Mikaela, Lily, Violet, and Ciri — a plural collective of nerdy, quoiromantic, poly, lesbian computer engineers and leftists.

Current media obsessions: Persona 5, RWBY, Cosmere


garak
@garak

The innovation of AI Art is that it has invented a novel method to take the existing users of your product, and treat them as capital instead of as human beings.

Of course we have realized and internalized the adage in the title of this post, it has been true for a long time. But most people (myself included) thought it was just about advertising. If you're not paying, then your eyeballs are being harvested by the second for fractions of a penny. But that's not all! Everything about you can be monetized.

This has been true for a while now. Google infamously operated a 411-like phone service, purely for the purpose of harvesting voice data, and shut it down when they had enough data (and were blindly copied by Microsoft). Facebook asks you to tag your friends so that you can provide training labels for its person-detection algorithm. When you are logged in to Chrome, your browsing activity is used to build anti-click-fraud models so Google can pay less money to publishers.

And now Github uses your code for Copilot, and every image hosting service will lay dubious claim to artwork for purpose of building their own proprietary image-generation framework. It is being treated as a commodity, a material precursor for a manufacturing process. Ore for the refinery, yarn for the loom, grist for the mill. The economic value of a collection of artwork will be quantified in terabytes.

The moral outrage against AI Artwork stems from the perception that it is dehumanizing. This perception is rooted in fact.


@Qyriad shared with:

You must log in to comment.

in reply to @garak's post:

Google infamously operated a 411-like phone service, purely for the purpose of harvesting voice data, and shut it down when they had enough data (and were blindly copied by Microsoft).

one small note here (since I worked for Tellme): Tellme Networks was operating their directory assistance line since 2003, and Google ripped the idea off of them; Tellme was then acquired by Microsoft in 2007, the year GOOG-411 launched

also Tellme's motivation wasn't purely harvesting voice data -- it was also operated as a lead-generation tool for their interactive voice response platform

more relevant and less direct side note: after Tellme's standalone voice command apps got wound down, I moved onto doing the OS-level voice command functionality on Windows Phone 7. WP7 would do voice recognition for voice dialing on-device -- this way we didn't have to send your contact information over the wire to a server in the clear, back in the days when SSL termination was still an expensive undertaking -- and only do open-grammar recognition1 on a network recognizer.

the first (and thankfully only, so far) time I ever had someone tell me I was making a "career-limiting move" was when I insisted that we shouldn't even send voice data to the network recognizer unless we were sure we would need it, on the grounds of user privacy -- not only would that make the speech recognition less responsive2, but the Continual Refinement team was counting on getting all that voice data!

thankfully, I did manage to at least talk people into not sending "dial" commands for pressing numbers on the dialer keypad3 over the wire, on the grounds that the utterance would be sent in the clear and could potentially be a Social Security number or PIN or something similarly sensitive.


  1. for things where the space of things you can say is unbounded, like sending text messages or doing web searches

  2. by like half a second

  3. required as an accessibility feature by the FCC, for people with visual impairments