perhaps a square with each point as one possible decision then we can begin to visualize at least four answers to a question. But logic may get fuzzy. For example, the question "what color is a bluebird?" has one obvious answer, but when we dig deeper, we see that a bluebird has more colors than blue. Even an ai generated image (Thank you Adobe Generative Recolor) has many hues of blue and even some reds.


Furthering this visualization, we can imagine each question to have many answers on many squared that make up cubes that multiply the possibile answers of each question. This is what the rug, Matmul is all about.


Matmul is the technical term for a matrix multiplication, which is an operator in machine learning. This combination of pathways that determine decision-making is depicted by cubes upon cubes with a multiplicity of patterns connecting them. When we ask an AI or large language model, such as ChatGPT, or Generative Recolor to generate something for us, it weighs connections between what it 'knows' to create a likely response.


My inspiration for building this design was inspired by a blog post by Anthropic’s Performance Team member, Simon Bohem. The nature of this operation is visualized to aid in understanding its mechanics to aid in implementation. The patterns that have emerged in trying to understand this aspect of machine learning were fun to make. Other ml connections between computer vision bounding boxes and other data-driven algorithms are easy to make when looking at Matmul, but the handmade nature of this rug ground it in the tangible world.


To build the rug, an empty burlap coffee sack was stretched on two frames and each strand of wool carefully pulled through loop by loop using a handheld hook. Once complete, the frames were removed and the border built. This traditional form of rug hooking makes the piece relatable to everyday life and reflects the culture of textile arts.

89 views0 comments

Recent Posts

See All

Comments

Couldn’t Load Comments
It looks like there was a technical problem. Try reconnecting or refreshing the page.
bottom of page