Re: <nettime> Computational Culture issue ten. Special Issue: Situated Bayes

[Felix Stalder via nettime-l <nettime-l@lists.nettime.org>]

Hi Matthew,

Congratulations! A great issue, a really timely and urgent extension of 
the line of thinking that I encountered first in Joque's book. The use 
of Bayesian statistics might create an opening towards very different 
political ends than those which is is currently used for and that 
exploring this opening might be a more productive than simply "resisting 
(AI)". We talked a bit about that over dinner recently.

In much of the philosophy/epistemology concerning Bayesian statistics 
the issue of the "prior" is absolutely central, and your intention to 
turn of it from a problem for objectivity into the foundation for 
situatedness is absolutely correct.

What is usually less discussed, perhaps because the issue not unique to 
Bayesianism, is the question of the threshold. When is the likelihood of 
an hypothesis being true strong enough to act as if it were true?

In ML, they try, as you write, minimize the situatedness by using 
"noninformative priors" despite the extra compute this requires, but 
they can at least to be non-subjective. In many ways, the prior is 
subjective only in a context where computation is scare. In a context 
where computation is treated as abundant, it's meaningless, a random 
starting point in a very long line of iterations. It's not subjective, 
but brute force ;)

But the situatedness creeps back in through the threshold. What degree 
of error is acceptable, which  is always also a question of who has to 
cover the costs of these errors. In this way, Bayesianism create a new 
type of externality.

I think this question of threshold, while not unique, is particularly 
urgent in Bayesian systems because they are less about generating 
knowledge (in a conventional scientific way, there the threshold is a 
stable p-value) than about enabling agency, on the spot, under a 
subjective risk/rewards ratio. In certain systems, say placement of 
advertisement, a 20% likelihood might be sufficient, in others, say, 
systems in HR departments, one would hope of a much higher threshold. 
The point being, the threshold is entirely subjective.

The consideration of the subjective/situated/political nature of 
threshold might open up less towards the issues you are concerned here, 
but more towards social justice question (how to distribute 
risks/rewards), but as a source of subjectivity it's a bit underrated.

Anyway, great issue!


all the best. Felix




On 7/25/25 09:28, Matthew Fuller via nettime-l wrote:
> Computational Culture, a journal of software studies
> Issue Ten, July 2025
> Special Issue: Situated Bayes
> Edited by Juni Schindler, Goda Klumbytė and Matthew Fuller
> Special Issue Introduction
> Juni Schindler, Goda Klumbytė, Matthew Fuller, [Situated Bayes – Feminist and pluriversal perspectives on Bayesian knowledge](http://computationalculture.net/situated-bayes/)


--
| |||||||||||||||| http://felix.openflows.com |
| |||||||||| https://tldr.nettime.org/@festal |
| for secure communication, please use signal |

--
# distributed via <nettime>: no commercial use without permission
# <nettime> is a moderated mailing list for net criticism,
# collaborative text filtering and cultural politics of the nets
# more info: https://www.nettime.org
# contact: nettime-l-owner@lists.nettime.org