NOTE
Algorithmic recommendation systems are sociotechnical systems. This is present even in current iterations of ML supported systems, which require fine tuning which is performed by the culture.
This concept of a “trap” emerging from a recommendation system is something I have felt. There’s a drive to purchase when seeing a constant stream of related items, all possible avenues of pleasures. This is often shown at checkout, or even post-checkout pressuring for another purchase.
In Computing Taste, Nick Seaver conducts an anthropological study of the technologists who design algorithmic music recommendation systems. He explores their ways of thinking and talking about music, taste, and computation to better understand their technological design approaches. By highlighting the humans behind the machines, Computing Taste shows how to think about computer algorithms as sociotechnical systems.
How designers think about “captivating algorithms” that they create.
- taste algorithms are often positioned as assisting audiophiles to navigate the overwhelming catalog of music that they can access. Humans are drowning in a sea of information, so these profiles are sought to navigate the problem of “too much.” How designers approach captivation.
- A designer needs a model of that which-is-to-be-trapped, because the trap needs to be able to lure its intend prey. This is why recommendation algorithms need to model their users including the different types that would exist. Machine learning remains a sociotechnical system. Human designers decide not only what to feed into the ML algo but to tune its outputs so that they conform to their expectations.
Those systems are shaped and continuously maintained by the beliefs, approaches, and discourses of the technologists who develop them.
digression
Captivating Algorithms - Recommender systems as traps - Nick Seaver
Jokes - Control Language - Islands overlap. Perform a query, and get a unending stream of related music..
As the company grew, so did the algorithm
Now, the ‘algorithm’ is not one algorithm at all, but ‘dozens and dozens’ of sub-algorithms, each of which parses a different signal: What does a song sound like? How often does a user click? What has a listener liked in the past? A master algorithm orchestrates the sub-algorithms’ outputs together into an ‘ensemble’ (Goldschmitt and Seaver, nd) that makes a simple decision: What song should be played next?
Cold start of recommendations:
But sophisticated recommendation requires data. New users pose a challenge that researchers call the ‘cold start problem’: they have no data yet and, without data, datadriven recommendations do not work.
The book is subtitled ‘How to Build Habit Forming Products’
Wandering or guided
I found that almost on a daily basis I would start with the intention of following one particular route of investigation and then find myself seduced by the aesthetics of one of the websites visited and moved by the simplicity of clicking to follow a link proffered by that site. A few more clicks would send me hurtling down some channels carved out of cyberspace by the sculptured links of these website creators, often to such a degree that it was hard to retrieve the original place from which this diversion had began
The question to ask of traps may not be how to escape from them, but rather how to recapture them and turn them to new ends in the service of new worlds.