@mattly yeah it feels like before we can responsibly consider the question of what a non-hostile recommendation algorithm looks like, we have some hard questions around transparency which need to be answered first! great point
Conversation
Notices
-
Embed this notice
tech? no! man, see... (technomancy@icosahedron.website)'s status on Tuesday, 25-Apr-2023 01:46:43 JST tech? no! man, see...
-
Embed this notice
tech? no! man, see... (technomancy@icosahedron.website)'s status on Tuesday, 25-Apr-2023 01:48:22 JST tech? no! man, see...
@mattly "it's too hard" but the thing that's hard is getting product people to care, not the technical difficulty of doing what must be done
-
Embed this notice
Joel VanderWerf 🐧 (joelvanderwerf@mastodon.social)'s status on Tuesday, 25-Apr-2023 02:19:25 JST Joel VanderWerf 🐧
As a user, I'd be happy with a personal construction set for an algorithm, allowing me to compose things like "if an account whom I follow AND frequently interact with had a popular post AND I did not see it AND it was within the last 24h, then put it in my timeline." Just checkboxes and number fields, nothing Turing complete!
tech? no! man, see... and Matthew Lyon repeated this. -
Embed this notice
tech? no! man, see... (technomancy@icosahedron.website)'s status on Tuesday, 25-Apr-2023 02:19:25 JST tech? no! man, see...
@joelvanderwerf @mattly right; so not to get too political here, but the value of software has always been able to be classified as one of two possibilities: either it makes it easier to control people, or it empowers people to do things they couldn't do before
modern megacorp software is firmly about the former camp while just making enough pretense towards the latter to retain a facade of legitimacy (sometimes it does accomplish the latter, but only incidentally; it's never the point)
but legitimately revolutionary software puts the second one front-and-center, and that has always meant putting control in the hands of the user, and since the distinguishing property of software is that it can be reprogrammed, it almost always means making it possible for the end user to reprogram it
I would argue these properties are fundamentally intertwined and we only miss the obvious because our imaginations have been shaped by spending too much time in a land where megacorps make the rules
-
Embed this notice
tech? no! man, see... (technomancy@icosahedron.website)'s status on Tuesday, 25-Apr-2023 02:26:31 JST tech? no! man, see...
@mattly @joelvanderwerf right; I know that's not your argument, sorry to hijack your thread but I guess I'm saying we need both
they certainly are closely related tho right? I mean, if you are going to provide the means to construct your own algorithm, you can't reasonably do that without providing the tools to debug it ... which may be the exact same tools of transparency and visibility you would need anyway for any algorithmic recommendation?
-
Embed this notice
tech? no! man, see... (technomancy@icosahedron.website)'s status on Tuesday, 25-Apr-2023 02:28:41 JST tech? no! man, see...
@mattly @joelvanderwerf this is actually very much in line with patterns that are common in other end-user-reprogrammable systems: the extensibility system used by the end user needs to be the same as the mechanisms used by the core maintainers to build the system, because if you're doing it right then the needs of the two groups overlap a lot
-
Embed this notice
tech? no! man, see... (technomancy@icosahedron.website)'s status on Tuesday, 25-Apr-2023 03:31:33 JST tech? no! man, see...
@lobotomy42 @mattly if you take a step back and frame it more broadly:
* attempt to improve your experience
* explain why it thinks these results will do that
* allow you to correct it when it's wrongeasier said than done obviously
but different people have different goals, so the question you asked cannot have a single answer
-
Embed this notice
Greggo (lobotomy42@mstdn.party)'s status on Tuesday, 25-Apr-2023 03:31:34 JST Greggo
@mattly @technomancy
I am not quite sure what a non-hostile recommendation algorithm even would do.Recommend me things that are “good” for me regardless of whether I engage with them?Recommend me things that only generate the “right kind” of engagement? Make me think more? Make me happier in the moment? Make me happier in ten years?
There is a question here about what it means to recommend something. For the hostile algorithm, this is conveniently answered — addiction.
-
Embed this notice