Foone🏳️⚧️
I imagine it's only a matter of time before some site like Facebook or discord adds a "feature" where they train an AI model on each of your contacts when you talk to them, so that if you ever ask them a question and they're away from the phone, it can have the AI model respond instead.
"your friend isn't here, but this is what we think they'd say:"
Foone🏳️⚧️
The only part of this that's fictional is the idea that they'd make it so absolutely clear when you're talking to the model instead of the human
Foone🏳️⚧️
@tswartz07 as a digital necromancer myself, I'm offended by the association
Tom Swartz
what about digital necromancy?
"your deceased friends memorial page was updated with some statuses based on what we'd think they'd say"
Foone🏳️⚧️
Anyway they have to turn it off and delete the model if the person dies, because Charlie Brooker has a patent
Foone🏳️⚧️
@niki now that's a fun wrinkle
squeak with your heart :ace:
@foone when someone blocks you they won't tell you and just keep them on your timeline as AI
Foone🏳️⚧️
@niki suggested that this could be used for blocking: you train a model on yourself, and instead of actually blocking someone, you just force them to only talk to the model.
Like the worst kind of shadow-block: they can't talk to you anymore, but they may never realize it
Foone🏳️⚧️
Game idea: a chat room is set up with N players, who all talk to each other normally. At some point, one of them will be replaced with an AI model trained on what they'd said before. Players can accuse each other of being the AI, but if wrong, they're penalized. They get points based on how close their accusation was to the flip.
Foone🏳️⚧️
The real strategy is in how players act before the replacement, because people know this is going to happen. So they're incentivized to act as neutral as possible to make themselves more likely to escape detection once AI-replaced, but at the same time they want to make other players act in hard-to-simulate ways, so it'll be easier to determine when they go AI
Foone🏳️⚧️
The final wrinkle: the AI models are being trained on the previous gameplay sessions... And the AI are playing by the same rules. They're allowed to accuse each other of being AI.
And if one AI goes undetected long enough, another AI is added. The AIs don't know who else is AI, so they can accuse other AIs of being AIs and win.
Foone🏳️⚧️
Anyway in the original idea, the only indication that you're talking to an AI model and not a human is that they use other AI image generation bullshit to re-render the person's avatar but with a goatee.
(Unless their avatar already has a goatee, in which case it'll be removed)
Foone🏳️⚧️
@Kiloku @niki I recently watched a security talk about AI manipulation and one of the examples an AI class was using was having them design a website which used an AI model to compose a shopping cast, and the security issue was that you could convince the model there was a special sale on today and get items for 100% off.
Kiloku
@foone @niki imagine the chaos when the model says something that can be interpreted as a binding agreement or something
Foone🏳️⚧️
@clacke I was really confused by this comment for a second, since "Turing" is my last name
a Claes unto himself 🇸🇪🇭🇰💙💛
０７６萌ＳＮＳ is a social network, courtesy of ０７６. It runs on GNU social, version 2.0.2-beta0, available under the GNU Affero General Public License.
All ０７６萌ＳＮＳ content and data are available under the Creative Commons Attribution 3.0 license.