Prevent dangerous AI: Difference between revisions

Prevent dangerous AI (view source)

1,174 bytes added , 7 March

no edit summary

3,082

edits

@@ Line 4: / Line 4: @@
 }}
 ===Not being able to trust pictures, video, audio===
-Let's draft up some policies that could stop this. Join the {{tp2}}.
+{{R|FAQ
+|''What about AI detection software?''<br />Even though generators and detectors both get more advanced over time, the generators ultimately win, unfortunately. ''Read&nbsp;more''
+}}
+What could be done:
+:; policy ideas:
+:* [[Ban unwatermarked generative AI]] {{talk|should watermarking be combined with an additional policy where every AI image generator company has to keep a database of all generated images, and offer a free service for anyone to verify any image on the internet, as in, to see if it matches something in the database?<br />what should we title such a policy?}}
+:; tech ideas:
+:* [[Anonymous camera digital signature]]s
+Status quo:<small>
+* Google's AI image generators produce images with a "synth ID" so they can be identified as fake. Not sure how foolproof it is though.
+* Sora AI (video generator) adds a subtle watermark to the video. Some people have been known to crop it out though.
+* There are plenty of image generators that don't produce any watermark at all.
+* Watermarks in general are not hard to remove, unfortunately. Still better than nothing though; most people probably won't bother removing them. So watermarking can at least reduce the total amount of misleading slop on the internet.</small>
 {{R|FAQ
 |Why not just create more jobs? {{p2|''Read&nbsp;more''